In-House AI Lead vs Consultancy vs Audit: How to Choose

Once a company decides it should be doing something with AI, the next question is not which model to use but who is going to do the work. Three routes are on the table: hire an in-house AI lead, engage a consultancy, or buy a productised assessment that tells you where AI pays back before you commit to either. They are usually framed as rivals. They are not really — they solve different problems, and each of them is the right answer in a scenario the other two handle badly. What follows is the scorecard, the scenario each route wins, the way each one fails, and the order most companies should take them in.

Quick Answer. Hire an in-house AI lead when AI work will run for a year or more and must compound inside the business; engage a consultancy for a bounded, one-off specialist build you would never staff permanently. Buy a productised assessment when you cannot yet name which processes to automate. All three fail without an internal owner.

Summary

Three routes to an AI capability — and the scenario each one wins
│
├─ Hire an in-house AI lead
│   ├─ You get — an owner inside the business, permanently
│   ├─ Wins when — AI work runs for a year or more and must compound
│   └─ Fails when — the role is vague, unsponsored, or hired too early
│
├─ Engage a consultancy
│   ├─ You get — senior specialist capacity, for a defined window
│   ├─ Wins when — a bounded build needs skills you will never keep
│   └─ Fails when — the knowledge leaves with the team at handover
│
├─ Buy a productised assessment
│   ├─ You get — a ranked shortlist and a phased roadmap, fixed scope
│   ├─ Wins when — you cannot yet name which processes to automate
│   └─ Fails when — nobody inside owns the roadmap once it lands
│
└─ How to choose
    ├─ Ask first — do we know WHAT to automate, or only THAT we should?
    ├─ Then ask — continuous work, or a one-off specialist build?
    └─ Usually a sequence — assess, then decide, then hire or engage

At a glance: the three-route scorecard

The table compares the routes on the dimensions that actually decide the call. Read it as a sorting tool, not a contest: the aim is to match your situation to a column, not to crown a winner in the abstract.

Three of those rows do most of the work: how long the AI work will last, how certain you need the cost to be, and who is left holding the knowledge when the work stops.

What the three routes have to solve

It helps to be precise about the problem, because the evidence says the binding constraint is rarely the technology. Among EU enterprises that considered AI and did not adopt it, the most commonly reported obstacle was a lack of relevant expertise, cited by around 71% — well ahead of legal uncertainty or data-protection concerns [1]. AI adoption itself remains uneven: about 20% of EU enterprises with ten or more employees used AI in 2025, but that splits into roughly 17% of small enterprises, 30% of medium-sized ones and 55% of large ones [1]. The gap is not access to models — the inference cost of a system performing at GPT-3.5's level of capability fell more than 280-fold between November 2022 and October 2024 [5]. The gap is knowing what to do and having someone able to do it.

That is one problem with two halves, and the three routes attack different halves. A productised assessment answers what to do. A consultancy and an in-house lead both supply someone able to do it — one rented, one owned. Conflating the halves is what produces the classic failure: hiring a capable person into a company that cannot yet tell them what the job is.

Route 1: hire an in-house AI lead

An in-house AI lead is a named person inside the business who owns the AI agenda: choosing which processes to change, running the pilots, managing vendors, and holding the standard for how AI is used day to day. In practice this is more often a promotion than an external hire. In the OECD's D4SME survey, reassigning existing staff was the most common way smaller firms met a digital-skills need — around 46% of small and 44% of medium-sized businesses — against 12% of medium-sized businesses that hired a new specialist [3].

There is a reason the promotion route dominates: the external hire is genuinely hard to close. Only 9.55% of EU enterprises recruited or tried to recruit ICT specialists in 2023, and of those, 57.5% had difficulty filling the vacancy — with a shortage of applicants the single most-cited reason, at around 43% [2]. Recruitment difficulty was broadly similar regardless of company size, but appetite was not: 6.23% of small enterprises tried to recruit an ICT specialist against 51.87% of large ones [2]. A smaller company that decides its answer is "hire someone" is entering the same market as everyone else with fewer of the things that win candidates.

The route's strength is durability. A lead with domain context knows which processes actually matter, can carry an unpopular change through the operations team, and gets better at the job every quarter. Its weakness is that the value is entirely contingent on the role being real: funded, sponsored by someone senior, and pointed at defined work. Hire into a vacuum and you get an expensive person waiting to be told what to do. That specific failure — and who inside your business is usually the better candidate — is the subject of our companion piece on the AI talent trap, which covers the hire route in depth; this article is about which of the three routes to take at all.

Route 2: engage a consultancy

Engaging a consultancy means renting expertise for a defined period: a diagnostic, an implementation, an integration, or a fractional senior role. The work is shaped to your situation, and it stops when the engagement stops.

This is a normal and well-established way for companies to fill capability gaps. The OECD notes that when smaller firms digitalise, they tend to outsource solutions, partly to compensate for weak internal capabilities and partly on cost grounds [4]. The same pattern shows up in survey data: 40% of medium-sized businesses in the D4SME survey said they contract external experts when they need digital skills, more than three times the share that hires a specialist [3]. Where firms look for those experts is telling — recommendations from other business owners (around 18%) and from digitalisation agencies (around 15%) dominate, while universities, innovation centres and government sources together account for about 1% of responses [3].

The route's strength is depth on demand. You get people who have done the specific thing before, at a seniority you could not justify hiring, for exactly as long as the problem lasts — and you stop paying when it ends. Its weakness is what happens at the end. Consultancy knowledge is portable by design, so unless the engagement is written with a named internal recipient and a real handover, the capability leaves when the team does, and the next question puts you back at the start. Cost is the other exposure: a day-rate engagement scales with scope, and scope is exactly what tends to grow once a diagnostic starts finding things.

Route 3: buy a productised assessment

A productised assessment is a fixed-scope diagnostic sold as a product rather than a project. The method is the same for every company that buys it, which is what makes the scope, the timeframe and the fee knowable in advance. The output is a prioritised view of your own processes — which ones repay automation, in what order, and what the first phase should contain. Our own AI Foundation Audit is one example of the category.

The case for buying the answer rather than reasoning your way to it is that "where do we start" is the problem most companies are actually stuck on. Lack of knowledge about digital tools and about how to start the transformation was the main barrier reported by non-digitalised businesses in the OECD survey, at around 43% [3], and 19% of respondents said they had no clear decision process for digitalisation at all [3]. That is a well-defined question with a repeatable method behind it — which is precisely the kind of work that productises well.

Now the limits, because they matter more than the pitch. An assessment does not build anything: it decides what to build, not how to build it, and the implementation still needs a lead, a vendor or both. It does not replace ongoing ownership — the roadmap covers a phase, not a permanent function. It is not a certification, a compliance audit, or legal or financial advice. And it is worth nothing at all if nobody inside the business owns the roadmap once it lands; a ranked list of opportunities with no owner is shelfware with better formatting. The honest claim for this route is narrow and specific: it removes the what question quickly and cheaply, so that the money you spend on the who question is spent on the right thing.

Side-by-side: where the three routes actually differ

The scorecard lists eight dimensions. Four of them decide most real cases.

Duration is the cost question. The rate comparison people reach for — salary versus day rate versus fee — is the wrong axis. What matters is how long the work lasts. A hire is a committed recurring cost that runs whether or not there is a year of work to justify it, plus recruitment and a ramp-up period before any output. A consultancy is a variable cost that stops when you stop. An assessment is a one-off with the scope fixed before you start. Continuous work makes the committed cost the cheapest per unit of output; intermittent work makes it the most expensive.

Speed to a usable answer differs by an order of magnitude. A hire is months away from its first decision — recruit or reassign, then ramp. A consultancy engagement is weeks from a first diagnostic. A productised assessment is days, because the method is already built and only your inputs are new. When a decision is blocking a budget cycle, that difference is often the whole argument.

Only one route leaves capability behind by default. The in-house lead accumulates context permanently. The consultancy accumulates it and takes it away unless handover is engineered into the contract. The assessment writes its reasoning down but cannot act on it. This is why the routes so often stack: the fastest way to get a durable capability is frequently to buy the answer, engage help to build the first thing, and put an internal owner in place to receive both.

Scope certainty runs opposite to flexibility. The productised route is certain because it is fixed — which is also its limitation, since it answers the questions its method asks and no others. The consultancy is flexible because it is negotiated, which is also why its cost is open-ended. The in-house lead is the most flexible of all and the least bounded: the role absorbs whatever AI work appears, for better and worse.

When each route wins

Each of the three is genuinely the best available answer in a specific situation. Here is which, and where each one breaks.

Choose an in-house AI lead when the AI work will be continuous for a year or more; when the value comes from repeated process change that has to compound; when domain context and internal credibility matter more than technical depth; and when you can define the role, fund it, sponsor it from the top, and keep the person. It fails when the role is created before the work is defined, when nobody senior sponsors it, or when a company enters a recruitment market where more than half of employers who try cannot fill the vacancy [2] without a plan for the likely outcome.

Choose a consultancy when the need is bounded and specialist — a one-off integration, a build that requires skills you will never keep in-house, a deadline set by someone else; when the seniority you need is not justifiable as a permanent hire; and when you want to convert a capability into a variable cost you can switch off. It fails when no internal person is named to receive the handover, when the brief is open enough for scope to expand indefinitely, or when the deliverable is a strategy nobody inside the company can actually operate.

Choose a productised assessment when you know AI matters but cannot yet name which processes to start with; when you need a defensible, evidence-based case for a board or a budget holder quickly; when scope and cost certainty matter more than bespoke depth; and when you want to de-risk a much larger commitment — a hire or an engagement — before making it. It fails when you already know exactly what to build (then you need build capacity, not another analysis), when the work needed is a deep bespoke engineering problem rather than a prioritisation problem, and above all when nobody inside owns the roadmap afterwards.

How to choose: the decision in order

Work through the questions in order rather than starting from the route you find most appealing.

In words, the tree branches like this:

• Can you name the processes AI should take on first? If not, assess first — a ranked shortlist and a roadmap — then come back to the next question with an answer.
• Will AI work run for a year or more? If yes, ask whether you can define, fund and keep the role; if no, go to the next question.
• Can you define, fund and keep the role? If yes, hire an in-house AI lead so ownership stays inside. If no, engage help now and name an internal owner later rather than creating a role you cannot sustain.
• Is it a bounded specialist build you could never staff? If yes, engage a consultancy on a fixed scope with a written handover.
• Neither continuous nor a bounded specialist build? Then no route is right yet — document and fix the process first. Automating an undocumented process just produces a poor outcome faster.

The order is the point. The first question is about the work, not the worker, and most companies that feel stuck are stuck there.

Sequence, not menu: how the three fit together

The framing of this comparison is a choice between three routes, because that is how the question arrives. In practice, the three are complementary far more often than they are mutually exclusive, and the sequence usually runs assess, then decide, then build or hire.

The reason is that the routes answer different questions. The assessment tells you which processes pay back and in what order, which is the input the other two routes need and neither reliably produces on its own. The consultancy supplies the hands to build the first thing on that list, at a depth you would not staff permanently. The in-house lead receives the work, runs the change through the business, and makes sure the second and third items on the roadmap actually happen. Skip the first step and you are hiring or contracting against a hunch. Skip the third and whatever gets built has no owner when the invoice is paid.

A worked shape makes the sequence concrete. Illustrative — the company below is a composite, not a client. Take a distributor with sixty staff that knows it is behind on AI but has never ranked its own processes. It assesses first, and the ranking comes back with supplier-invoice matching top, quotation drafting second, and the customer portal it had been talking about a distant fifth. It then engages a specialist firm on a fixed scope to build the invoice-matching integration, because that job needs an ERP connector nobody internally has built before and never will again. In parallel it names its existing operations manager as the internal owner, with time protected for it, so that when the specialists leave the second item on the roadmap still has someone to run it. Three routes, one order — and the total spend is likely to be lower than the same company would have committed by hiring first and finding out later that its top opportunity was not the one it had assumed.

There is one honest exception. If you already know exactly which process to automate and why, the assessment step adds cost without adding information — go straight to the who question. That situation is rarer than it feels, but it is real, and a comparison that pretended otherwise would not be worth reading.

Whichever route you take, the decision improves when it is made against evidence rather than instinct. That is the job our fixed-scope AI Foundation Audit does: it scores every repeatable process on return, suitability and risk, names the top three opportunities, and hands back a phased implementation roadmap you can put in front of a board — so the question of who does the work is answered against a real list. For the underlying method, start with our cornerstone guide to an AI opportunity assessment for SMBs.

Related insights

• AI Opportunity Assessment for SMBs — the parent method that ranks where automation pays back first.
• The AI Talent Trap — the deep dive on the hire route, and who inside your business is usually the better candidate.
• How to Prioritise AI Use Cases — how the ranking that precedes any of these routes is actually done.
• AI Implementation Roadmap for SMBs — what the roadmap an owner receives should contain.
• Which Processes to Automate First — the shortlist question, one level down.

Last updated: July 2026. Version 1.0.