What to do when your AI hire and your data team disagree
7 minute read. Updated 2026-08-08.
When a fractional AI hire and an in-house data team disagree, don't referee it by expertise, referee it by accountability. Give the specialist the final say on whether the method fits the business goal, and give the data team the final say on production and everything they'll own after the engagement ends. Most of these disagreements exist because nobody wrote decision rights down before the work started.
The argument in the room
You bring in a freelance or fractional AI specialist to move a stalled project forward. Within a few weeks they recommend a change: a different way of matching customer records, a model in place of a rules-based process, a shift in what data feeds a decision. Your in-house data team pushes back hard, and their objection sounds just as credible as the recommendation. You are now the person in the middle, refereeing a disagreement between two people who each have a legitimate claim to being right, on a subject you brought the specialist in precisely because you could not judge it yourself.
The instinct is to treat this as a competence question: who actually knows more about the data, the model, the risk. That instinct is the wrong starting point. In most cases the disagreement is not about competence at all. It is about authority, specifically about who was ever given the right to make this particular call, and in most organisations nobody wrote that down before the argument happened.
What the disagreement is actually about
Neither side is usually wrong on the point they are each defending. The specialist knows the method and where it has worked elsewhere. The data team knows the mess in the underlying data that never made it into any specification document. The clash is not knowledge against ignorance. It is two accurate, incomplete pictures of the same system, produced by two people looking at it from different angles.
Part of the reason this keeps happening is that the work gets bought under a job title that does not map to any real authority. Nothing at all is published for titles like AI governance or AI risk and compliance in the advertised UK market, which means people doing that kind of work are routinely hired under adjacent titles such as AI lead or data scientist, with the actual decision rights left undefined. If the title does not describe the job, the org chart will not describe it either.
The data team is often hearing about the decision after it has already been made, not before. Their objection in the meeting is frequently the first chance they have had to register that they were never consulted on scope. That reads as territorial. It usually is not.
Who should win
Give the specialist the final call on whether the method fits the stated business goal. That is what you are paying for. Give the data team the final call on anything to do with production, data access, and what happens once the specialist has moved on to the next client. Draw that line before you need it, not during the argument.
The data team's case at its strongest: they inherit the system, they get paged when it breaks, and a specialist who is gone in 4 months has less at stake in getting the production detail right than the person who will own it for years afterwards.
The specialist's case at its strongest: you hired them because your internal view was limited enough to need an outside judgement, and a judgement that folds the first time it meets internal pushback was not worth paying for in the first place.
When that split does not resolve it cleanly, escalate to whoever is still accountable for the outcome in 12 months. That is rarely the most senior person in the room, and it is never simply whoever argues hardest.
Fix it before it starts
Put decision rights in the scope document, not in your head. Name who signs off on model or method choice, who signs off on whether something is fit to go into production, and what happens if those two people disagree.
Do this while everyone is still being polite, before delivery pressure is on the table. A decision-rights paragraph written before the engagement starts costs nothing. Refereeing a live argument once a deadline is attached costs weeks, and usually costs trust as well.
When it means something else
Sometimes a recurring objection from the data team is not really about fitness for purpose, it is about control. A fractional hire is, functionally, doing work that used to sit with someone permanent, and that person is defending territory. That is a management conversation, not a technical one, and treating it as technical wastes everyone's time.
Sometimes it runs the other way. A specialist who keeps overriding operational concerns without engaging with them, rather than addressing them, is telling you how they will behave in the next disagreement too. Technical brilliance does not offset that, and it tends to get worse, not better, once the specialist feels secure in the role.
What to check this week
If you already have a live disagreement, do not wait for a scheduled review. Ask both parties, separately, what decision they think is theirs to make, and compare the answers. A gap between those two answers is usually the entire problem, dressed up as a technical dispute about models or data quality.
If you do not have a live disagreement yet, write the decision-rights paragraph now, while it costs nothing. It takes one meeting. The alternative is finding out where the gaps are while a delivery date is already fixed and everyone is under pressure to be right rather than to be clear.
What to do about it
- Write decision rights into the scope document before the engagement starts, not after the first disagreement.
- Give the specialist the final call on method, and the data team the final call on production and ongoing ownership.
- Treat a recurring objection about control, rather than fitness for purpose, as a management problem, not a technical one.
- Don't hire under a borrowed title and expect an org chart to sort out authority on its own.
- Escalate stuck disagreements to whoever is accountable for the outcome in a year, not to the most senior or most persuasive voice in the room.
Questions people also ask
Should I always side with my in-house data team over an external hire?
No. Side with whoever's argument concerns the part of the system they'll actually own. On production and inherited operational risk, that's usually your data team; on whether the recommended method fits the business goal, that's what you brought the specialist in to judge. Blanket loyalty to whichever side is internal treats the disagreement as a politics problem rather than a decision-rights problem, and it will recur.
What if the disagreement happens midway through a live project with a deadline?
Resolve it against the decision-rights split you should have written down at the start, even if you didn't. If nothing was written down, make the call yourself based on who owns the consequence once the specialist leaves, then write the rule down immediately afterwards so the next disagreement doesn't have to reach you.
Is a disagreement like this a sign I hired the wrong specialist?
Not on its own. A specialist who never gets pushback from your data team is either always agreeing with existing assumptions or not being properly heard. Worry if the specialist can't explain their reasoning clearly to a technical peer, or if they override operational objections without engaging with them at all.
How specific should the decision-rights section of the contract be?
Specific enough to survive a real argument. Name the individual who signs off model or method choice, the individual who signs off production readiness, and what happens if those two people disagree. A general clause about collaborative working will not hold up once a deadline is attached and the disagreement turns real.
Where the figures come from
Every rate and salary quoted in this article is a median or percentile of figures advertised in UK job postings over the six months to 8 August 2026. They are not rates paid, and the gap widens at the top of a range.
- IT Jobs Watch, UK contract rates, 6 months to 8 August 2026, read 2026-08-08.
- IT Jobs Watch, UK permanent salaries, 6 months to 8 August 2026, read 2026-08-08.