AI Consulting: When It Makes Sense and What to Expect
"AI consulting" describes both work that changes a company's direction and an 80-slide deck nobody opens again. The difference isn't in the word — it's in what gets delivered, and when you hire it. This article explains what it is, when it makes sense for a mid-size company, what to expect from a good engagement, and the signs you're buying the wrong one.
What is AI consulting, really?
It's the work of deciding where AI is worth it in your company — and in what order — before you spend money building. Good consulting connects four things that usually live apart: the business (where the money is), the processes (where time leaks), the data (what's accessible) and the technology (what's feasible). The output isn't an abstract strategy; it's a short list of projects prioritised by impact and feasibility, with a sense of cost and risk for each.
The scope goes well beyond "train a model." It includes identifying real opportunities, assessing the company's data maturity, prioritising use cases, guiding architecture decisions and — in good consulting — staying through implementation so the thing actually generates value.
When does it make sense to hire?
In three concrete situations:
- You have high-volume repetitive processes. If the team spends hours a day answering the same questions, processing the same documents or generating the same reports, there's work with measurable ROI — and consulting tells you where to start.
- You've tried and nothing reached production. Several prototypes, lots of enthusiasm, nothing running for real. Here the value is diagnosing why — almost always a lack of reliability, not a lack of model.
- You're about to invest and don't want to waste it. Before committing budget, having someone separate what pays off from what's hype saves far more than it costs.
When it doesn't make sense: if you already know exactly which process to automate and it's simple and isolated, you may not need consulting — you need someone to build it well. Honest consulting tells you that.
What to expect from good AI consulting
A good engagement delivers verifiable things, not impressions:
- A diagnostic of the real state — which processes exist, which data is accessible, where the biggest return is.
- A prioritised list of projects — with estimated impact, feasibility and risk for each, not a wish list.
- An order to start in — which project is first and why, with the ROI maths done before you begin.
- Reliability and governance criteria — how it'll be measured, escalated to a person, and how sensitive data is handled.
- An honest sense of what not to do — the projects that look attractive but don't pay for themselves.
The sequence we always recommend is the same: start with one high-impact workflow, make it genuinely reliable, measure the ROI, and only then scale. The opposite — ten fronts at once — is the pattern we most often see fail. We lay it out in the AI transformation roadmap for SMEs.
The red flags of bad AI consulting
- It ends in a PDF, not an executable plan. If the delivery is strategy that never touches production, the value evaporates.
- It talks technology before your business. Whoever starts with the tool, not the problem, will sell you the tool.
- It promises ROI without seeing your numbers. Guaranteed return multiples before the diagnostic are marketing, not engineering.
- It recommends building everything from scratch when a ready solution exists. Honest consulting makes the buy-vs-build call in your favour, not its own.
Consulting that builds vs consulting that only advises
The distinction that matters most: some advise and disappear, and some stay to build what they recommended. The advantage of the second model is that no one recommends projects they can't deliver — the advice stays honest because whoever gives it will have to make it work. It's how we run AI strategy: the diagnostic points the way and the same engineering team builds the AI development and automation that follow. The cost trade-offs of building in-house vs outsourcing are in how to choose an AI development company.
To see the kind of projects a good diagnostic tends to surface, 40 AI project examples has the list by function and by sector.
Our format: the technical diagnostic
In practice, our "consulting" starts with a 30-minute technical diagnostic with a principal engineer — not a salesperson. We map your operation and you leave with three candidate projects, an order to start in, and a sense of cost and risk. No slides, no pitch. If it turns out it isn't time to invest in AI yet, we say so. Often the real blocker isn't strategy — it's reliability, which we cover in why your AI agent isn't reliable enough to scale.
Frequently asked questions
How much does AI consulting cost?
It varies a lot with the format. A focused initial diagnostic can be short and low-cost (ours is free the first time); a deeper strategy engagement, with data assessment and a roadmap, is a separate project. Be wary of anyone charging a lot for a PDF with no path to execution.
What's the difference between AI consulting and implementation?
Consulting decides what and in what order; implementation builds. The risk of splitting them across different vendors is getting recommendations that aren't feasible to build. That's why we prefer the same team to do both.
Do I need my data organised before starting?
Not for the diagnostic — part of the work is assessing the state of your data. For some more ambitious projects, organising data will be a prerequisite, and good consulting tells you that instead of pushing ahead regardless.
Is consulting for a small company or only for large ones?
It's mostly for SMEs, because that's where a bad bet hurts most. The format changes — for a smaller company, a focused diagnostic is worth more than a long strategy project — but the need to choose the first project well is equal or greater.