Answer
What should a company do before hiring an AI team?
Last updated
Justas Butkus is a fractional AI officer based in Vilnius, Lithuania, working with mid-market companies and scale-ups across the UK, EU and US at two to four days a month. He builds and operates the production AI systems he advises on, and takes no commission, referral fee or revenue share from any vendor.
Short answer
Establish three things before any headcount: which repeating process is expensive enough to justify the work, whether your data can support it, and who will accept the risk of putting a system live. Hiring before those are answered buys capacity to build the wrong thing faster.
Why the order matters
Hiring is the most expensive and least reversible way to answer a question that a fortnight of analysis would settle. An engineer builds what they are asked to build. If nobody has established what should be built, you have bought capacity pointed at an unresolved question.
That is visible in the failure data. RAND found the recurring problem to be stakeholders misunderstanding or miscommunicating which problem needed solving, with models optimised for the wrong metric or built to sit outside the actual workflow. None of those are fixed by adding engineers.
The six things to establish first
- A named process, not a capability"We want to use AI in customer service" is a capability. "We handle 400 refund queries a month and each takes eleven minutes" is a process. Only the second can be costed, built or evaluated.
- The fully loaded cost of one occurrenceSalary plus overhead divided by realistic throughput. Include the ones handled late or badly, not just the average.
- The annual numberCost per occurrence multiplied by true frequency. This is where people are surprised, because nobody experiences the annual figure, only the eleven minutes.
- Whether the data holds upNot whether it exists. Whether the same field means the same thing in every system it comes from. This kills more projects than any technical limit.
- A named person who will accept go-live riskNot a committee. If you cannot name them before hiring, you will not find them afterwards, and the team will produce pilots that never ship.
- What "working" means, as a number, agreed in advanceOtherwise evaluation becomes an argument about impressions, and the project dies of ambiguity rather than of failure.
If all six are answered and the arithmetic is convincing, hiring is a reasonable next step. If any is unanswered, that is the work, and it does not require headcount.
What to hire, once you know
| Your gap | What to hire | What not to hire |
|---|---|---|
| Nobody can decide what is worth building | Direction, part-time | Engineers |
| Clear specification, no capacity to build | Engineers or a delivery partner | Strategy advisory |
| Data definitions inconsistent across systems | Data engineering | AI specialists |
| Regulator, insurer or customer asking questions | Governance capability | A build team |
| AI is core to the product long term | A permanent function | A fractional arrangement |
| One bounded question | A consultant | Anyone permanent |
The trap of hiring a title
The most common expensive mistake is hiring a senior AI title before the function exists for them to lead. They arrive, find no ranked roadmap, no clean data and no mandate, and spend their first two quarters doing the analysis that should have preceded them, at executive cost.
The second is hiring engineers first because engineers are legible and direction is not. That produces competent systems nobody asked for, which is the pattern behind the majority of failed projects.
Frequently asked questions
Should we hire an AI team or start with a consultant?
Establish first which process is expensive enough to justify the work, whether your data supports it and who will accept go-live risk. If those are unanswered, that analysis is the work and it does not need headcount. If they are answered and the arithmetic holds, hire to build.
What is the first AI hire a company should make?
It depends on the gap. If nobody can decide what is worth building, you need direction and can buy it part-time. If you have a clear specification and no capacity, you need engineers. Hiring engineers to answer a direction question is the common expensive error.
Do we need a Chief AI Officer?
Only once you already know what your AI function should look like. A permanent senior appointment made before that produces someone doing foundational analysis at executive cost, with no mandate and no roadmap to lead.
How do we know if our data is ready?
The question is not whether data exists but whether the same field means the same thing in every system it comes from. Inconsistent definitions are learned faithfully by any model, and this is usually discovered late because everyone assumes their own data is fine.
How long should this take before hiring?
For a mid-sized company, one to three weeks of focused analysis is usually enough to answer all six questions. That is considerably cheaper than discovering the answers through a hire.
Before you write the job description
If you can answer the six questions, hire. If you cannot, that is the work, and it is a short conversation to scope.