Which AI Partner Do You Need?

Three kinds of help get sold under the same words, and they cost wildly different amounts. An AI consultant sells you a decision. An AI agency sells you a deliverable. An outsourced AI department sells you an outcome and stays accountable for it. Everything else — price, contract shape, who you meet, what happens in month nine — follows from that one difference. Here is how to tell which one your company actually needs, including the cases where the answer is none of them.

What is the difference between an AI consultant, an AI agency, and an outsourced AI department?

The difference is where accountability stops. A consultant’s obligation ends when the plan is accepted. An agency’s ends when the build passes acceptance testing. An outsourced AI department’s does not end — it reports against your numbers for as long as the engagement runs. That single line explains the price gap and predicts which one will disappoint you.

The distinction matters because the failure modes are different, and each one is invisible at the point of sale. A consulting engagement fails quietly: the strategy is sound, nobody executes it, and eighteen months later the deck is still a deck. An agency engagement fails at the handover: the system works on the day it ships and slowly stops working, because monitoring and retraining belonged to nobody. A department engagement fails when the provider cannot actually build, and the monthly report becomes a status update rather than a measurement. Ask each candidate which of those three failures they have personally been part of. The honest ones have an answer.

	Consultant	Agency	Outsourced AI department
Gives you a plan	Yes	No	Yes
Builds working systems	No	Yes	Yes
Trains your team	Sometimes	Rarely	Yes
Owns the outcome after go-live	No	No	Yes
Reports measured results monthly	No	No	Yes
Engagement ends when	The plan is accepted	The build ships	You end it
Typical published rate	$400–900/hr (strategy houses)	$90–200/hr (engineering-led)	Monthly retainer

Rate figures are BD Emerson’s published August 2026 bands, described below.

What does each one cost in 2026?

BD Emerson’s August 2026 rate guide is the most specific public pricing anyone in this category has published: strategy houses $400–900 per hour, global integrators $150–350 per hour blended, engineering-led firms $90–200 per hour, and senior-staffed boutiques $200–400 per hour. It puts one production use case in the mid-market at $250,000 to $1.5M.

Read those bands carefully, because the hourly rate is the least informative number in any proposal. A senior-staffed boutique at $300 an hour frequently produces a lower program total than an integrator at $200, because the headcount is a fraction. What drives the total is how many people are on the engagement and for how long, and the biggest single swing inside that $250,000-to-$1.5M range is the condition of your data — not the model, not the vendor, and not the rate card.

Retainer-priced work is harder to compare because so little of it is published. That is a real gap and it cuts against buyers: when we probed ChatGPT for fractional AI leadership pricing in September 2026, it answered using fractional CTO benchmarks and said plainly that the AI-executive market is too immature to have standardised rates. Our own retainer and assessment prices are on the pricing page, and we publish them for exactly this reason. If a provider will not put a number in front of you before a second meeting, that is information.

How long before any of this pays back?

Longer than the vendors imply. Deloitte’s 2025 survey of 1,854 senior executives found most organisations reach satisfactory ROI on a typical AI use case in two to four years — against the seven to twelve months normally expected of a technology investment. Only 6% reported payback inside a year. Plan the engagement around that, or you will cancel a working programme at month ten.

The distribution is worth seeing rather than summarising: 6% under a year, 27% at one to two years, 37% at two to three, 25% at three to four. Deloitte’s respondents also gave a consistent reason it takes that long, and it is not the technology. AI is almost never deployed alone — it arrives alongside data cleanup, a team reorganisation, or a process change, which makes its specific contribution genuinely hard to isolate. One executive quoted in the study said they could only reach a ballpark estimate for exactly that reason.

This is the number that should decide your contract shape. If payback is realistically two years out, a three-month consulting engagement cannot deliver it and should not be sold as if it could. What a short engagement can deliver is a decision — which workflow, what the data looks like, whether the economics work at all. That is worth buying on its own, and it is a different purchase from ownership.

When is each one the wrong call?

A consultant is wrong when the problem is that nobody will execute. Consulting resolves uncertainty about what to do. If your company already knows what to do and has not done it for eight months, another plan will not change that, and you will have paid four figures an hour to document the stall.

An agency is wrong when the specification is not settled. Agencies price and staff against a defined build. Hand them an unsettled problem and you get change orders, because discovery you did not buy is happening inside a fixed-price contract. If you cannot write down what the system does in a paragraph, you are not ready to brief an agency.

An outsourced AI department is wrong when you already have an owner. This is the one we have the most reason to soft-pedal and will not. If a named person inside your company owns AI decisions and has genuine time for them, you do not need a department — you need capacity or a specific skill, and you should buy that narrowly. Paying an ongoing retainer for ownership you already have is the most common way companies overspend here.

All three are wrong when the underlying decision is contested. If operations and finance disagree about whether the dispatch process should change, no external party can settle that for you. They can only document the disagreement expensively. Settle it, then buy help executing it.

And here is where we specifically are the wrong answer: if you need a partner across forty countries, or a regulated model-risk programme that has to survive an examiner, we are not it. Those practices exist at Deloitte and IBM for good reasons. We work with $10M–$50M companies on one workflow at a time.

How do you choose in one question?

Ask who is accountable in month nine. Not who builds it, not who advises on it — who is answerable when the system has been live for a quarter and the number has or has not moved. If the answer is someone inside your company, buy a consultant or an agency. If the answer is nobody, that is the gap, and it is the only gap an outsourced AI department is worth paying for.

Before you brief anyone, it helps to arrive with something concrete. Knowing which workflow, what the data looks like, and what “better” would be worth in dollars changes the conversation from a capability pitch into a scoping exercise — and it makes competing proposals comparable for the first time.

Where to go next

Start with what an outsourced AI department actually is if the category itself is new, and what one costs for the banded figures. If you are weighing named firms, how to choose an AI consulting firm covers the seven tests and why most “top firms” lists are written by firms on them. To arrive at any of these conversations with real numbers, what an AI readiness assessment measures and how to measure AI readiness are the groundwork, and the free AI Opportunity Assessment maps one workflow end to end. More in the outsourced AI department hub.

Sources

• BD Emerson, Best AI Consulting Firms in 2026, by Drew Danner, 21 August 2026 — all four published rate bands and the $250,000–$1.5M mid-market use-case figure.
• Deloitte, AI ROI: The paradox of rising investment and elusive returns, 22 October 2025 — the two-to-four-year payback finding, the 6% under-one-year figure, the full distribution, and the attribution difficulty. Survey of 1,854 senior executives across Europe and the Middle East, 15 August to 5 September 2025, with 24 follow-up interviews.
• MIT Sloan Management Review, Three Approaches to Measuring and Managing AI ROI, by Mika Ruokonen and Paavo Ritala, 23 June 2026 — that two companies making near-identical AI investments routinely define success in entirely different ways, from interviews with more than 30 CEOs and senior leaders.

Method note. Updated 6 September 2026. The rate bands and payback figures are quoted from the named published sources above and are not our own survey data; treat them as the most specific public reference points available rather than as market averages. The observation about AI-executive pricing comes from a scripted ChatGPT probe we run weekly and logged on 6 September 2026 — it describes what one model returned on one date, not a market rate. The accountability framing and the four “wrong call” cases are our own judgment from scoping conversations, not third-party findings.