AI consulting has become a label that gets attached to everything from a genuine strategy engagement to a thinly disguised sales pitch for whatever the firm happens to build. Here is what a real AI consulting process should cover, so you know what to expect and what to push back on if it is missing.
It should start with your data, not with a model
Every AI project lives or dies on the data behind it, not on which model gets used. A proper consulting engagement starts by auditing what data you actually have, how clean it is, how much of it there is, and whether it is even structured in a way that a model could learn from. If a consultant jumps straight to recommending GPT-4 versus Claude versus a custom model before asking a single question about your data, that is a sign they are selling a solution rather than diagnosing your actual problem.
It should tell you when AI is not the answer
A lot of problems that get labelled as "AI opportunities" are actually solved more cheaply and more reliably with a simple rules-based system or a basic automation, no machine learning required. Good AI consulting includes an honest answer of "you do not need AI for this, a straightforward workflow tool will do the job for a fraction of the cost," even though that is not the answer that gets a consultant more billable work. If every conversation ends with a recommendation to build something complex, be skeptical.
What a real engagement typically covers
- Data readiness assessment: what data exists, its quality, and what is missing
- Use case prioritisation: which problems are worth solving first, based on cost saved or revenue gained versus effort required
- Build vs buy analysis: whether an existing tool solves the problem already, versus needing a custom build
- Risk assessment: where the system could fail, and what happens when it does
- A realistic cost and timeline estimate, not an optimistic one designed to win the deal
Red flags to watch for
Be cautious of any AI consulting engagement that promises a finished, production-ready AI system in a matter of days, that cannot explain in plain language what the model is actually doing, or that has no answer for what happens when the AI gets something wrong. AI systems fail differently to regular software, often confidently and silently rather than with an obvious error message, so any real engagement should include a plan for catching and correcting those failures, not just a plan for building the initial version.
When to bring in a consultant
The right time is before you have committed budget to a specific tool or vendor, when you still have room to change direction based on what the assessment finds. Bringing in consulting after you have already bought a platform and it is not working tends to be a much more expensive and limited conversation.
If you are trying to work out whether a specific process in your business is genuinely a good fit for AI, or whether a simpler solution would do the job just as well, our AI team starts every engagement with exactly that honest assessment before recommending a build.