The data guy will resist. Here's how to bring him along

The data guy will resist your AI rollout.

Good news: the resistance means he's paying attention

That's the most rational person in your building doing exactly the job you hired him for. The question isn't how to overcome him – it's how to bring him along instead of going around him. Because I've watched both versions, and only one of them ends with AI your company actually trusts.

He can block it – and going around him is worse

Let's be honest about the power map. If you run AI adoption – or you're the CEO driving it personally – the org chart says you outrank your data person. Reality says otherwise.

In a 200-person company, the one person who owns the data can veto your rollout without ever saying no. Access requests that take three weeks. Security reviews that surface new concerns each round. A pilot that stays "almost ready" through two quarters. And the kill shot: being right, once, about one wrong number the new tool produced. After that, every skeptic in the company has a flag to rally around, and your AI program becomes the thing people mention with a smirk in planning meetings.

None of this makes him a saboteur. 

Friction is how a professional without formal veto power exercises professional judgment. If the org gave him no way to say "this isn't safe yet" that anyone listens to, slow-walking IS the safety process.

So the tempting move is to go around him – buy the tool, connect it quietly, apologize later. That's worse. Now you have shadow AI answering business questions with no adult supervision over the logic, and when the first hallucinated figure lands in a deck – it will – there's nobody whose job was to catch it. The program doesn't just die; it dies discredited, and the data guy gets to say he told you so. Because he did.

The only winning path runs through him. Here's the map.

Why he resists: control, blame, security

Before you can answer his objections, respect where they come from – because all three are professionally sound.

Control. He's spent years making your numbers mean something – chasing definitions, fixing joins, cleaning the quality issues nobody else sees. An AI that redefines metrics at question-time threatens to undo that work at scale, anonymously.

Blame. When the AI is wrong about the data, whose name is attached? His. There's a whole thread of practitioners watching AI hallucinate on basic data retrieval – his fear isn't Luddism, it's evidence. He's seen the confident nonsense firsthand, probably this week.

Security. "We're putting company data into a chat tool?" is not paranoia. It's the exact question your auditor would ask, arriving early.

Notice what these three have in common: they're the objections of someone who takes the numbers seriously. You don't want to defeat that instinct. You want to arm it.

The three objections, answered honestly

Here's the kit. Not talking points – mechanisms. Each answer has to survive him actually checking, because he will.

‍

The pattern across all three rows: every answer gives him more explicit control than he has today, when his control lives in tribal knowledge and hope. That's the pitch, and it has the advantage of being true – this is governance with his name on the authorship line, enforced at the query level rather than promised in a policy doc.

Gatekeeper, not bottleneck

Now the part that turns him from reluctant approver into architect: what the role becomes.

Today he's a help desk. Every question in the company queues through his keyboard; his week is other people's tickets; his craft is spent on fetching. After: he defines the truth once – the governed data marts, the joins, the metric logic – and the answering happens without him, at whatever volume the company needs, always running on his logic. He stops being the person who pulls numbers and becomes the person whose judgment is embedded in every number anyone pulls.

There's a visibility upgrade too, and it matters more than it sounds. Today, leadership sees his output – reports appear, dashboards exist – but not his judgment; the definitions, the catches, the quality saves are all invisible until something breaks. In the governed setup his judgment IS the visible artifact: the published surface, the authored joins, the query log. It's the difference between being the plumbing and being the architecture.

Concretely, his week changes shape: the "quick questions" evaporate, and what's left is the work that actually compounds – modeling, quality, turning one-off requests into reusable data marts, reviewing the AI's query log. The industry line for this is blunt: analysts aren't replaced by AI, they're replaced by analysts who use AI. Your data person, running a governed surface an entire org self-serves from, is the second kind – with the audit trail to prove it.

The analyst who tried to break it

The rollout I keep coming back to – the US ecommerce brand whose CEO famously got ten years of answers in five minutes – didn't actually stick because of the CEO's enthusiasm. Enthusiasm doesn't survive the first wrong number.

It stuck because of what the analyst did next. He went through the governed joins line by line. He asked the AI questions designed to break it – edge cases, ambiguous phrasings, metrics he knew had traps in them. He watched what it did when it didn't have the data. And then he signed off. His skepticism wasn't the obstacle to the rollout – it was the certification of it. The CEO trusted the numbers precisely because the person whose job is distrust had gone first and come back satisfied.

That's what "bringing him along" buys you that going around him never can: a validator whose sign-off means something. When someone later questions a number – someone always does – the answer isn't "the AI said so." It's "our analyst certified the logic, and here's the audit trail." One of those survives scrutiny.

The replacement fear, versus the data

Under the three spoken objections sits the unspoken one, so let's put numbers on it.

McKinsey finds 78% of companies use AI to augment their analytics teams, not replace them, and the U.S. Bureau of Labor Statistics projects 34% growth in data science openings through 2034. What AI is actually absorbing is the mechanical slice of the week – exports, repetitive report assembly, the CSV grind he already resents.

The honest caveat, because he'll spot a sales pitch instantly: roles that were only report generation are genuinely contracting. Which is precisely the argument for the gatekeeper move now – it relocates his value from the shrinking part of the job to the compounding part while the choice is still his to make.

"This replaces me"

If he says it out loud, take it seriously – and answer with career logic, not comfort.

The governed setup makes him more load-bearing, not less: every answer the company acts on runs on logic he authored; every metric definition is his call; and for the first time, his oversight is visible – leadership can open Run History and see the control he's exercising, instead of assuming reports appear by magic. The work that stays human is the work that was always the point: validating data integrity, defining what metrics mean, investigating the variances, translating ambiguity into questions data can answer. The exports were never his value. Now the exports are gone and the value is legible.

And here's the champion's move that beats any pitch deck: bring a gift before you bring a proposal. 

We built model.owox.com – a free visual data-modeling canvas, open format, no signup wall – because the fastest way to show a data person respect is to invest in their craft before asking for their sign-off. Send it with one line: "saw this, thought of you." Let him form his own opinion about the product behind it afterward, in the role it actually offers him.

The one-conversation playbook

When you sit down with him, the script is four moves, and none of them is a demo.

Name his queue as your problem: "the line in front of you is the most expensive queue in my company, and it's not your fault." Hand him the canvas – the gift, not the pitch. Give him explicit veto power over the governed surface: nothing gets published that he didn't author. And propose the smallest possible start: one data mart, one connected chat, his edge-case testing as the acceptance gate.

You're not asking him to trust AI. You're asking him to govern it – which is what he's been trying to do to your data chaos all along. Start free when he's ready. He'll know before you do whether it's real.