Explorez tous les épisodes du podcast TheModernCDO

For fifty years, the rule was simple: garbage in, garbage out. Bad data produced output that obviously looked bad, so someone caught it and fixed it. AI broke that bargain. Feed a copilot flawed data and it returns a fluent, confident, well-structured answer that happens to be wrong, and nothing on the screen looks broken.

In this episode, Richard Muirhead makes the case that data quality is the one problem AI both makes more dangerous and is genuinely built to solve. He covers why the failure mode went invisible, how AI-assisted profiling, entity matching, observability, and cataloging actually find bad data, where the return on investment shows up, and the three traps that separate a fix that holds from a demo that doesn't.

In this episode:

- Why "garbage in, confidently out" is the new failure mode

- The numbers behind AI-readiness (Precisely 2025, Gartner)

- Four things AI-assisted data quality does that hand-written rules never could

- How to build the ROI case your finance team will actually fund

- Three traps: tooling without ownership, overconfidence, and treating it as a one-time project

Read the full whitepaper: https://themoderncdo.gumroad.com/l/s3e1-Garbage-In-Confidently-Out

Work with Richard: https://perigee.pro

The Modern CDO is written and hosted by Richard Muirhead, fractional Chief Data Officer and founder of Perigee.

Most AI governance ends at the moment the real risk begins. The day a model goes live is the day it starts to drift.

A model learns a snapshot of the world, and the moment it ships that snapshot starts to age. Accuracy and fairness degrade silently, with no alarm, because the governance was all spent at the gate and none on the operation. Watch the model for as long as it runs.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e29-no-post-deployment-monitoring

Generative AI is most dangerous at the exact moment it is most convincing.

It fails fluently. It produces a confident, polished, authoritative falsehood with no signal that it is wrong, leaks whatever an employee feeds it, and varies its output from one run to the next. Govern the real failure modes, not the ones traditional software taught you to expect.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e28-the-generative-ai-traps

A model can be 95 percent accurate and systematically unfair to the people who matter most. The aggregate number hides it.

Treated as a late-stage checklist item, bias gets discovered too late to fix, rationalized away under deployment pressure, or missed entirely. A single accuracy number is not a fairness measure. Measure by subgroup, from the start, where you can still act on what you find.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e27-bias-as-an-afterthought

The moment a real decision can be blamed on a model, you have built a machine for laundering accountability.

A model makes a consequential decision and no human owns it. The data scientist points to the business, the business points to the model, the sponsor treats the output as objective. Everyone is adjacent and nobody is accountable, which means nobody is watching when the decisions drift wrong.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e26-the-accountability-vacuum

When a model starts making bad decisions and you cannot trace what it learned from, you do not have a problem. You have an outage with no off switch.

A model runs beautifully for a year, then starts producing bad outcomes nobody can diagnose, because its lineage was never captured. An untraceable model is an unfixable model. Make lineage a deployment gate, the way tests gate code, and capture it at build time.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e25-model-risk-without-lineage

Standing up a separate AI governance function is how you rebuild data governance from scratch and get it wrong twice.

Most AI risk is data risk with higher stakes: bias from training data, privacy from data use, unexplainability from missing lineage. Wall off a separate tower and you rebuild lineage, quality, and ownership badly, then watch the two versions diverge and open a seam where the worst risks hide.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e24-ai-governance-in-a-separate-tower

Most companies wrote their AI policy before they had any idea how much AI they were running.

Writing the policy is the easy part. The AI it governs is mostly invisible: shadow models, AI features inside SaaS tools, employees pasting company data into public generative tools. You cannot govern what you cannot see. Discovery comes first, and it never stops.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e23-policy-without-an-inventory

AI governance that only knows how to say no does not make AI safer. It drives the risky work into the shadows.

A brake is not a steering wheel. A function that can only stop makes itself an obstacle, and obstacles get routed around. The risky AI work continues with no oversight at all, while the function congratulates itself on what it blocked. Build the steering wheel instead.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e22-governance-as-a-brake

MDM is never finished. It is operated. And the operating budget is the line item every program forgets.

Every MDM business case has a detailed build budget and a stewardship line that is tiny or missing. Go-live is not the end. It is the start of the longest and most important phase. The exception queue that needs human judgment does not shrink after launch. It grows.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e21-underestimating-stewardship

Master data management promises a single version of the truth. But a golden record built from wrong source data is just a single version of a wrong truth, and now it is official.

In this contrarian standalone, Richard argues that survivorship rules pick a value, they do not verify one, so a golden record can launder bad data into a clean, trusted, authoritative error. The fix is not in the master data hub. It is upstream, at the source. And he does the math on what it costs to fix the source against what it costs to keep perpetuating the problem.

What you'll learn:

- Why a golden record promotes bad data instead of fixing it

- How survivorship picks a value without ever verifying it

- Why the real fix is upstream: quality at capture, provenance, and source accountability

- The 1-10-100 rule: about a dollar to prevent, ten to correct, a hundred to let it fail

- Where AI actually fits: it can flag the error, it cannot make bad data true

Read the whitepaper: https://themoderncdo.gumroad.com/l/a-single-version-of-the-wrong-truth

Every MDM program assumes everyone already agrees what a customer is. That assumption is where it dies.

Is one customer a person, a household, a legal entity, an account. The definition lives entirely in the edge cases the kickoff skips, and the buried disagreement erupts deep in the build, when the matching rules force a decision. Decide what one customer is before you try to master it.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e20-no-definition-of-master

MDM does not fix bad data. It gives bad data a gold star and removes the warning label.

MDM matches and merges. Neither operation creates correct data. Feed it bad inputs and it selects one of them, stamps it golden, and broadcasts it everywhere with more authority than it ever had. Quality comes before matching, or the master just promotes your mess.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e19-ignoring-upstream-quality

Writing code was the scarce skill in software for fifty years, and vibe coding just made it nearly free. Andrej Karpathy named it in early 2025; by November it was Collins Dictionary's word of the year. Google and Microsoft now say a quarter to a third of their new code is AI-generated. So here's the uncomfortable question for every engineering leader: your developers can produce more code than ever, but how well do they actually understand the vertical they're writing it for?

In this standalone episode, Richard Muirhead argues that the bottleneck in software was never the typing, it was understanding the problem, and AI automated the translation while leaving the understanding exactly where it was. The result is teams shipping flawless, fluent code that solves the wrong problem. He covers why domain judgment is now the scarce skill, where the real return is, and the traps that turn fast code into confident mistakes.

In this episode:

- Why "vibe coding" bakes the trap into its own definition

- The wrong-problem failure mode: code that compiles, demos, ships, and is still wrong

- Why two-thirds of developers say their top AI frustration is "almost right, but not quite"

- The fix: value domain judgment, embed engineers in the business, point AI at understanding

- The ROI, the speed illusion, and three traps including why you can't vibe-code a domain

Read the full whitepaper: https://themoderncdo.gumroad.com/l/the-code-that-solved-the-wrong-problem

Work with Richard: https://perigee.pro

The Modern CDO is written and hosted by Richard Muirhead, fractional Chief Data Officer and founder of Perigee.

The most consequential decision in your MDM program is being made by whoever is configuring the tool.

Which source wins when two systems disagree decides whose version of reality is authoritative. Too often an implementer picks a sensible default under deadline pressure. That is a business judgment, attribute by attribute, and it belongs to the domain owner, validated against real records before go-live.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e18-survivorship-without-ownership

Without governance underneath it, your MDM hub is an expensive argument that runs on a server.

You reached the agreement, configured the hub, and the golden records were trusted at launch. Six months later they have drifted back into disorder, because mastering data is not a thing you finish. The survivorship rules are business judgments that need an owner forever, not just at setup.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e17-mdm-with-no-governance-underneath

Mastering customer, product, vendor, and asset at the same time is how you master none of them.

Each master data domain is its own deep, contentious negotiation, and contentious negotiations do not parallelize. Start them all at once and eighteen months later you have four half-mastered domains, which is to say four nobody trusts. Master one completely first.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e16-mastering-every-domain-at-once

Master data management is a business agreement enforced by software. Skip the agreement and the software just automates the fight.

The hard part of MDM was never the hub. It was getting the business to agree on what a customer is and which source wins. Automate before you agree and you do not get a single version of the truth. You get an automated version of your disagreement, running at machine speed.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e15-mdm-as-a-technology-project

Governance built only to pass an audit dies the moment the auditor leaves the building.

A program born from an audit gets optimized to survive audits. It passes, the regulator moves on, and it loses its only reason to exist. Use compliance as the trigger. Then build for the value, because the same inventory and lineage the regulator wants are what the business runs on.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e14-compliance-only-governance

If your data quality work has an end date, you are paying to clean data that will be dirty again by next year.

The cleanup crosses its finish line, everyone celebrates, and a year later the data is back where it started, because the sources that made the errors never changed. Quality is not a state you reach. It is an operation you run, with monitoring and a loop back to the source.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e13-data-quality-as-a-one-time-project

A steward with no time and no authority is a name in a spreadsheet, not a control.

You can name forty stewards in a kickoff. It feels like progress. It is a list. A steward needs protected time, real authority to decide, and recognition in how they are evaluated. Take away any one of the three and you have built a costume, not a control.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e12-stewards-without-authority

A data catalog is not a governance program. It is a filing cabinet for a program you never built.

A tool scales a working operating model. It cannot create one. Buy the catalog first, hoping it does the hard work for you, and you get an empty machine harvesting technical metadata into a stale, misleading glossary. Build the program. Then file it.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e11-tool-led-governance

IT cannot tell you what an active customer is. So IT cannot own your governance.

Ownership drifts to IT because the catalogs and the tooling live there. But whether someone counts as an active customer is a business judgment, not a technical one. The business owns the what. IT owns the how. Invert that and the catalog is complete and quietly ignored.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e10-it-owned-governance

Governing four domains at once is how you govern none of them.

Named stewards everywhere, untrained, with no time. Processes defined in parallel with no proven template. A year later, there is no single place you can point to and say governance works here. Prove the operating model in one high-value domain. Then scale the proof.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e9-the-big-bang-rollout

If people describe governance as something done to them, they are already building the shadow systems that will gut it.

Built as a tax, governance gives the data user nothing they can feel, just forms and approvals, so they route around it. Every shadow spreadsheet and unofficial extract is a vote against your process. Make the right thing the easy thing, or get evaded.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e8-governance-as-a-tax

A capability that is built and never used is not worth zero. It is worth less than zero.

Datasets cataloged, users provisioned, models deployed. Every number goes up and to the right, and none of them tells you whether any of it is used. Then a skeptical new leader asks the adoption question, and the whole green dashboard inverts in an afternoon.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e7-vanity-metrics

You cannot build an AI future on foundations you were too proud to audit.

Every ambitious roadmap rests on an assumption no one has checked: that the foundations are good enough. They rarely are. "We will clean the data as part of the project" is where the program goes to die, seventy percent in, with leadership watching the AI initiative stall on data it cannot trust.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e6-ignoring-data-debt

Most data strategies do not fail in design. They die in the first ownership fight nobody was empowered to settle.

You can write a brilliant strategy and still stall in month three, when two units both claim the customer domain and no one is allowed to decide. The missing layer is the operating model. Skipping it does not avoid the fight. It defers the fight to the worst possible moment, the middle of execution.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e5-no-operating-model

If no executive will put their name on a number that your data team produces, the program is already dying.

Data value is hard to attribute, so programs retreat to platitudes about being an asset. Budget committees do not fund platitudes. They fund numbers with a name attached. Tie every serious thing you build to a business owner who will defend it when you are not in the room.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e4-no-line-of-sight-to-value

If your data strategy can be summarized as the name of a platform, you bought infrastructure and called it a strategy.

Platform decisions feel like progress because they are concrete, and a vendor hands you a glossy reference architecture. But if your biggest competitor adopted the same platform tomorrow, what advantage would you have left. Outcome first, then the capability it requires, then the technology. Never the reverse.

Long-form whitepaper is at https://themoderncdo.gumroad.com/l/s2e3-technology-first-strategy

Fifteen workstreams running at once is not ambition. It is the most reliable way to ship nothing.

Boiling the ocean does not look like failure while it is happening. It looks like effort; a lot of smart people are busy on many fronts. Then funding fatigue arrives, and nothing is finished. Thirty percent of a capability is worth zero, not thirty percent of the value.

Long-form whitepaper is at https://themoderncdo.gumroad.com/s2e2-boiling-the-ocean

The most dangerous day for a data strategy is the day it gets approved.

Approval feels like the finish line. It is the moment the strategy starts to die. This episode makes the case that a data strategy you approved is worthless, and the only one that counts is the one that made you kill something. The test is simple and brutal: name one initiative your strategy caused you to decline. If you cannot, you have a wish list with a cover page.

Long form can be read at: https://themoderncdo.gumroad.com/s2e1-strategy-as-a-document

Board-level scrutiny of AI is no longer theoretical. Audit committees and risk committees at major enterprises are now asking pointed, specific questions about AI risk exposure, model accountability, and regulatory readiness — and getting vague answers from data and technology leaders who have not built the infrastructure to respond with precision. Richard Muirhead examines what board-ready AI accountability looks like in practice, and the four specific artefacts that separate CDOs who can answer the question from those who get found out by it.

In this episode:

· How two major US financial institutions received identical AI risk questions from their audit committees in 2024 — and why one CDO lost authority over AI governance as a direct consequence of being unable to answer them

· Why a leading US health insurer's Chief Data and Analytics Officer could not confirm, in the boardroom, whether prior authorisation AI systems had a documented human review pathway — and what the three-week investigation that followed revealed

· How a FTSE 100 retailer spent ninety days building a board AI accountability dashboard they should have had in place before a regulatory inquiry arrived

· The four board-ready deliverables every CDO needs: AI System Inventory with Risk Classification, Model Accountability Matrix, Regulatory Exposure Map, and Board AI Monitoring Cadence

The complete templates for all four deliverables are available for purchase at https://gumroad.com/products/lnffqo

Eighty percent of enterprise AI pilots never reach production. The failure mode is consistent across sectors, and it is almost never about the model. Richard Muirhead examines why the production gap is a data infrastructure and governance problem masquerading as an AI problem — and what the organisations closing that gap have done structurally to change the ratio.

In this episode:

· How three major European telecoms operators ran simultaneous AI pilots for network optimisation — with dramatically different production outcomes — and why the gap traced to data pipeline architecture, not model quality

· Why a top-five US bank's AI credit decisioning model failed production validation for regulatory data lineage reasons that were visible six months before the project started

· What a healthcare network's AI diagnostic pilot revealed about the gap between research-grade and production-grade data infrastructure — and the eighteen-month remediation that followed

· The production readiness checklist: six data infrastructure criteria that determine whether an AI pilot will reach production — and when to apply them in the development cycle

The full production readiness checklist is available in the paid tier at themoderncdo.substack.com.

The majority of enterprise data strategies are technically coherent and strategically invisible. They describe capabilities, platforms, and architectural ambitions that mean nothing to the executives who fund them. Richard Muirhead examines why data strategies fail at the board level — not because the thinking is wrong, but because the translation is — and lays out a three-layer framework that connects data investment to outcomes a CFO will sign off on.

In this episode:

·  Why two major US retail banks' multi-year data modernization programs were defunded mid-execution — and the board communication failure that caused it

·  How a leading European CPG manufacturer rebuilt its data strategy narrative around revenue attribution and cut its budget approval cycle from nine months to six weeks

·  What a healthcare system's data strategy pivot from infrastructure investment to clinical AI revenue enabled — and why the framing change mattered more than the technical change

·  The three-layer data strategy framework: business outcome mapping, capability sequencing, and the board narrative that makes both defensible under scrutiny

A policy document, an ethics committee, and a model registry that nobody updates is not AI governance — it's the appearance of governance. In this episode, Richard Muirhead dissects three documented enterprise AI governance failures across banking, telecoms, and retail, and defines the three non-negotiable components of a program that will hold under regulatory scrutiny and operational stress.

In this episode:

·  How Credit Suisse's risk governance apparatus failed to convert model alerts into executive decisions — and what that architecture failure cost

·  Why Vodafone's technically sophisticated AI deployment was grounded for eight months by an unanswered accountability question

·  How EU-jurisdiction retailers are facing consumer protection actions for AI outputs their governance teams didn't know were happening

·  The three operational governance components that separate real programs from theater: accountability mapping, monitoring with teeth, and regulatory readiness as a standing posture.

Most enterprise data governance programs are failing — not because of bad policy or weak tooling, but because of a single structural mistake: they were designed as compliance functions. Richard Muirhead makes the case that governance-as-compliance is an AI delivery bottleneck that no technology investment can solve, and lays out the organizational moves that fix it.

In this episode:

·  Why top retail organizations with policy-complete governance programs had 11-week data access cycles — and what that costs in AI delivery capacity

·  How Capital One built governance as a product function with SLAs, time-to-trust metrics, and a measurably different AI outcome

·  Why HIPAA has done structural damage to healthcare data governance — and what the health systems succeeding in AI did differently

·  Three organizational moves: change the reporting line, introduce time-to-trust as an operational metric, make governance the prerequisite for AI not the constraint on it.

Hosted by Richard Muirhead — Fractional CDO, AI strategist, and former IBM Distinguished Engineer with 30+ years of enterprise experience and 130+ patents — each episode delivers a single sharp thesis, three sector-specific evidence cases, a practical decision framework, and one falsifiable prediction.

No guest roundtables. No vendor content. No hedging. Episodes run 12–18 minutes. Published weekly.

Coverage spans three domains that define the modern data leader's mandate:

→  Data Strategy — how to build, defend, and evolve a data capability that outlasts the hype cycle.

→  Data Governance — the structural decisions that separate organizations that scale AI from those that stall at the pilot stage.

→  AI Governance — the emerging discipline that will determine which organizations earn trust in the AI era and which get regulated into compliance.

Sectors covered: Financial Services · Healthcare & Pharma · Retail & CPG · Telecoms.

If you're a CDO, Chief AI Officer, CIO, or senior data leader navigating what AI actually means for your organization — this is built for you.