MBA Training Data

Is Reddit's move to make most of the site login-only a panic response to Google AI Mode cutting search referrals, or a pricing move? The episode argues the wall matters less for casual readers than for AI companies, since it closes the free logged-out route scrapers use while Reddit renegotiates its data licensing deal with Google. The host pushes back that lurkers are the top of the funnel.

You come away able to separate a vanity metric like daily active uniques from revenue, see how one archive supports two money streams (targeted ads to logged-in users and licensing to model builders, including OpenAI and Google), and weigh the flywheel risk of a stale archive and what it means for brands' SEO.

Key takeaways

• Before judging a traffic loss, check which users actually generate revenue; Reddit says over four fifths of monetization comes from logged-in users.
• Treat open, logged-out access as a free substitute that weakens your price in any data licensing negotiation.
• Expect headline metrics like daily active uniques to fall when you deliberately shrink a low-value audience, and explain that trade to investors in advance.
• Protect the data flywheel: if casual readers never sign up and post, the archive you license goes stale.
• If your SEO plan relies on Reddit threads appearing in AI answers, recognize that visibility now depends on licensing deals you are not part of.

Chapters
0:00 Google AI Mode cuts Reddit search traffic
1:40 Who the Reddit login wall targets
2:27 Reddit's data licensing talks with Google
4:07 Daily active uniques versus engaged users
6:13 Ads, licensing and the data flywheel
6:54 What the wall means for SEO

Go deeper, free lessons

• Data monetization: three modes & the data flywheel

Sources

• Reddit puts logged in experience first as Google AI hits traffic
• Reddit shares sink on weak sales guidance and plans login requirement test as Google search traffic shrinks
• Earnings call transcript: Reddit beats Q3 2026 estimates but stock sinks By Investing.com
• Reddit to Limit Access for Logged-Out Users in Strategy Shift
• Reddit will make users log in as Google’s AI search cuts its traffic
• Reddit puts logged in experience first as Google AI hits traffic

Full transcript: https://www.mba-training.com/en/podcast/data/reddit-login-wall-data-leverage

The article: https://www.mba-training.com/en/blog/reddit-login-wall-data-leverage

MBA Training, mba-training.com

How did one compromised login give an intruder nearly four months inside the SFTP server where California swapped student test data with Educational Testing Service (ETS)? The files held names, test results, disability status, Section 504 status and records on students who left as far back as 2006. The hosts split on blame: ETS for the dwell time, or the state for pooling every child's record at one vendor behind one login.

You walk away able to audit your own vendor file drops: name an owner, check whether MFA is on, enforce data minimisation with a scheduled deletion job, and treat contract clauses as no substitute for controls that block a bulk download at midnight.

Key takeaways

• Give every SFTP file drop that feeds a vendor a named owner, since folders often belong to someone who left long ago.
• Enforce data minimisation with a scheduled deletion job, not only a written policy.
• Do not pool every record at one vendor behind one login when the job does not need it.
• Treat a contract clause as a promise on paper and require controls that detect unusual bulk downloads.
• Check whether multifactor authentication is on for every file transfer account.

Chapters
0:00 ETS breach via stolen SFTP login
1:29 Disability and Section 504 records exposed
2:33 Records since 2006 and data minimisation
4:03 Blame debate: ETS or state design
6:12 Notification delays, lawsuits and fallout
7:26 Unowned file drops and missing MFA

Sources

• Hackers Steal Personal Data of Millions of California Students in Massive Breach
• Millions of California student records breached, some dating back to 2006
• ETS Sued Over Data Breach Affecting Current, Former Calif. Students
• One Compromised Login Exposed Millions of California Students' Records - SecurityWeek
• Cotchett, Pitre & McCarthy Files Class Action Lawsuit Against Educational Testing Service for Unlawfully Exposing Private Information of California Students

Full transcript: https://www.mba-training.com/en/podcast/data/ets-breach-student-data-retention

The article: https://www.mba-training.com/en/blog/ets-breach-student-data-retention

MBA Training, mba-training.com

Who pays when a fund's NAV is wrong and the manager never saw the error? Under CSSF Circular 24/856 the investment fund manager corrects and compensates, even when the administrator calculated the price. The episode argues that managers should own one position and pricing record, backed by a written pricing policy, rather than inheriting whatever the administrator used.

Listeners learn to back-solve reconciliation tolerances from the 0.5% NAV materiality threshold for fixed income UCITS funds: a tenth of it per break for same-day ownership, a quarter of it in open breaks as a NAV gate. Also covered: the aggregation clause, EU T+1, Invesco's identifier consolidation, circular matching, Bloomberg pricing drift and the SEC's Madison Capital Funding settlement.

Key takeaways

• Set reconciliation tolerances from the fund's regulatory NAV error threshold instead of copying vendor default settings.
• Give any single break worth more than a tenth of the NAV threshold a named owner the same day.
• Do not release the NAV while open breaks add up to more than a quarter of the threshold, and log, own and approve any override.
• Check quantities against the custodian and prices against an independent challenger vendor, because IBOR matching the administrator can be two copies of the same mistake.
• Report breaks to the board as money against threshold, age of the oldest break and its owner, not as a raw count.

Chapters
0:00 Who pays for NAV errors under CSSF 24/856
0:59 Custodian, administrator and IBOR disagree on positions
3:23 Back-solving tolerances from the NAV threshold
5:56 NAV gate versus T+1 cut-offs
7:04 Identifiers, circular matching and price challenge
9:40 Reporting breaks to the board

Go deeper, free lessons

• Benchmarking golden-source pricing and valuation confidence
• Measuring data quality with completeness and accuracy metrics
• Running practical data checks and controls
• Mapping the market and portfolio data stack
• Master Data Management in practice: styles, tools, and the Golden Record

The article: https://www.mba-training.com/en/blog/custodian-reconciliation-nav-tolerance-budget

MBA Training, mba-training.com

How did hackers send a ransom note through the real Asos app? Reporting points to a conned employee and, per researcher Kevin Beaumont, Simon AI, a composable customer data platform owned by Monetate that reads straight from Snowflake with MFA optional and off by default. That route is unconfirmed. The episode argues the real failure is ownership: marketing bought it, data engineering wired it to the warehouse, and nobody owned the login.

You leave with a concrete audit: list every third-party tool holding warehouse credentials, check whether MFA is enforced or only available, check whether it can message customers, and fix the risky ones with mandatory MFA or single sign-on before Friday.

Key takeaways

• Ask your data engineering lead for a list of every third-party tool that holds credentials into your warehouse.
• For each tool, check whether MFA is enforced or just available, and whether it can also send messages to customers.
• Put any tool that answers optional MFA and can message customers at the top of the list and fix it before Friday.
• Turn on mandatory MFA or put the tool behind single sign-on through your company's identity system.
• Treat MFA as the floor, not the ceiling, since it stops replayed infostealer passwords but is weaker against a live con.

Chapters
0:00 Asos app push notification ransom note
1:55 Social engineering and the Simon AI CDP
3:31 Optional MFA on Snowflake-connected tools
6:17 Why composable CDPs are a data leader problem
7:31 Breach scale, extortion deadline, open questions
8:28 Auditing third-party warehouse credentials this week

Sources

• Asos Hackers Claim Breach of Snowflake-Connected Simon AI
• Asos Shares Plunge Following Public Data Breach Threat
• ASOS breach update: Hackers stole customer details and shopping searches
• Asos confirms breach of customer data after hackers send rogue app notification
• ASOS Breach Reveals the Risks in Customer-Facing SaaS

Full transcript: https://www.mba-training.com/en/podcast/data/asos-breach-simon-ai-mfa

The article: https://www.mba-training.com/en/blog/asos-breach-simon-ai-mfa

MBA Training, mba-training.com

Agents dominated dbt Summit, but does smart orchestration or a context layer make data agent-ready on its own? The episode argues no. Fanatics Betting and Gaming saw about 0.2% model reuse from dbt State until it declared source freshness and model-level SLAs, which raised reuse to roughly 15%. The same pattern runs through Fivetran's Context Layer, AI audit trails and token costs.

Listeners learn how dbt State is billed per daily active target table, including tests, why the Context Layer's beta status needs checking in writing, how git history serves as an audit trail, and how dbt models served through the MCP server cut Gong transcript tokens by 20x or more.

Key takeaways

• Declare source freshness and model-level SLAs before expecting dbt State to reuse models and cut warehouse costs.
• Get the release status and pricing basis of Fivetran's Context Layer in writing before you evaluate it.
• Write metric definitions with semantic name, owner, source models, grain, filters, time basis, unit rules, freshness and exclusions.
• Keep transformation logic in git so you can show a diff when the CFO asks why a number was revised.
• Pick the metric your AI pilot is asked about most and give it a named owner, written definition, freshness tolerance and one test before Friday.

Chapters
0:00 Fanatics and the 0.2% dbt State reuse rate
2:21 dbt State pricing and per-test charges
3:53 Fivetran Context Layer: early access risks
6:26 Auditing AI agents without a rulebook
7:49 Why token costs belong to data teams
9:17 Declare meaning before agents act

Go deeper, free lessons

• dbt (data build tool): industrialized SQL transformation
• The metrics & semantic layer
• Active metadata and the data catalog
• Data FinOps: controlling cloud data cost
• Shift-left data quality: embedding governance in the engineering pipeline

The article: https://www.mba-training.com/en/blog/agent-ready-data-dbt-summit

MBA Training, mba-training.com

How did an attacker using ARTEX, an open-source agentic AI penetration-testing tool, take data on more than 67,000 customers from seven South Korean firms, including Shinhan, KB Kookmin and Hana, without touching core banking? The episode argues the failure was data access design. Broker portals and employee apps returned income, loan limits and resident registration numbers to callers who did not need them, despite nearly 124 billion won in security spending.

You come away knowing what the Financial Services Commission ordered, how CrowdStrike linked the attack to Claude Code and DeepSeek, and why basic cyber hygiene matters more than an AI defense product. You also get a concrete plan to inventory every externally reachable system that returns customer data and remove unneeded fields.

Key takeaways

• List every system reachable from outside your network that returns customer data, including partner portals and employee mobile apps.
• For each system, record the fields it returns and who outside the company can call it.
• Remove any field the caller does not need, starting with income, credit limits and national ID numbers.
• Make sure the data office is in the room when partner and broker APIs are specified, so a loan status is returned instead of a full customer profile.
• Have your red team run an agentic penetration-testing tool against partner portals before an outsider does.

Chapters
0:00 Shinhan breach through a loan broker portal
1:13 ARTEX agentic AI tool and the human attacker
2:38 Side-system data access as a CDO problem
4:42 FSC emergency order on exposed systems
6:06 Hygiene failure or AI-driven attack speed
8:16 Weekly task: audit customer data endpoints

Sources

• CrowdStrike: How a Cyber Attacker hit South Korean Banks
• AI-powered attacks on banks expose technological lag in Korea's financial cyber defenses - The Korea Times
• S. Korean regulators call emergency meeting after financial data breaches - UPI.com
• Financial authorities hold emergency meeting after data breaches at Shinhan Bank, KB Kookmin Bank - The Herald Business
• AI-linked hacks hit Korean banks through loan-agent sites

Full transcript: https://www.mba-training.com/en/podcast/data/artex-korea-bank-breaches-data-access

The article: https://www.mba-training.com/en/blog/artex-korea-bank-breaches-data-access

MBA Training, mba-training.com

What happens when a data platform is legally defensible, operationally embedded and publicly distrusted at the same time? More than 44,000 people have filed Article 21 objections under UK GDPR against NHS England's Palantir-powered Federated Data Platform, live in 142 trusts, weeks before a February 2027 contract break clause comes into reach. The position here: objection handling is a throughput problem, not a policy document.

You come away able to pressure-test your own platform. Check the published benefits claims, 117,000 extra operations, a 14.3% cut in discharge delays, against Foxglove FOI data showing 13 of 41 trusts doing fewer operations, and name the owner and service level for mass objections before someone else does.

Key takeaways

• Write down the exact workflow, named owner and turnaround time for handling ten thousand simultaneous objections to your most visible data platform.
• Check your last published benefits number: if the methodology is before and after rather than controlled comparison, add the causality caveat yourself.
• Treat your supplier's public reputation as part of your lawful basis, because objections scale when a vendor becomes a political symbol.
• Expect published benefits claims to be audited by people with FOI rights and a motive, so check whether an aggregate hides a bimodal distribution.
• Remember that when people cannot opt out of the system, they opt out of the service, degrading data no access control can fix.

Chapters
0:00 44,000 Article 21 objections to NHS data
1:40 What the Federated Data Platform runs
2:31 Palantir politics versus legal mechanism
3:01 Why patients cannot opt out of FDP
4:09 Testing Palantir's NHS benefit claims
7:28 What transfers to any CDO

Sources

• Thousands demand NHS halt use of patient data on Palantir platform
• 44,000 object to Palantir’s NHS data platform, the Guardian reports
• MPs tell NHS to start packing Palantir's bags ahead of 2027 contract break
• Pressure mounts from MPs on Palantir’s role in the NHS
• NHS Palantir claims face scrutiny after data suggests uneven results
• Patients Could Opt Out of Care Over NHS Data Platform Fears, Councillors Warned - This is the Coast
• 44,000 object to Palantir’s NHS data platform, the Guardian reports
• NHS Patients Can't Opt Out of Palantir. Their Hospitals Can.

Full transcript: https://www.mba-training.com/en/podcast/data/palantir-nhs-data-objections

The article: https://www.mba-training.com/en/blog/palantir-nhs-data-objections

MBA Training, mba-training.com

dbt State went generally available in September 2026, so rebuilding every model on a cron schedule is now a choice rather than a default. The position here: the tool cannot tell you how stale each model is allowed to be, and the lag_tolerance config most teams set once and globally is a business decision about decision cadence, not an engineering setting.

You leave with a method: meter wasted rebuilds using BigQuery INFORMATION_SCHEMA.JOBS or Snowflake credit consumption, tie tolerance to the decision each model feeds, keep source freshness checks running, read the Explain tab in audit mode, and cap agent queries with resource monitors and maximumBytesBilled. Includes the RxBenefits 59% result, dbt's 15 to 30% range, FinOps Foundation and Flexera 2026 findings.

Key takeaways

• Calculate last month's share of scheduled runs that rebuilt models whose upstream sources had not changed, and treat that number as your addressable budget.
• Set lag_tolerance per model to the cadence of the decision it feeds, never shorter, and flag models where nobody can name the decision.
• Keep source freshness checks on their own schedule, because reused models skip their tests and silent semantic drift passes volume checks.
• Run dbt State on one non-critical job with conservative tolerances and read the Explain tab for a week before trusting it widely.
• Set Snowflake resource monitors with suspend triggers and BigQuery maximumBytesBilled so agent-generated queries cannot run up an overnight surprise.

Chapters
0:00 Why 60% of scheduled model builds repeat
1:13 What dbt State GA actually changes
2:57 Metering wasted rebuilds before flipping the switch
4:18 Silent semantic drift when models are reused
5:41 Committed spend contracts hide the savings
6:57 Capping autonomous agents in the warehouse

Go deeper, free lessons

• Data FinOps: controlling cloud data cost
• dbt (data build tool): industrialized SQL transformation
• Performance & scalability: partitioning, clustering & cost optimization
• Cloud data infrastructure: services, costs & migration
• Chargeback & showback models

The article: https://www.mba-training.com/en/blog/cloud-data-costs-dbt-state

MBA Training, mba-training.com

California's No Robo Bosses Act bars employers from disciplining or firing a worker on an automated system's output alone, and from 1 July 2027 a human must independently corroborate that output in writing. The position here is that this reads as an HR matter but lands on data teams, because the statute covers statistical modeling and data analytics, which means attrition models and attendance scores, not only frontier AI.

You come away knowing what SB 947 and SB 951 demand, how Cal-WARN notices must now name AI-driven displacement, what the Chamber of Commerce and Senator McNerney argued, why the SB 7 inventory clause was dropped, and how to audit every system scoring a named employee.

Key takeaways

• Inventory every system that outputs a score, ranking or flag about a named employee, including attendance, quality and attrition models.
• For each system, record in writing whether you retain the input features and output for that individual on that date, and who the human reviewer was.
• Stop scoring pipelines overwriting prior outputs, since a decision may be questioned years later.
• Treat internal automation ROI decks and headcount-avoided trackers as future legal exhibits under SB 951's Cal-WARN disclosures.
• Check models for proxy inference of protected characteristics such as postcode or shift pattern, which SB 947 prohibits.

Chapters
0:00 Why SB 947 is a data team problem
0:53 What the No Robo Bosses Act requires
1:52 Proving model lineage for firing decisions
4:17 SB 951 and Cal-WARN AI layoff notices
5:46 Chamber opposition and the vetoed SB 7
7:40 Enforcement odds and your logging checklist

Sources

• No Robo Bosses Act: California's Essential AI Firing Warning
• Gov. Newsom wraps California term by enacting 11 more laws on AI safety — Transparency Coalition. Legislation for Transparency in AI Now.
• California Passes No Robo Bosses Act
• California Governor Signs 3 Bills Targeting AI and Workplace Surveillance - Ogletree
• New California Law Requires That Humans Decide Firings, Not AI
• California Legislature Advances AI Employment Bills
• California Expands Cal-WARN Notice Requirements for AI-Related Workforce Reductions
• California Employers Will Soon Need to Say When AI Causes a Mass Layoff: 5 Steps to Prepare

Full transcript: https://www.mba-training.com/en/podcast/data/sb-947-ai-firing-human-review

The article: https://www.mba-training.com/en/blog/sb-947-ai-firing-human-review

MBA Training, mba-training.com

Why do churn models produce accurate predictions that never move retention numbers? The position here is that most telcos predict churn the week it happens, when the customer has already price-shopped and gone. Telefonica's work came from years of accumulating unglamorous signals such as degraded data speed at a single cell tower, payment timing drift and the gap between plan and actual usage, combined with contract stage and competitor pricing by postcode.

You walk away able to separate accuracy from uplift, size the movable middle, defend a holdout control group to a revenue VP, and wire a score into a pre-approved retention offer the agent can act on. References include MIT Sloan Management Review on unactionable analytics and dbt Labs on data cleaning time.

Key takeaways

• Stop scoring churn in the week it happens and build features from months of early signals like payment timing drift and network quality.
• Treat network quality as a retention input, not only an engineering metric, and get those two teams talking.
• Measure uplift against a holdout control group instead of accuracy, so you spend only on the movable middle.
• Wire the churn score into a pre-approved offer the frontline agent can give with no approval chain.
• Audit your last retention campaign for a control group; without one you have a discount program, not a churn program.

Chapters
0:00 Catching churn twelve hours too late
1:00 Boring signals that predict churn early
1:54 Why years went into data plumbing
2:31 Accurate scores nobody can act on
3:10 Uplift and the movable middle
4:04 Monday check: did you run a control group

Go deeper, free lessons

• Decoding the telecom data goldmine: CDRs, network telemetry, and usage signals
• Advanced analytics: CLV, churn prediction & demand forecasting
• Models in production: drift, monitoring & MLOps
• Governing rich telecom data under GDPR, ePrivacy, and lawful intercept
• Benchmarking network and service KPIs that leadership tracks

The article: https://www.mba-training.com/en/blog/telefonica-churn-prediction-subscriber-behavior

MBA Training, mba-training.com