Excited to share a Startup Spotlight on Plural, a company in our Microsoft for Startups Pegasus portfolio tackling one of the toughest infrastructure problems in retail: keeping Kubernetes clusters healthy across thousands of stores, from the cloud to the edge.

The results from one national retailer's deployment stood out to me: fleet-wide upgrades that used to take six months now take two weeks, incident remediation twelve times faster, and a small platform team now supporting twice the store footprint without adding headcount.

For any retailer or CPG brand running physical infrastructure at scale, this is worth a read.

I took a closer look at PulpoAR, a beauty tech company using AI and AR to change how people shop for makeup and skincare. Their virtual try-on, shade matching, and skin analysis tools are producing real results, including a 12% average add-to-cart conversion rate.

Mordor Intelligence projects the virtual try-on market will grow from $15 billion in 2025 to $48 billion by 2030, yet McKinsey & Company found that only 10% of beauty executives use AI regularly today.

That gap is where Microsoft for Startups partners like PulpoAR come in

Retail and CPG run on a deceptively simple promise: 𝒕𝒉𝒆 𝒓𝒊𝒈𝒉𝒕 𝒑𝒓𝒐𝒅𝒖𝒄𝒕, 𝒊𝒏 𝒕𝒉𝒆 𝒓𝒊𝒈𝒉𝒕 𝒑𝒍𝒂𝒄𝒆, 𝒂𝒕 𝒕𝒉𝒆 𝒓𝒊𝒈𝒉𝒕 𝒑𝒓𝒊𝒄𝒆, at the moment a shopper is ready to buy.

Delivering on it has never been harder.

I've been spending time mapping the Microsoft for Startups Pegasus portfolio against the challenges retailers and brands actually face, from consumer insights and demand forecasting to shelf intelligence, supplychain resilience, and the agentic platforms reshaping customer experience.

The result is a market map of 𝟓𝟓 𝒆𝒏𝒕𝒆𝒓𝒑𝒓𝒊𝒔𝒆-𝒓𝒆𝒂𝒅𝒚 𝒔𝒕𝒂𝒓𝒕𝒖𝒑𝒔 𝒂𝒄𝒓𝒐𝒔𝒔 𝒇𝒊𝒗𝒆 𝒔𝒐𝒍𝒖𝒕𝒊𝒐𝒏 𝒕𝒉𝒆𝒎𝒆𝒔, plus a walkthrough of the companies I find most compelling for this industry.

Two things worth noting:

- Most of these startups reach well beyond Retail and CPG. A retailer's forecasting engine is a manufacturer's demand planner. A CPG brand's content automation is a bank's compliance accelerator.

- This portfolio isn't static. We're constantly listening to where retailers and brands see capability gaps, then recruiting through our tier-one VC network to fill them with proven startups.

If your team is working through a specific challenge, reach out and we can organize a private startup showcase built around your priorities.

Satya Nadella recently argued that the real #AI advantage isn't which model you pick, but whether your organization builds a learning loop that gets better at your business every time it's used.

This piece applies that idea to #retail: how a learning loop paired with a world model can turn everyday signals (transactions, weather and local events, and in-store sensor data on traffic, dwell time, and shelves) into a compounding advantage competitors can't buy. It also looks at how Microsoft's stack and the #startups in the Microsoft for Startups ecosystem can help #retailers and #brands build it faster.

Spent this morning with the team at Brain Co., and I had to write it up.

They're tackling a problem every enterprise knows too well: AI pilots that look great in a demo and then never make it to production. What stood out to me is their forward deployed model. Instead of handing over software and wishing you luck, they embed engineers and product specialists right alongside your operators to get complex AI actually working in the real world.

Their Retail and CPG work is a great example, including a roofing distributor where they blended external signals like weather and competitor pricing into forecasting, pricing, and sales, and got real results in weeks rather than months.

Brain Co. is one of the hashtag#startups in our Microsoft for Startups Pegasus portfolio, and there are hundreds more we've curated based on what we're seeing across our customer ecosystem. Reach out if you'd like the list or want to organize a Startup Showcase with some of the founders.

Had a great conversation with Shameel Abdulla, CEO of Clootrack, this morning, and came away energized about what his team is building.

Most brands have gotten very good at measuring customer experience but not at driving outcomes from it. Clootrack closes that gap. It's an AI Super Agent for Voice of the Customer that doesn't just tell retailers and CPG brands what happened, but why it happened and what to do next, connecting customer feedback with transactions, churn, returns, and operational data to actually move the needle.

I'm also excited about their new MCP server, which brings this customer intelligence directly into Microsoft Copilot and other AI work fronts, so teams can ask questions in plain language and get governed, evidence-backed answers.

Clootrack is a great example of the kind of Agentic AI company we love supporting through the Microsoft for Startups Pegasus Program.

Most frontierAI research isn't built with retail in mind, but a handful of models in Microsoft Foundry Labs map onto real retail problems remarkably well. I went through the catalog and picked out the ones with a genuine use case, from product imagery and conversational commerce to search, optimization, and 3D shopping. I also dig into why in-house retail teams can treat these as building blocks rather than off-the-shelf products, and how pairing them with startups in the Microsoft for Startups portfolio can accelerate the work. If you're a retailer or brand thinking about where to build, I'd love to hear from you.

I spent the last month building autonomous projects: a hashtag#retail operation that designs and sells t-shirts off trending phrases, a pipeline that turns photos into 3D-printed figurines, and a newsletter that translates retail news into insights and actions rather than just headlines.

I learned a lot along the way. That you no longer need coding skills to build real things, and that you can build them in a fraction of the time. That debugging can be handed to hashtag#AI, and watching it explain its reasoning is genuinely educational. That the hard part has quietly moved from writing code to saying clearly what you actually want.

But the biggest realization was this. All of this technology is democratizing the ability to create. The challenge no longer lives in the tools or the skill to code something. It lives in creativity, in coming up with ideas worth building that no one else has thought of. That is the new skill all of us need to develop, and it is what becomes our unique strength.

What if your news feed answered "𝒘𝒉𝒂𝒕 𝒅𝒐𝒆𝒔 𝒕𝒉𝒊𝒔 𝒎𝒆𝒂𝒏 𝒇𝒐𝒓 𝒎𝒆" instead of "what happened"? I wrote a job description for an #AI agent using #MicrosoftScout, that reads 30+ #retail news sources every week and delivers trends and takeaways rather than headlines. New post on how I built it, the exact instructions that power it, and a link to the first live issue.

Earlier today I had a great conversation with Georges F. Mirza, creator of the ARS² Framework and founder of ComTask, and it prompted me to write down something I keep seeing with the #retailers and #brands I work with.

Around 70,000 B2B Tech #startups launch every year and most fail within two years. Yet the capabilities retailers need most right now, especially in agentic AI, live inside those young companies. The result is a familiar cycle: a promising demo, a quick assessment, a pilot that stalls, and the search starts over.

In this article I lay out two parallel paths out of that cycle. The first is the curated route through Microsoft for Startups and the Pegasus program, where solution categories are defined through direct customer engagement, sourced from tier 1 #VC portfolios, and vetted for enterprise readiness. The second is for the startups you find yourself, at trade shows or through your own scouting, where two frameworks let you do the vetting independently: ARS² to confirm the solution delivers accurately, repeatedly, at scale, and with speed, and the Azure Well-Architected Framework to confirm the engineering underneath can carry it.

Thanks to Georges for the conversation that sparked this. The full article is linked below, and I'd welcome perspectives from others who evaluate startups for enterprise deployment. What does your process look like?