Sign up to save your podcasts
Or
Share Cleaning Data for AI Starts With Context, Not Perfection
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Cleaning Data for AI Starts With Context, Not Perfection
Today on the Salesforce Admins Podcast, we talk to Chris Emmett, Salesforce Solution Architect at Capgemini. Join us as we chat about how to clean up your data to prepare your org for Agentforce, and why data without context is useless.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Chris Emmett.
AI requires clean data
I caught up with Chris hot on the heels of his TDX London session, “Prep Like a Pro: Clean Data and Metadata for Agentforce.” He’s an experienced Salesforce consultant who has helped countless organizations of all sizes reboot their business processes.
As Chris explains, unless you have a company of five people that started last week, your org probably needs some data cleanup. And if you want to get started with Agentforce, you need to do the work to make sure the agents you build can understand your data and use it to generate actionable insights. After all, if you can’t derive useful information from your data, then it’s useless.
Why cleaning data can feel like boiling the ocean
When I worked in sales, we used a CRM that was so complicated that only one guy at our company knew how to use it. Talk about a bottleneck!
The truth is, if your business has been around for a little while, you’ve probably inherited all sorts of legacy data. Maybe it’s some random field created by that one guy in the 90s who didn’t document anything, or a legacy system like SAP or MSX that is essential to your day-to-day operations.
Chris has seen it all, and it can often feel like cleaning up all that data is akin to boiling the ocean. It’s a monumental task with no end in sight, let alone getting the organizational buy-in to do it in the first place.
A practical way to start cleaning your data
Chris recommends focusing your data cleanup strategy on the functionality you want to build in Agentforce. For example, if you want an agent to email a customer when their opportunity is five days from the close date and still unsigned, what data do you actually need?
You don’t need the 300 fields that might be on the opportunity page, or the 300 fields in that account. You might need the opportunity’s name, the stage of the opportunity, the close date, the account, and maybe the primary contact of that account. That’s five pieces of information.
Suddenly, you don’t need to boil the entire ocean—you just need to boil a cup of water. So start small, focus on the functionality your data cleanup project will deliver, and get the ball rolling. Trust that the things you build with Agentforce will speak for themselves, and you’ll be able to generate momentum to clean up your data project by project.
Make sure to listen to our full conversation with Chris to learn more about how to clean up your data and provide context for AI agents. And don’t forget to subscribe to the Salesforce Admins Podcast so you never miss an episode.
Podcast swag
Learn more
Admin Trailblazers Group
Social
Full show transcript
Mike:
This week on the Salesforce Admins Podcast, we’re joined by Chris Emmett, consultant, data enthusiast and Salesforce evangelist accidentally. Well, Chris is passionate about data, and he takes us on a journey from legacy systems and those DOS screens to databases and AI-powered actions, all while sharing practical advice on how to clean up your data without making it feel like you’re boiling the ocean. So if you ever wondered how to prepare your org for Agentforce or why data without context is basically useless, this one is for you. So listen in, share it with somebody who maybe is swimming in a sea of records. So let’s have Chris swim his life raft over and let’s get him on the podcast.
So Chris, welcome to the podcast.
Chris Emmett:
Thanks for having me.
Mike:
Yeah. Well, it’s good to be here. I got a note from Jennifer Lee. She saw the session that you and Jonathan did at TDX London about cleaning data and cleaning metadata, and last week, we talked with Jonathan about cleaning metadata, so this week, of course, we’re going to clean our data because who doesn’t have clean data? But let’s start off first with a little bit about you, Chris. How did you get into Salesforce and why are you so passionate about clean data?
Chris Emmett:
Yeah. Before I start that, you said who doesn’t have clean data? It really should be the other way around.
Mike:
I know. It was a rhetorical question to get people to listen and be like, oh, but there could be that one person that tunes into the podcast, like I don’t need to listen to this. My data is sparkling.
Chris Emmett:
You know what? If there is a company out there with five people who just opened up last week, their data is going to be impeccable, and they do want to listen to this podcast. They can tune out.
Mike:
Right.
Chris Emmett:
I started in the ecosystem in a weird and wonderful way because a lot of … you’ve obviously got your accidental admins that find a weird and wonderful way into Salesforce. I was an accidental consultant. So I started out fresh from university as a desktop support engineer, just fixing Windows and fixing printers and fixing Office, and it was great, fun. Then I did a bit of work as a developer, cutting my teeth on Visual Basic 6.0 and .NET all the way back in 2008. But that entire team was made redundant to make way for an off-the-shelf manufacturing system. And as an SME, as an expert in how that old system worked, I was brought in, and then the project manager on that project quit, and I became a project manager, and I was a project manager for 10 years.
And then about 10 years later, around 2016, I was working for a company and we were really interested in changing how we managed products. We wanted a brand new system to manage our projects, and one of the products that we looked at was Salesforce. And my word, I was blown away. Literally within one day of using Salesforce, I was creating formula fields and workflow rules as they were back then, and I just fell in love. And this is coming from a person who had spent the best part of a decade dealing with systems where if you needed to add a field, it would take two, three months to get through, and I was dealing with a system where I could start a sentence, explaining to someone what Salesforce was, and by the time I’ve finished that sentence, I could have created a field. I could have expanded that data model, and I fell in love with it.
The company I was working for did not fall in love with it, and that pilot failed. It fell by the wayside, but I was hooked. I was like, man, I need to change my career. So I start looking for project management jobs within the Salesforce space. I had never project managed software before. I had project managed big old factory systems that were very waterfall in their approach. I had never done agile before, so I was applying and applying and applying, getting nowhere. I was probably applying for about eight months, and then this small company in Cambridge were like, dude, you are not great for a project manager but you seem really enthusiastic. Had you considered being a consultant? And I went, but yeah, because it’s not managing the projects that I love. It’s the system that I love. Salesforce is a platform that I love. I want to be able to empower other companies to improve themselves through Salesforce.
So I got a job there as a consultant, and that was 2017, and I have just been building up and flourishing, and since then, I’ve got 600 odd badges on Trailhead and I think 23 certs now. So I’ve just gone all in and built my career up.
Mike:
Wow. Holy cow. That’s a lot. I feel such a kinship with you because I joined … when I started doing Salesforce stuff, I worked at a publishing company, and we had this Apple-based CRM. I can’t even remember what it was called, but there was one guy in the office that still knew how to use it, and he would create views or lenses, and that was basically the only way we got things done. All of the salespeople, I was one of them, had to go to him and be like, please, Mr. Jay, would you create a view of … because none of us knew how to use it. It was incredibly complex.
And then we tried to go to this other, this was 2004, we tried to go to this other web-based CRM thing. It was called absoluteBUSY. And if you wanted a field created, you had to log a ticket, and then the person that created the CRM in Sweden would create the field. And it would be like you. It was sometimes months. I just need another phone field. How hard can that be?
Chris Emmett:
Yeah.
Mike:
And then I went to another company and they had Salesforce, and I was like, oh, click, click, boom. And I remember, I was like, this is almost too easy to create a field. This is dangerous.
Chris Emmett:
Dangerous is the right word. It is. I’d like to think most Salesforce professionals go through some … it’s not quite like the seven stages of grief, but it’s the seven stages of Salesforce acceptance. You can’t go in. You’re very skeptical, and then you’re like, oh, wow, this can create whatever fields I want, whatever data points I want, whatever automation I want, whatever reporting I want. And then at a certain point, you go, oh, wait a minute, I have just created a monster for myself, and then you learn to think before you build.
Mike:
I feel like that’s a good starting point. Is that where you find most people go off the rails, is maybe they get Salesforce and like, you know what, let’s start fresh and they create too many fields? And then because there’s too many fields and they’re in such a hurry that they get to bad data.
Chris Emmett:
I look at it a different way. It may be my exposure. I deal with a lot of existing companies. I’ve not really dealt with a lot of brand new companies. So a lot of existing companies, especially if they’re at least 40 or 50 years old. They’ve got a lot of older systems. They might have a bit of SAP. They might have mainframes kicking around. They might have things written in COBOL or FORTRAN. I would even deal with companies today that have things in MSX, believe it or not.
So the danger isn’t, oh, let’s just create everything in Salesforce. The danger is 14 years ago, Derek created this field. We don’t know what it does. We don’t know where it’s hooked up to. It’s not documented anywhere, but we feel like we should pull it over. So the real danger is actually migrating everything. If you don’t know what that data point is, you don’t know what use it is, you can’t validate it, and you can’t use it in any meaningful way. Because if you don’t understand, then to bring it to the point of this pod, if you don’t understand what that data is, what it means, what it’s doing for your business, how can an AI agent understand that? An AI agent is not magical. It’s not telepathic. It reads the information as if it’s a human. It tries to interpret that information. So you’ve got to know what it means so your AI agent can be told what it means as well.
Mike:
Yeah. I think people forget that new systems or new features won’t save them if they haven’t started to save themselves.
Chris Emmett:
Yeah. This is genuine, by the way. I was thinking this morning as I was just leaving the gym, so I can sound like I’m healthy.
Mike:
Oh, wow. Fancy. My weightlifting for this morning has been coffee cups.
Chris Emmett:
Yeah. So I was thinking this morning like, oh, what intelligent things can I say on this pod? I was actually thinking about the meaning of information technology as a term for a department within a business. Information technology, what does that actually mean? It’s not about the data. It is, but it’s not about the data. It’s not data technology. It’s not technology that drives data to help a business. It’s information technology. And the information that you derive from that data is the most important thing because you can throw data into a data lake or just into a database and just have it stored there. But if you can’t derive useful information from it, and similarly, if an AI agent can’t derive useful information from it, it’s actually pointless.
Mike:
Yeah. I’ve once heard information described as data plus context.
Chris Emmett:
Yeah, absolutely.
Mike:
And so if your data is bad, then it really lacks context, or you could say it doesn’t provide context.
Chris Emmett:
Yeah. It is all about that context. And again, just to bring it back to the whole Agentforce thing. Agentforce needs to understand that context. It needs to be able to derive some meaning from it. So just as a random example, let’s say you’ve got an opportunity record, which is great, and you’ve got a date on that opportunity. Cool. Agentforce, can you tell me if there’s a date on this record? Agentforce might go away and it might find that date and it might go, cool, Chris, I found a date. It’s July the 24th, 2025.
Now, if that’s written in a note without any context, and I say Agentforce, what does this mean? It’s going to go, well, I don’t know. A date in the future. But if it’s against the close date field, it’s immediately got context, and it immediately can derive something from that. It can say, oh, right, okay, so this is the close date. I can see that we are in a negotiation stage. We’ve got one more stage after this, which is sign contract. It’s the 24th. It’s a few weeks away, or at least from when we’re recording. I understand a bit of context. I understand that there’s another stage ahead of this, and I’ve immediately got more information other than just a record with a random date. That context is everything, especially for an LLM.
Mike:
But the irony is, and I’ve done other episodes on this, you could have a close date, continuing your idea, of July 24th. In the close date field, except your company doesn’t use that, that’s the date that this part of the opportunity is going to close. But then maybe there’s a follow-on implementation stage, and that’s the context in which you use that. However, if you haven’t given your AI any kind of information about that, the context at which you use the close date, ironically, because you have a date in that field, it’s bad data, even though it looks like good data.
Chris Emmett:
Yeah, absolutely. And the point of the TDX talk I did with Jonathan, and I’m sure Jonathan has already mentioned this in last week’s episode, the data is important, the data has to be valid, but it also has to sit within contextual metadata. Because if you have a field on your opportunity that is called installation date or delivery date or deployment date, whatever, and you have an accurate data against that, again, you’re giving meaning to that data. You’re giving meaning to Agentforce so it can interpret it and give you useful information because it is about information. It’s not about the data. It’s the information. In fact, if I had a time machine, I would go back three weeks, redo my TDX talk, and it would be called prepping your information like a pro, and I guess your meta information.
Mike:
Yeah. But let’s pivot into that because it’s time to get actionable and less heady. And I’ve heard this a thousand times, it can feel like boiling the ocean to clean your data. What is your approach that you would suggest people use to start cleaning their data in preparation for deploying Agentforce?
Chris Emmett:
Yeah, sure. So again, unless you are that three-day old company with three people in and the data is perfect, it’s probably safe to assume you are sat on a mountain of data. If you are a relatively small company, maybe it’s tens of thousands of records. If you are a medium or enterprise, you might have hundreds of thousands or millions of records. I certainly worked for a company that had 30 million accounts in the system. That is a lot of data, and you cannot possibly begin to go through that top to tail, making sure that every field is correct and accurate and has meaning.
So how do you actually break that down? You’re right. You cannot boil the ocean. We start off by thinking about the actions and the intents that you want your agents to do. So if you want your agents to write an email to a customer if their opportunity is within five days of the close date and they’ve not signed yet, well, what do you need for that? You don’t need 300 fields that might be on the opportunity page or 300 fields in that account. You might need the opportunity’s name. You need the stage of the opportunity. You need the close date. You need the account, and maybe the primary contact of that account. That’s five pieces of information.
And then you do not really need to think about all of the old opportunities because this is about an action where you are emailing people for opportunities that are about to expire or about to close. So immediately, you’ve gone from a million records, let’s say, you’ve got it down to a thousand records, and then you’re only looking at those five pieces of data. So you got it down to a thousand records and you got it down to five pieces of information on those thousands records. So I’m not going to do the math in my head because I’m terrible at that, but it’s-
Mike:
Nobody should do math live.
Chris Emmett:
Yeah. You’re probably looking at about less than 1% of data because you’re thinking about the intent of that action, that AI action. You’re thinking about exactly what pieces of information you need to help that agent. And you’re making sure that that specific data is accurate.
My theory behind this, my thesis, is by doing that, by getting rid of 99% of the data that you don’t need to worry about today, you can get agents and agent actions out to your users quicker, which means they’re happier. They’re trying new things. You’re getting feedback on those new things. And it means that you can improve more processes because you’re getting more feedback. You’re getting more insight into what is helping people, what’s hindering people. And you’re doing that because you’re just targeting the data that matters, and everything else can wait until it’s actually needed.
Mike:
That is probably the most concise spot-on answer I have heard in a long time.
Chris Emmett:
I felt like I was talking for about 20 or 30 minutes on that.
Mike:
No, you weren’t. So targeting the data that matters.
Chris Emmett:
Yes.
Mike:
Those words, put that on a shirt. Somebody needs to wear that shirt at Dreamforce because … so that as you were talking through, I was realizing you’re really making it from boiling the ocean to boiling a cup of water. Everything doesn’t have to be perfect for you to start this project. Just the part that you need to worry about.
And I was thinking back to last week. So in Iowa, in the US where I live, we had a really bad frost all winter. We didn’t get the snow cover that we usually do, and some of my landscaping plans didn’t make it because the roots were just burned by the frost, but not all of them. Some of them were hardy and they’re fine. And so I called my landscape company. I was like, I need to replace all these, and plus I want different ones anyway. He’s like, good, because if we go through another winter like this, we’re just going to be replacing them. And I promise you this gets somewhere.
But much like your analogy, so the landscape company came out, and just in the area that they needed to replant those bushes, they scraped all the gravel, leveled the bed, put new tarp down, replanted the bushes, put the gravel back. They didn’t have to clean the entire planter bed and scoop all the gravel out and dig up all the bushes and start from scratch. They only had to do the part that mattered. And I felt like that was like, wow, that’s like a real life scenario of if we’re going to implement this and we’re going to really laser focus on this one part of it, let’s do that.
But the other key thing that you said that nobody has said on a podcast about cleaning data is you improve the process while you do it. Because if you are not going to improve the process that led you to the bad data, you’re always going to be cleaning data. It’s almost like sending a janitor out to a sports stadium to pick up trash because there’s no trash cans. Well, if you’re not going to sit down and say, okay, how do we put trash cans out so that trash isn’t everywhere? All you’re doing is sending the janitor back out to clean up trash. You’re not actually fixing the problem that led to the trash being everywhere.
Chris Emmett:
Yeah. It’s interesting you say that because I genuinely hadn’t really thought that far ahead.
Mike:
Yeah, you had-
Chris Emmett:
Probably not.
Mike:
… at night in your brain while you were sleeping. Chris’ brain is like, I got this idea about this process and then it was going to sound smart, and then it surfaced while you were working out.
Chris Emmett:
Yeah. People are going to be so disappointed when they see a picture of me because I’m making it sound like I’m some sort of gym monkey. Gym monkey, that’s probably not the right word.
Mike:
Well, I could throw your picture in AI.
Chris Emmett:
Yeah, yeah, yeah. You should.
Mike:
You’re like Arnold Schwarzenegger.
Chris Emmett:
You should. Yeah. I’ll look exactly like Arnold Schwarzenegger.
Mike:
Yeah, absolutely.
Chris Emmett:
I look more like Captain America at the start of the film.
Mike:
That is the best description. I look more like Captain America, but at the start of the film.
Chris Emmett:
Thanks. I’m glad you agree. I’m really glad you agree. Oh my, what was the point I was trying to … yeah, yeah. Okay. Yeah, back on track.
So the whole point of IT, the whole point of computer systems is to improve things. So if you genuinely are taking an existing process and you are just throwing at an agent and say, robots, do my bidding, without changing anything, you really are missing the point of IT. It’s about evolution, it’s about simplification, and it’s about making things more efficient. So why wouldn’t you try and make it more efficient in the process of moving it over? Because if you don’t, you are just lifting and shifting a process, and that is not going to help anyone. It’s not going to fix any errors. The worst case of that is that you program an agent with all of the issues and problems that a human would have, so you are just replicating those problems. You’ve got to improve. You’ve got to evolve.
Mike:
Do you find, because you mentioned you work with mature companies that often when they’re doing a Salesforce implementation or they’re bringing you in or they’re thinking about something, that they’re stuck thinking within the constraints of the legacy system that they have?
Chris Emmett:
I was going to be kind. That sounds terrible.
Mike:
Well, you don’t have to … I don’t want you to-
Chris Emmett:
I was going to be kind and say the majority of companies will try and hold onto legacy, and there’s a few that don’t. But actually, I would be as bold as to say every single customer I’ve worked with, not necessarily through any fault of their own, holds onto the past. And it is either because there’s things that they don’t know, there might be fields or tables or data points or integrations that no one knows why they exist, and they’re afraid of removing them. And in the data, they don’t know what the data means or whether it’s useful. It’s just there. So they don’t don’t know know how to cleanse it. They don’t know if it’s got any meaning.
But then you’ve got companies who are just in a bind where they’ve got a 30-year-old system and they do not have the budget to replace it, but Salesforce has got to integrate with it, and the design decisions 30 years ago for that system that Salesforce has to adhere to. So sometimes, it is because the systems that you have to hook Salesforce into are just bound by old design, and then you have to introduce those bad data decisions into Salesforce.
Mike:
I could see that. I was also thinking back as you were saying that. About a week ago, I was standing at a rental counter and I happened to see their screen and it was still a DOS screen, like the whole … I was like, wow, you still work on that? I’m fairly certain the car you’re about to hand me keys to has a stronger computing system in it than what you’re sitting at right now.
Chris Emmett:
Probably. The phone in that pocket is probably more powerful.
I definitely worked for a company, I can’t say their name, I’m pretty sure, but I worked for a company where they were using an MS-DOS program because the person who wrote it, he wrote it in his garage and then retired in the ’90s. And if he’s still with us, he’s probably, I’d like to think, on a yacht somewhere because the software developers in the ’70s probably earned quite a lot of money. He’s enjoying life and not thinking about it. But that company was stuck with that MS-DOS program because it did a vital thing and it can’t be removed. It can’t be replaced because they know it does a vital process, but they don’t know how it works. And then it becomes a business risk. It’s like, do you risk the operation of the business to try and rebuild this or do you just leave it?
And then another thing I was thinking about this morning, I was thinking a lot of stuff at the gym.
Mike:
Apparently. Holy cow. But to be fair, you’re always thinking like this. This wasn’t just this morning.
Chris Emmett:
You are right. You are right. I’m always thinking this. My thought process flips between IMDb trivia and Salesforce. I was thinking that I have worked with a lot of companies where it’s common practice to buy old equipment from eBay.
Mike:
Oh, wow.
Chris Emmett:
If you’re a retailer and you’ve got systems that haven’t been made in 10, 15 years, but you need that equipment, your only course is to scour eBay. I’ve worked with two or three companies that have done that, and it becomes a risk. But again, not to be too mean against them, it’s more often than not a budget constraint. So bad systems and bad data with those bad systems is not necessarily because people have just made bad choices or due diligence isn’t being put into it. More often than not, it’s about legacy systems and budget, and that’s a difficult thing to overcome.
Mike:
Especially the budget part.
Chris Emmett:
Yeah.
Mike:
I was just thinking back. I worked retail in the ’90s. And retail in the ’90s, if you were a smaller retailer, it was common practice for smaller retailers, when the big box companies would change systems, they would have a clearing out, and smaller retailers could buy enough POS systems that worked and then enough that didn’t work for parts and literally upgrade their whole system, even though it was used systems. And that was common in retail in the ’90s. Now, it’s all changed. It’s all online. I’m sure that doesn’t happen. But I remember working for a couple retailers and like, oh, we’re getting new cash registers, and they would show up, and why are some of the keys sticky? And they’re like, well, they’re not new. THey’re new to us.
Chris Emmett:
Yeah.
Mike:
I wonder what’s creeping around inside of those old systems.
You bring up a good point. I think I’ve done a lot of podcasts on cleaning data and cleaning metadata, but what’s interesting about this one, to put a point on it, is you pointed at something that none of the others have brought up, which is it’s not necessarily bad user’s fault that we have bad data. It’s sometimes a systematic failure of bad decisioning or a lack of budgeting priorities that leads to it. It’s not just Bob, the sales guy, who puts his call notes in the phone field because that’s what he does because he pays no attention. It’s sometimes working within the constraints of the budget. You’re probably doing that now, even doing a Salesforce implementation with AI. You’re like, well, we would love to do all this, one through five, but we can afford one now and maybe two later. They maybe even start on the second project, and then that’s as far as they get on their list of one through five.
Chris Emmett:
Yeah. You know what? If Bob wants to put his data in the wrong field, that is, in my opinion, that is probably a low impact. It’s a quick fix because either you can tell him to move it or you can, if it’s consistent, you can go in with Data Loader, export it in Excel, copy it from one column into another column, re-upload, you’re done. For me, it’s about all of this historical data that you may be pulling into Salesforce where you don’t necessarily understand its context. You don’t necessarily understand its validity.
And that’s where the whole point of trying to identify what’s the actual action that you want your AI agents to carry out, and what are the data points that that action needs to interact with? Okay. And then let’s tackle those data points and turn those data points, to go way back to the start of the conversation, you turn those data points into information, because you cannot boil the ocean. And the majority of companies, at least the ones I’ve dealt with, have a sea of data that just either makes no sense or comes from old systems or comes from unnecessary decisions. And you cannot … there’s no business case that will ever be put forward to say, we need to spend 10 years improving all of this data. But there definitely would be a business case that says, we need to spend a month chipping away at this opportunity data or this account data so we can deliver this agentic functionality.
Mike:
Right. Spot on. Now, I want to end on a fun note. You said IMDb data trivia.
Chris Emmett:
I like IMDb.
Mike:
Is it trivia about the website IMDb or is it trivia about movies and TV shows?
Chris Emmett:
Oh, wow. I’m really upset that I don’t know the history of the website now.
Mike:
Oh, I kind of do.
Chris Emmett:
Oh, do tell.
Mike:
So the Cliff Notes version is a guy started, I don’t know when, but it was started pretty early in the … maybe even pre-internet. I want to say he either had a huge collection of VHS tapes or he ran a rental movie store, and he created this literal database of movies and actors and actresses and roles and directors, because he found it fascinating to see what, like we’ve all heard, the seven, six degrees or Six Degrees of Kevin Bacon, like connecting people to different movies. He found that fascinating. And I want to say early on, Amazon bought it and curated it into the big website that it is now. That’s the amount of history that I know about it.
Chris Emmett:
Well, I can add to that.
Mike:
But it was one person’s passion project.
Chris Emmett:
That’s crazy. I can add to that in a related way, in a very Salesforce related way.
Mike:
Oh.
Chris Emmett:
And if anything, that’s why we’re all here. So during COVID, we were all watching a lot of TV at home, I’m pretty sure. I was. And I wanted to track what I was watching. I wanted to be able to give it some fun ratings. And I could have done that in a notebook or on a spreadsheet or in my notes on my phone. But 23-time certified, of course, I’m going to build it on Salesforce.
Mike:
Yeah. You have 600 badges. You can’t not.
Chris Emmett:
So I built it in a developer edition, salesforce.org. I was tracking all of the movies, all of the TV programs I was watching during COVID. And to just take it a step further, I didn’t build it, I ripped it off the internet, an integration into IMDb to pull the correct title, the director, the Rotten Tomatoes rating, I think, and the synopsis. So all I needed to do is type in the title and my own personal rating. And when it saved, it pulled back the poster, all of the cast, the directors, genre, everything. And that took one weekend. If people are listening to this and they never touched Salesforce before, that’s how quick it is to build something in Salesforce. I did that in a weekend.
Mike:
So you’re my long-lost brother, I swear. No lie. And I can show this to you. I’ll show this to you at Dreamforce, hand to god. I was in the same scenario in December of 2013. I took a long vacation break. And for many years, up until that point, I was hooked on Top Gear, and I thought I should keep track, because you could also get all of the Top Gear episodes off of iTunes, off of that, it was called iTunes at that point.
Chris Emmett:
Yeah.
Mike:
And I was like, I should keep track of this so that I know which ones to buy, which ones I like and have a whole … And so I literally spent 30 seconds. I was like, I could do this on a spreadsheet, but wait a minute, let’s do this in a developer org in Salesforce. And I created a Top Gear app, and I branded the whole salesforce.org Top Gear with Jeremy Clarkson and all of them. I had cars and stars and lap times and episode ratings and titles. I ended up, for almost a whole weekend straight, just non-stop binging Top Gear and building this org.
Chris Emmett:
That is awesome.
Mike:
And now, it’s one of those where I have to remember every month I log into it to make sure that that org dies. I would cry if that org went away. I seriously would. But yeah, I’m not fancy and smart like you. I don’t know what I’d integrate it to. I guess I could integrate it to IMDb to pull that stuff in, but I more just wanted it of my own information. But it was so old. It was back in the day when you used to stick image files in resources.
Chris Emmett:
Oh, yeah.
Mike:
And then you would do an image. You would create a formula field that returned an image, and that’s how I did star ratings on the page.
Chris Emmett:
That is still a valid way to do it. I don’t care what anyone says.
Mike:
It just feels so old school. But I remember showing it to somebody and they’re like, how did you get stars to show up on the page? It was like, well, let me tell you.
Chris Emmett:
That’s what it’s about. I don’t care what anyone says. Salesforce for me is about, as a profession, you are obviously going into businesses and building out their systems to improve the way they work, but for me, the beauty of a dev or a developer org is to be able to just do my own personal random projects where I might have an idea where it’s a weekend. I want to see if it works. And nine out of 10 ideas may burn and fail, but that one idea out of 10 might be just gold. And all of a sudden, I have a movie tracker that I can use, and I’ve learned something. And you know what? I can then take some of that integration knowledge to my next customer. It’s about just trying stuff out, exploring, being bold, being crazy.
Mike:
I would argue even those other nine are still fun when they fail because there’s a moment when you’re like, I think this is going to work. Well, I know it’s not going to work now, and now I know what not to do.
Chris Emmett:
Exactly.
Mike:
Chris, this was a blast.
Chris Emmett:
It was amazing.
Mike:
I’m so glad … I think we talked about some clean data.
Chris Emmett:
Data is important. Data and information, very important. That’s the takeaway.
Mike:
Yeah. We got to get some shirts made for Dreamforce. Thanks so much for being on the podcast. And I’m fairly certain you and Jonathan will have long legs in presenting your clean data and metadata information to the world.
Chris Emmett:
Yeah, no doubt. Thanks for having me.
Mike:
All right. So that’s a wrap for today’s therapy session with Chris Emmett. I can’t help but say we laughed, we learned, we IMDb’d. I will say after the call, we probably went down a rabbit hole of movie fun facts from television shows from the ’80s. That was incredibly fun. But again, if you’re secretly running a DOS system or still mourning a lost developer org, just remember, not all your data deserves your attention. Boil a cup of water, not the ocean. And maybe for fun, build a movie tracker while you’re at it or a television show tracker.
So if you love this episode, be sure to give us a review on iTunes. Share it with another admin who needs a nudge in the better direction of cleaning data. And until next time, we’ll see you in the cloud.
The post Cleaning Data for AI Starts With Context, Not Perfection appeared first on Salesforce Admins.
View all episodes
By Mike Gerholdt
4.7
200200 ratings
Today on the Salesforce Admins Podcast, we talk to Chris Emmett, Salesforce Solution Architect at Capgemini. Join us as we chat about how to clean up your data to prepare your org for Agentforce, and why data without context is useless.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Chris Emmett.
AI requires clean data
I caught up with Chris hot on the heels of his TDX London session, “Prep Like a Pro: Clean Data and Metadata for Agentforce.” He’s an experienced Salesforce consultant who has helped countless organizations of all sizes reboot their business processes.
As Chris explains, unless you have a company of five people that started last week, your org probably needs some data cleanup. And if you want to get started with Agentforce, you need to do the work to make sure the agents you build can understand your data and use it to generate actionable insights. After all, if you can’t derive useful information from your data, then it’s useless.
Why cleaning data can feel like boiling the ocean
When I worked in sales, we used a CRM that was so complicated that only one guy at our company knew how to use it. Talk about a bottleneck!
The truth is, if your business has been around for a little while, you’ve probably inherited all sorts of legacy data. Maybe it’s some random field created by that one guy in the 90s who didn’t document anything, or a legacy system like SAP or MSX that is essential to your day-to-day operations.
Chris has seen it all, and it can often feel like cleaning up all that data is akin to boiling the ocean. It’s a monumental task with no end in sight, let alone getting the organizational buy-in to do it in the first place.
A practical way to start cleaning your data
Chris recommends focusing your data cleanup strategy on the functionality you want to build in Agentforce. For example, if you want an agent to email a customer when their opportunity is five days from the close date and still unsigned, what data do you actually need?
You don’t need the 300 fields that might be on the opportunity page, or the 300 fields in that account. You might need the opportunity’s name, the stage of the opportunity, the close date, the account, and maybe the primary contact of that account. That’s five pieces of information.
Suddenly, you don’t need to boil the entire ocean—you just need to boil a cup of water. So start small, focus on the functionality your data cleanup project will deliver, and get the ball rolling. Trust that the things you build with Agentforce will speak for themselves, and you’ll be able to generate momentum to clean up your data project by project.
Make sure to listen to our full conversation with Chris to learn more about how to clean up your data and provide context for AI agents. And don’t forget to subscribe to the Salesforce Admins Podcast so you never miss an episode.
Podcast swag
Learn more
Admin Trailblazers Group
Social
Full show transcript
Mike:
This week on the Salesforce Admins Podcast, we’re joined by Chris Emmett, consultant, data enthusiast and Salesforce evangelist accidentally. Well, Chris is passionate about data, and he takes us on a journey from legacy systems and those DOS screens to databases and AI-powered actions, all while sharing practical advice on how to clean up your data without making it feel like you’re boiling the ocean. So if you ever wondered how to prepare your org for Agentforce or why data without context is basically useless, this one is for you. So listen in, share it with somebody who maybe is swimming in a sea of records. So let’s have Chris swim his life raft over and let’s get him on the podcast.
So Chris, welcome to the podcast.
Chris Emmett:
Thanks for having me.
Mike:
Yeah. Well, it’s good to be here. I got a note from Jennifer Lee. She saw the session that you and Jonathan did at TDX London about cleaning data and cleaning metadata, and last week, we talked with Jonathan about cleaning metadata, so this week, of course, we’re going to clean our data because who doesn’t have clean data? But let’s start off first with a little bit about you, Chris. How did you get into Salesforce and why are you so passionate about clean data?
Chris Emmett:
Yeah. Before I start that, you said who doesn’t have clean data? It really should be the other way around.
Mike:
I know. It was a rhetorical question to get people to listen and be like, oh, but there could be that one person that tunes into the podcast, like I don’t need to listen to this. My data is sparkling.
Chris Emmett:
You know what? If there is a company out there with five people who just opened up last week, their data is going to be impeccable, and they do want to listen to this podcast. They can tune out.
Mike:
Right.
Chris Emmett:
I started in the ecosystem in a weird and wonderful way because a lot of … you’ve obviously got your accidental admins that find a weird and wonderful way into Salesforce. I was an accidental consultant. So I started out fresh from university as a desktop support engineer, just fixing Windows and fixing printers and fixing Office, and it was great, fun. Then I did a bit of work as a developer, cutting my teeth on Visual Basic 6.0 and .NET all the way back in 2008. But that entire team was made redundant to make way for an off-the-shelf manufacturing system. And as an SME, as an expert in how that old system worked, I was brought in, and then the project manager on that project quit, and I became a project manager, and I was a project manager for 10 years.
And then about 10 years later, around 2016, I was working for a company and we were really interested in changing how we managed products. We wanted a brand new system to manage our projects, and one of the products that we looked at was Salesforce. And my word, I was blown away. Literally within one day of using Salesforce, I was creating formula fields and workflow rules as they were back then, and I just fell in love. And this is coming from a person who had spent the best part of a decade dealing with systems where if you needed to add a field, it would take two, three months to get through, and I was dealing with a system where I could start a sentence, explaining to someone what Salesforce was, and by the time I’ve finished that sentence, I could have created a field. I could have expanded that data model, and I fell in love with it.
The company I was working for did not fall in love with it, and that pilot failed. It fell by the wayside, but I was hooked. I was like, man, I need to change my career. So I start looking for project management jobs within the Salesforce space. I had never project managed software before. I had project managed big old factory systems that were very waterfall in their approach. I had never done agile before, so I was applying and applying and applying, getting nowhere. I was probably applying for about eight months, and then this small company in Cambridge were like, dude, you are not great for a project manager but you seem really enthusiastic. Had you considered being a consultant? And I went, but yeah, because it’s not managing the projects that I love. It’s the system that I love. Salesforce is a platform that I love. I want to be able to empower other companies to improve themselves through Salesforce.
So I got a job there as a consultant, and that was 2017, and I have just been building up and flourishing, and since then, I’ve got 600 odd badges on Trailhead and I think 23 certs now. So I’ve just gone all in and built my career up.
Mike:
Wow. Holy cow. That’s a lot. I feel such a kinship with you because I joined … when I started doing Salesforce stuff, I worked at a publishing company, and we had this Apple-based CRM. I can’t even remember what it was called, but there was one guy in the office that still knew how to use it, and he would create views or lenses, and that was basically the only way we got things done. All of the salespeople, I was one of them, had to go to him and be like, please, Mr. Jay, would you create a view of … because none of us knew how to use it. It was incredibly complex.
And then we tried to go to this other, this was 2004, we tried to go to this other web-based CRM thing. It was called absoluteBUSY. And if you wanted a field created, you had to log a ticket, and then the person that created the CRM in Sweden would create the field. And it would be like you. It was sometimes months. I just need another phone field. How hard can that be?
Chris Emmett:
Yeah.
Mike:
And then I went to another company and they had Salesforce, and I was like, oh, click, click, boom. And I remember, I was like, this is almost too easy to create a field. This is dangerous.
Chris Emmett:
Dangerous is the right word. It is. I’d like to think most Salesforce professionals go through some … it’s not quite like the seven stages of grief, but it’s the seven stages of Salesforce acceptance. You can’t go in. You’re very skeptical, and then you’re like, oh, wow, this can create whatever fields I want, whatever data points I want, whatever automation I want, whatever reporting I want. And then at a certain point, you go, oh, wait a minute, I have just created a monster for myself, and then you learn to think before you build.
Mike:
I feel like that’s a good starting point. Is that where you find most people go off the rails, is maybe they get Salesforce and like, you know what, let’s start fresh and they create too many fields? And then because there’s too many fields and they’re in such a hurry that they get to bad data.
Chris Emmett:
I look at it a different way. It may be my exposure. I deal with a lot of existing companies. I’ve not really dealt with a lot of brand new companies. So a lot of existing companies, especially if they’re at least 40 or 50 years old. They’ve got a lot of older systems. They might have a bit of SAP. They might have mainframes kicking around. They might have things written in COBOL or FORTRAN. I would even deal with companies today that have things in MSX, believe it or not.
So the danger isn’t, oh, let’s just create everything in Salesforce. The danger is 14 years ago, Derek created this field. We don’t know what it does. We don’t know where it’s hooked up to. It’s not documented anywhere, but we feel like we should pull it over. So the real danger is actually migrating everything. If you don’t know what that data point is, you don’t know what use it is, you can’t validate it, and you can’t use it in any meaningful way. Because if you don’t understand, then to bring it to the point of this pod, if you don’t understand what that data is, what it means, what it’s doing for your business, how can an AI agent understand that? An AI agent is not magical. It’s not telepathic. It reads the information as if it’s a human. It tries to interpret that information. So you’ve got to know what it means so your AI agent can be told what it means as well.
Mike:
Yeah. I think people forget that new systems or new features won’t save them if they haven’t started to save themselves.
Chris Emmett:
Yeah. This is genuine, by the way. I was thinking this morning as I was just leaving the gym, so I can sound like I’m healthy.
Mike:
Oh, wow. Fancy. My weightlifting for this morning has been coffee cups.
Chris Emmett:
Yeah. So I was thinking this morning like, oh, what intelligent things can I say on this pod? I was actually thinking about the meaning of information technology as a term for a department within a business. Information technology, what does that actually mean? It’s not about the data. It is, but it’s not about the data. It’s not data technology. It’s not technology that drives data to help a business. It’s information technology. And the information that you derive from that data is the most important thing because you can throw data into a data lake or just into a database and just have it stored there. But if you can’t derive useful information from it, and similarly, if an AI agent can’t derive useful information from it, it’s actually pointless.
Mike:
Yeah. I’ve once heard information described as data plus context.
Chris Emmett:
Yeah, absolutely.
Mike:
And so if your data is bad, then it really lacks context, or you could say it doesn’t provide context.
Chris Emmett:
Yeah. It is all about that context. And again, just to bring it back to the whole Agentforce thing. Agentforce needs to understand that context. It needs to be able to derive some meaning from it. So just as a random example, let’s say you’ve got an opportunity record, which is great, and you’ve got a date on that opportunity. Cool. Agentforce, can you tell me if there’s a date on this record? Agentforce might go away and it might find that date and it might go, cool, Chris, I found a date. It’s July the 24th, 2025.
Now, if that’s written in a note without any context, and I say Agentforce, what does this mean? It’s going to go, well, I don’t know. A date in the future. But if it’s against the close date field, it’s immediately got context, and it immediately can derive something from that. It can say, oh, right, okay, so this is the close date. I can see that we are in a negotiation stage. We’ve got one more stage after this, which is sign contract. It’s the 24th. It’s a few weeks away, or at least from when we’re recording. I understand a bit of context. I understand that there’s another stage ahead of this, and I’ve immediately got more information other than just a record with a random date. That context is everything, especially for an LLM.
Mike:
But the irony is, and I’ve done other episodes on this, you could have a close date, continuing your idea, of July 24th. In the close date field, except your company doesn’t use that, that’s the date that this part of the opportunity is going to close. But then maybe there’s a follow-on implementation stage, and that’s the context in which you use that. However, if you haven’t given your AI any kind of information about that, the context at which you use the close date, ironically, because you have a date in that field, it’s bad data, even though it looks like good data.
Chris Emmett:
Yeah, absolutely. And the point of the TDX talk I did with Jonathan, and I’m sure Jonathan has already mentioned this in last week’s episode, the data is important, the data has to be valid, but it also has to sit within contextual metadata. Because if you have a field on your opportunity that is called installation date or delivery date or deployment date, whatever, and you have an accurate data against that, again, you’re giving meaning to that data. You’re giving meaning to Agentforce so it can interpret it and give you useful information because it is about information. It’s not about the data. It’s the information. In fact, if I had a time machine, I would go back three weeks, redo my TDX talk, and it would be called prepping your information like a pro, and I guess your meta information.
Mike:
Yeah. But let’s pivot into that because it’s time to get actionable and less heady. And I’ve heard this a thousand times, it can feel like boiling the ocean to clean your data. What is your approach that you would suggest people use to start cleaning their data in preparation for deploying Agentforce?
Chris Emmett:
Yeah, sure. So again, unless you are that three-day old company with three people in and the data is perfect, it’s probably safe to assume you are sat on a mountain of data. If you are a relatively small company, maybe it’s tens of thousands of records. If you are a medium or enterprise, you might have hundreds of thousands or millions of records. I certainly worked for a company that had 30 million accounts in the system. That is a lot of data, and you cannot possibly begin to go through that top to tail, making sure that every field is correct and accurate and has meaning.
So how do you actually break that down? You’re right. You cannot boil the ocean. We start off by thinking about the actions and the intents that you want your agents to do. So if you want your agents to write an email to a customer if their opportunity is within five days of the close date and they’ve not signed yet, well, what do you need for that? You don’t need 300 fields that might be on the opportunity page or 300 fields in that account. You might need the opportunity’s name. You need the stage of the opportunity. You need the close date. You need the account, and maybe the primary contact of that account. That’s five pieces of information.
And then you do not really need to think about all of the old opportunities because this is about an action where you are emailing people for opportunities that are about to expire or about to close. So immediately, you’ve gone from a million records, let’s say, you’ve got it down to a thousand records, and then you’re only looking at those five pieces of data. So you got it down to a thousand records and you got it down to five pieces of information on those thousands records. So I’m not going to do the math in my head because I’m terrible at that, but it’s-
Mike:
Nobody should do math live.
Chris Emmett:
Yeah. You’re probably looking at about less than 1% of data because you’re thinking about the intent of that action, that AI action. You’re thinking about exactly what pieces of information you need to help that agent. And you’re making sure that that specific data is accurate.
My theory behind this, my thesis, is by doing that, by getting rid of 99% of the data that you don’t need to worry about today, you can get agents and agent actions out to your users quicker, which means they’re happier. They’re trying new things. You’re getting feedback on those new things. And it means that you can improve more processes because you’re getting more feedback. You’re getting more insight into what is helping people, what’s hindering people. And you’re doing that because you’re just targeting the data that matters, and everything else can wait until it’s actually needed.
Mike:
That is probably the most concise spot-on answer I have heard in a long time.
Chris Emmett:
I felt like I was talking for about 20 or 30 minutes on that.
Mike:
No, you weren’t. So targeting the data that matters.
Chris Emmett:
Yes.
Mike:
Those words, put that on a shirt. Somebody needs to wear that shirt at Dreamforce because … so that as you were talking through, I was realizing you’re really making it from boiling the ocean to boiling a cup of water. Everything doesn’t have to be perfect for you to start this project. Just the part that you need to worry about.
And I was thinking back to last week. So in Iowa, in the US where I live, we had a really bad frost all winter. We didn’t get the snow cover that we usually do, and some of my landscaping plans didn’t make it because the roots were just burned by the frost, but not all of them. Some of them were hardy and they’re fine. And so I called my landscape company. I was like, I need to replace all these, and plus I want different ones anyway. He’s like, good, because if we go through another winter like this, we’re just going to be replacing them. And I promise you this gets somewhere.
But much like your analogy, so the landscape company came out, and just in the area that they needed to replant those bushes, they scraped all the gravel, leveled the bed, put new tarp down, replanted the bushes, put the gravel back. They didn’t have to clean the entire planter bed and scoop all the gravel out and dig up all the bushes and start from scratch. They only had to do the part that mattered. And I felt like that was like, wow, that’s like a real life scenario of if we’re going to implement this and we’re going to really laser focus on this one part of it, let’s do that.
But the other key thing that you said that nobody has said on a podcast about cleaning data is you improve the process while you do it. Because if you are not going to improve the process that led you to the bad data, you’re always going to be cleaning data. It’s almost like sending a janitor out to a sports stadium to pick up trash because there’s no trash cans. Well, if you’re not going to sit down and say, okay, how do we put trash cans out so that trash isn’t everywhere? All you’re doing is sending the janitor back out to clean up trash. You’re not actually fixing the problem that led to the trash being everywhere.
Chris Emmett:
Yeah. It’s interesting you say that because I genuinely hadn’t really thought that far ahead.
Mike:
Yeah, you had-
Chris Emmett:
Probably not.
Mike:
… at night in your brain while you were sleeping. Chris’ brain is like, I got this idea about this process and then it was going to sound smart, and then it surfaced while you were working out.
Chris Emmett:
Yeah. People are going to be so disappointed when they see a picture of me because I’m making it sound like I’m some sort of gym monkey. Gym monkey, that’s probably not the right word.
Mike:
Well, I could throw your picture in AI.
Chris Emmett:
Yeah, yeah, yeah. You should.
Mike:
You’re like Arnold Schwarzenegger.
Chris Emmett:
You should. Yeah. I’ll look exactly like Arnold Schwarzenegger.
Mike:
Yeah, absolutely.
Chris Emmett:
I look more like Captain America at the start of the film.
Mike:
That is the best description. I look more like Captain America, but at the start of the film.
Chris Emmett:
Thanks. I’m glad you agree. I’m really glad you agree. Oh my, what was the point I was trying to … yeah, yeah. Okay. Yeah, back on track.
So the whole point of IT, the whole point of computer systems is to improve things. So if you genuinely are taking an existing process and you are just throwing at an agent and say, robots, do my bidding, without changing anything, you really are missing the point of IT. It’s about evolution, it’s about simplification, and it’s about making things more efficient. So why wouldn’t you try and make it more efficient in the process of moving it over? Because if you don’t, you are just lifting and shifting a process, and that is not going to help anyone. It’s not going to fix any errors. The worst case of that is that you program an agent with all of the issues and problems that a human would have, so you are just replicating those problems. You’ve got to improve. You’ve got to evolve.
Mike:
Do you find, because you mentioned you work with mature companies that often when they’re doing a Salesforce implementation or they’re bringing you in or they’re thinking about something, that they’re stuck thinking within the constraints of the legacy system that they have?
Chris Emmett:
I was going to be kind. That sounds terrible.
Mike:
Well, you don’t have to … I don’t want you to-
Chris Emmett:
I was going to be kind and say the majority of companies will try and hold onto legacy, and there’s a few that don’t. But actually, I would be as bold as to say every single customer I’ve worked with, not necessarily through any fault of their own, holds onto the past. And it is either because there’s things that they don’t know, there might be fields or tables or data points or integrations that no one knows why they exist, and they’re afraid of removing them. And in the data, they don’t know what the data means or whether it’s useful. It’s just there. So they don’t don’t know know how to cleanse it. They don’t know if it’s got any meaning.
But then you’ve got companies who are just in a bind where they’ve got a 30-year-old system and they do not have the budget to replace it, but Salesforce has got to integrate with it, and the design decisions 30 years ago for that system that Salesforce has to adhere to. So sometimes, it is because the systems that you have to hook Salesforce into are just bound by old design, and then you have to introduce those bad data decisions into Salesforce.
Mike:
I could see that. I was also thinking back as you were saying that. About a week ago, I was standing at a rental counter and I happened to see their screen and it was still a DOS screen, like the whole … I was like, wow, you still work on that? I’m fairly certain the car you’re about to hand me keys to has a stronger computing system in it than what you’re sitting at right now.
Chris Emmett:
Probably. The phone in that pocket is probably more powerful.
I definitely worked for a company, I can’t say their name, I’m pretty sure, but I worked for a company where they were using an MS-DOS program because the person who wrote it, he wrote it in his garage and then retired in the ’90s. And if he’s still with us, he’s probably, I’d like to think, on a yacht somewhere because the software developers in the ’70s probably earned quite a lot of money. He’s enjoying life and not thinking about it. But that company was stuck with that MS-DOS program because it did a vital thing and it can’t be removed. It can’t be replaced because they know it does a vital process, but they don’t know how it works. And then it becomes a business risk. It’s like, do you risk the operation of the business to try and rebuild this or do you just leave it?
And then another thing I was thinking about this morning, I was thinking a lot of stuff at the gym.
Mike:
Apparently. Holy cow. But to be fair, you’re always thinking like this. This wasn’t just this morning.
Chris Emmett:
You are right. You are right. I’m always thinking this. My thought process flips between IMDb trivia and Salesforce. I was thinking that I have worked with a lot of companies where it’s common practice to buy old equipment from eBay.
Mike:
Oh, wow.
Chris Emmett:
If you’re a retailer and you’ve got systems that haven’t been made in 10, 15 years, but you need that equipment, your only course is to scour eBay. I’ve worked with two or three companies that have done that, and it becomes a risk. But again, not to be too mean against them, it’s more often than not a budget constraint. So bad systems and bad data with those bad systems is not necessarily because people have just made bad choices or due diligence isn’t being put into it. More often than not, it’s about legacy systems and budget, and that’s a difficult thing to overcome.
Mike:
Especially the budget part.
Chris Emmett:
Yeah.
Mike:
I was just thinking back. I worked retail in the ’90s. And retail in the ’90s, if you were a smaller retailer, it was common practice for smaller retailers, when the big box companies would change systems, they would have a clearing out, and smaller retailers could buy enough POS systems that worked and then enough that didn’t work for parts and literally upgrade their whole system, even though it was used systems. And that was common in retail in the ’90s. Now, it’s all changed. It’s all online. I’m sure that doesn’t happen. But I remember working for a couple retailers and like, oh, we’re getting new cash registers, and they would show up, and why are some of the keys sticky? And they’re like, well, they’re not new. THey’re new to us.
Chris Emmett:
Yeah.
Mike:
I wonder what’s creeping around inside of those old systems.
You bring up a good point. I think I’ve done a lot of podcasts on cleaning data and cleaning metadata, but what’s interesting about this one, to put a point on it, is you pointed at something that none of the others have brought up, which is it’s not necessarily bad user’s fault that we have bad data. It’s sometimes a systematic failure of bad decisioning or a lack of budgeting priorities that leads to it. It’s not just Bob, the sales guy, who puts his call notes in the phone field because that’s what he does because he pays no attention. It’s sometimes working within the constraints of the budget. You’re probably doing that now, even doing a Salesforce implementation with AI. You’re like, well, we would love to do all this, one through five, but we can afford one now and maybe two later. They maybe even start on the second project, and then that’s as far as they get on their list of one through five.
Chris Emmett:
Yeah. You know what? If Bob wants to put his data in the wrong field, that is, in my opinion, that is probably a low impact. It’s a quick fix because either you can tell him to move it or you can, if it’s consistent, you can go in with Data Loader, export it in Excel, copy it from one column into another column, re-upload, you’re done. For me, it’s about all of this historical data that you may be pulling into Salesforce where you don’t necessarily understand its context. You don’t necessarily understand its validity.
And that’s where the whole point of trying to identify what’s the actual action that you want your AI agents to carry out, and what are the data points that that action needs to interact with? Okay. And then let’s tackle those data points and turn those data points, to go way back to the start of the conversation, you turn those data points into information, because you cannot boil the ocean. And the majority of companies, at least the ones I’ve dealt with, have a sea of data that just either makes no sense or comes from old systems or comes from unnecessary decisions. And you cannot … there’s no business case that will ever be put forward to say, we need to spend 10 years improving all of this data. But there definitely would be a business case that says, we need to spend a month chipping away at this opportunity data or this account data so we can deliver this agentic functionality.
Mike:
Right. Spot on. Now, I want to end on a fun note. You said IMDb data trivia.
Chris Emmett:
I like IMDb.
Mike:
Is it trivia about the website IMDb or is it trivia about movies and TV shows?
Chris Emmett:
Oh, wow. I’m really upset that I don’t know the history of the website now.
Mike:
Oh, I kind of do.
Chris Emmett:
Oh, do tell.
Mike:
So the Cliff Notes version is a guy started, I don’t know when, but it was started pretty early in the … maybe even pre-internet. I want to say he either had a huge collection of VHS tapes or he ran a rental movie store, and he created this literal database of movies and actors and actresses and roles and directors, because he found it fascinating to see what, like we’ve all heard, the seven, six degrees or Six Degrees of Kevin Bacon, like connecting people to different movies. He found that fascinating. And I want to say early on, Amazon bought it and curated it into the big website that it is now. That’s the amount of history that I know about it.
Chris Emmett:
Well, I can add to that.
Mike:
But it was one person’s passion project.
Chris Emmett:
That’s crazy. I can add to that in a related way, in a very Salesforce related way.
Mike:
Oh.
Chris Emmett:
And if anything, that’s why we’re all here. So during COVID, we were all watching a lot of TV at home, I’m pretty sure. I was. And I wanted to track what I was watching. I wanted to be able to give it some fun ratings. And I could have done that in a notebook or on a spreadsheet or in my notes on my phone. But 23-time certified, of course, I’m going to build it on Salesforce.
Mike:
Yeah. You have 600 badges. You can’t not.
Chris Emmett:
So I built it in a developer edition, salesforce.org. I was tracking all of the movies, all of the TV programs I was watching during COVID. And to just take it a step further, I didn’t build it, I ripped it off the internet, an integration into IMDb to pull the correct title, the director, the Rotten Tomatoes rating, I think, and the synopsis. So all I needed to do is type in the title and my own personal rating. And when it saved, it pulled back the poster, all of the cast, the directors, genre, everything. And that took one weekend. If people are listening to this and they never touched Salesforce before, that’s how quick it is to build something in Salesforce. I did that in a weekend.
Mike:
So you’re my long-lost brother, I swear. No lie. And I can show this to you. I’ll show this to you at Dreamforce, hand to god. I was in the same scenario in December of 2013. I took a long vacation break. And for many years, up until that point, I was hooked on Top Gear, and I thought I should keep track, because you could also get all of the Top Gear episodes off of iTunes, off of that, it was called iTunes at that point.
Chris Emmett:
Yeah.
Mike:
And I was like, I should keep track of this so that I know which ones to buy, which ones I like and have a whole … And so I literally spent 30 seconds. I was like, I could do this on a spreadsheet, but wait a minute, let’s do this in a developer org in Salesforce. And I created a Top Gear app, and I branded the whole salesforce.org Top Gear with Jeremy Clarkson and all of them. I had cars and stars and lap times and episode ratings and titles. I ended up, for almost a whole weekend straight, just non-stop binging Top Gear and building this org.
Chris Emmett:
That is awesome.
Mike:
And now, it’s one of those where I have to remember every month I log into it to make sure that that org dies. I would cry if that org went away. I seriously would. But yeah, I’m not fancy and smart like you. I don’t know what I’d integrate it to. I guess I could integrate it to IMDb to pull that stuff in, but I more just wanted it of my own information. But it was so old. It was back in the day when you used to stick image files in resources.
Chris Emmett:
Oh, yeah.
Mike:
And then you would do an image. You would create a formula field that returned an image, and that’s how I did star ratings on the page.
Chris Emmett:
That is still a valid way to do it. I don’t care what anyone says.
Mike:
It just feels so old school. But I remember showing it to somebody and they’re like, how did you get stars to show up on the page? It was like, well, let me tell you.
Chris Emmett:
That’s what it’s about. I don’t care what anyone says. Salesforce for me is about, as a profession, you are obviously going into businesses and building out their systems to improve the way they work, but for me, the beauty of a dev or a developer org is to be able to just do my own personal random projects where I might have an idea where it’s a weekend. I want to see if it works. And nine out of 10 ideas may burn and fail, but that one idea out of 10 might be just gold. And all of a sudden, I have a movie tracker that I can use, and I’ve learned something. And you know what? I can then take some of that integration knowledge to my next customer. It’s about just trying stuff out, exploring, being bold, being crazy.
Mike:
I would argue even those other nine are still fun when they fail because there’s a moment when you’re like, I think this is going to work. Well, I know it’s not going to work now, and now I know what not to do.
Chris Emmett:
Exactly.
Mike:
Chris, this was a blast.
Chris Emmett:
It was amazing.
Mike:
I’m so glad … I think we talked about some clean data.
Chris Emmett:
Data is important. Data and information, very important. That’s the takeaway.
Mike:
Yeah. We got to get some shirts made for Dreamforce. Thanks so much for being on the podcast. And I’m fairly certain you and Jonathan will have long legs in presenting your clean data and metadata information to the world.
Chris Emmett:
Yeah, no doubt. Thanks for having me.
Mike:
All right. So that’s a wrap for today’s therapy session with Chris Emmett. I can’t help but say we laughed, we learned, we IMDb’d. I will say after the call, we probably went down a rabbit hole of movie fun facts from television shows from the ’80s. That was incredibly fun. But again, if you’re secretly running a DOS system or still mourning a lost developer org, just remember, not all your data deserves your attention. Boil a cup of water, not the ocean. And maybe for fun, build a movie tracker while you’re at it or a television show tracker.
So if you love this episode, be sure to give us a review on iTunes. Share it with another admin who needs a nudge in the better direction of cleaning data. And until next time, we’ll see you in the cloud.
The post Cleaning Data for AI Starts With Context, Not Perfection appeared first on Salesforce Admins.
More shows like The Salesforce Admins Podcast
View all
WSJ What’s News
4,362 Listeners
Bloomberg Surveillance
1,200 Listeners
Motley Fool Money
3,195 Listeners
Bloomberg Businessweek
431 Listeners
Masters of Scale
3,998 Listeners
Modern Wisdom
3,815 Listeners
Wall Street Breakfast
1,029 Listeners
The Journal.
5,942 Listeners
Morning Brew Daily
2,969 Listeners
The Prof G Pod with Scott Galloway
5,414 Listeners
Salesforce for Everyone by Talent Stacker
227 Listeners
The Hiring Edge
22 Listeners
AI For Humans: Making Artificial Intelligence Fun & Practical
248 Listeners
Salesforce Simplified
1 Listeners
Prof G Markets
1,127 Listeners