Guide: How-to
How to Evaluate AI Consulting Proposals Before Signing: A Guide to Going In With Your Eyes Open
NILG.AI · November 9, 2026
The 70,000-euro difference nobody explains to you
Three proposals on the table. One at 50,000, another at 80,000, another at 120,000. All promise AI. All look professional. And you can't work out what justifies the difference.
This happens every day. And the reason it's confusing has nothing to do with a lack of intelligence on your part. Most AI consultancy proposals were not written to help you decide. They were written to impress you.
What you need is not more technology in the document. It's knowing what to ask.
What a serious proposal has to answer before you get to the price
Before you look at the numbers, check whether the proposal answers these three questions directly:
What specific business problem does this solve?
If the proposal talks about "implementing a machine learning model" without saying which process will change, which inefficiency will be eliminated, and what the expected impact is in money or time, send it back and ask for clarity. Technology without a defined business problem is decoration.
How will you know it worked?
A serious proposal defines success metrics before the project starts. Not after. Without clear acceptance criteria (X% reduction in process time, Y% drop in errors, Z% increase in conversion rate), there is no way to hold anyone accountable for the outcome.
What is included and what is not?
This is where most money is lost. Does the proposal include integration with your existing systems? Training for the team that will use the solution? Maintenance of the model once it goes into production? A project at 50,000 that excludes integration can end up costing 90,000. One at 120,000 that includes everything can be the cheaper option.
Five signs a proposal has problems
1. The solution is defined before the diagnosis
If in the first meeting they proposed a specific solution without asking about your data, your processes, or your team, be cautious. No serious AI solution is defined without first understanding the context. It's like a doctor prescribing before examining.
2. Results are guaranteed in round percentages
"We'll increase efficiency by 40%." Nice. Where does that number come from? Based on what data? Under what conditions? Real results from AI projects come with ranges, assumptions, and caveats. If the numbers are too clean and lack context, they are marketing estimates, not technical commitments.
3. There is no mention of the post-production phase
The moment a project goes into production is not the end. It's the beginning. AI models degrade over time as data changes. A proposal that says nothing about ongoing monitoring, model retraining, or handover to your team is selling you half the work. You will find out about the other half later, at additional cost.
4. The price is not tied to clear delivery milestones
An 80,000-euro project should have defined milestones: what is delivered after one month, after three, after six. Payments tied to concrete deliverables protect you. Fixed payments on a schedule, regardless of progress, offer no protection at all.
5. There is no reference to your sector or your specific context
If you can swap your company's name for any other and the proposal still makes complete sense, it's a generic proposal. A consultancy with real experience in your sector will reference your type of data, your typical processes, and the common constraints of your industry. That level of specificity cannot be faked.
How to compare proposals with very different prices
The price difference between proposals is almost never unjustified. The problem is that the justification is rarely made explicit. These are the questions that make comparison possible:
Ask each consultancy: "If the project does not achieve the promised results, what happens?"
The answer to that question tells you more than any written proposal. A serious consultancy has a prepared answer: additional iteration at no cost, root cause analysis, scope adjustment. One that hesitates or steers the conversation elsewhere is telling you that when things go wrong, the responsibility will be yours.
Ask for a list of completed projects in contexts similar to yours. Not client logos on a slide deck. Projects with a description of the problem, the solution, and the outcome. If you cannot speak with someone who has worked with them, factor that in.
Ask who will actually work on your project. Many consultancies sell with senior profiles and deliver with junior ones. Ask for the CVs of the team doing the work, not the team that wrote the proposal.
The cost that never appears in the proposal
There is a cost that rarely shows up in any proposal but is often the largest one: the cost of a project that never reaches production.
A pilot that runs well in a controlled environment and then cannot be integrated into real systems, or that the team never adopts, or that stalls because the data quality is not there, that project is not worth what it cost. It is worth nothing. And sometimes it costs more than nothing, because it created expectations that were not met and left people inside the organisation distrustful of AI.
The real criterion for choosing a consultancy is not the price in the proposal. It is the likelihood that the project will reach production and generate real impact. That likelihood depends on methodology, experience, and the honesty with which you are told what can go wrong, before you sign.
If you want to understand how AI projects fail before reaching production so you can avoid the same mistakes, or how a demand forecasting case in the construction industry works in practice, those cases help calibrate what is reasonable to expect.
A good proposal is neither the cheapest nor the most expensive. It is the one that tells you clearly what will happen, when, with what success criteria, and with what responsibilities on each side. If you cannot find that in the proposals you have in front of you, ask for it. And if you cannot get a clear answer, you already have your answer.