AI consultant: what to demand before you sign anything
By Jacob Slot · Published
An AI consultant should deliver four things: a map of your processes, a solution that runs in production, the documentation that makes it maintainable, and ownership of all of it. If the fourth is missing, you have bought a subscription, not an asset.
That is the only way to make the term useful. "AI consultant" now covers everything from a management firm selling a report to a freelancer who builds a flow in a week, and those two have neither the same price, the same risk, nor the same result.
We build this kind of solution for Danish small and midsize businesses, so we see both sides of the table: the proposals customers turned down, and the projects they paid for once without getting anything they could use. This guide is written to help you avoid the second kind.
What an AI consultant does, and what the role does not cover
An AI consultant maps the processes in your business that cost time without creating value, designs a solution where AI or automation takes over the work, builds it, and hands it over to your organization. The role is half business understanding and half engineering, and a good consultant spends more time on the first than on the second.
Three things look like the role but are something else. A staffing agency sells you hours from a specialist you still have to manage. An AI trainer raises your employees' skill level but leaves no solution running. And a management firm delivers a strategy and a roadmap that still require you to find someone to build.
Any of the three can be the right purchase. They are just not the same purchase, and a vendor who does not make the difference clear in the first meeting will not make it clear in the proposal either.
The need is real. The use of artificial intelligence in Danish companies with more than ten employees rose from 15% in 2023 to 28% in 2024, and 70% of users name more efficient workflows as the positive effect, according to Statistics Denmark. But the spread is uneven: 96% of companies with more than 250 employees are using it, against 57% of those with fewer than ten, according to Danish Chamber of Commerce's analysis of AI use in 2025. The gap is rarely technology. It is that nobody in the small company has had time to take the first step.
The four deliverables an AI consultant must hand over
Almost every proposal describes what the consultant will do. Very few describe what you hold in your hands when the work is over. That is the list you need to negotiate.
| Deliverable | What it looks like | When you get it |
|---|---|---|
| Process map | An overview of the processes examined, with weekly time spent and a reasoned priority order | After the initial analysis, before building starts |
| Live solution | The automation runs on your data in your environment, not in the vendor's demo | At handover, after a period of bug fixing |
| Documentation | What the flow does, where it runs, what happens when it fails, and who gets notified | Together with the solution, not as a later deliverable |
| Ownership | Code, flows, accounts, and access are held in your company's name | At payment, written into the agreement |
The process map is the most important and the one most often missing. It should be readable by someone who was not in the meetings, and it should include the processes the consultant advised against automating, and why. A document that only describes what was sold is a sales document.
Documentation decides what the solution is worth in a year. A flow nobody can explain gets switched off the first time it fails, and then you have paid for a one-time gain. Require that the documentation covers error handling, not just the normal path through the system.
Advice or building: two different purchases
You can buy advice, and you can buy a solution. Advice is the right route if you have many departments that must agree on a direction and the decision itself is the hard part. Then you are paying for a priority list and a plan.
If instead you have one specific process that costs time every week, advice is typically an unnecessary middleman. You know the problem. What you lack is someone who builds the solution and makes sure it keeps running without them.
You can see the difference in the proposal on one line: what is the deliverable? If it says "workshop," "analysis," "strategy," and "recommendations," you are buying advice. If it says "live flow," "integration with named systems," and "handover," you are buying a solution. We have broken down what AI automation costs in both cases, because the gap is usually larger than the choice of vendor.
Six questions to ask before you hire an AI consultant
References and case studies tell you what went well. The questions below tell you how the vendor works when something goes wrong. Ask them in the first meeting and notice how quickly the answer becomes concrete.
1. Which process would you start with, and why that one?
A good answer names a process in your business, an estimate of the time it costs, and a reason it is the easiest win to take first. A bad answer describes a platform or a technology. If the vendor cannot point to a process after one conversation about your company, they have not listened or have not understood.
2. What does the solution do when it fails?
Automations fail. An invoice is missing a field, a system does not respond, a customer writes something unexpected. A good answer describes what happens: the case is queued, a person is notified, and nothing is passed on half finished. A bad answer says it rarely happens.
3. Who owns the code, the flows, and the accounts afterward?
There is only one right answer: you do, at payment, and it is in the agreement. Anything else is lock-in, however kindly it is phrased. Ask directly whether the solution runs in your own account with the tools used, or in the vendor's.
4. What do we need to provide ourselves?
A good answer is precise and slightly uncomfortable: an internal project owner with decision authority, access to the relevant systems, and someone who can explain how the process actually runs today. An answer that "we'll handle everything" means someone on your side will need the time anyway, just unplanned.
5. What would you advise us not to automate?
This is the most revealing question. An experienced AI consultant has a list ready: processes that change every month, tasks with too few repetitions to earn back the investment, and judgment calls that need a person. A vendor who thinks everything can be automated has never cleaned up after a project.
6. What does the handover look like, and what does help afterward cost?
A good answer has a date, a named recipient on your side, and a short training session. It also has a price for support afterward that you can choose to take or leave. A bad answer makes support a condition for the solution working.
Ownership and exit: what you must own when the engagement ends
The costliest mistake in this category is not about technology. It is about whose name is on the account. A solution that runs in the vendor's environment, on the vendor's keys, with the vendor's documentation, is not yours, no matter what you paid.
Five things must stand in your company's name before you sign:
- Source code, flows, and configuration, including an export of everything in a format you can open.
- Accounts with the tools and models the solution uses, and the keys that belong to them.
- Access to the data the solution produces, including the history.
- Documentation, and the training your employees received, in writing.
- The right to let another vendor continue the work without starting over.
The test is simple. Ask the vendor what happens to the solution if you end the engagement in three months. The answer "it keeps running, and you or someone else can maintain it" is the only one that holds up. With us it is part of the agreement, not a negotiation, and it is one of the reasons we dare to write ROI within 90 days into the contract.
What a realistic engagement looks like
A focused project in a smaller company is short. We work for about 3 weeks from audit to launch on a defined process, and that pace is possible because the scope is small and the decisions sit with one person. Our entire process is built around that sequence.
If the proposal runs six months, the scope is too big for a first project. Split it up, and get the first piece live before you decide on the next. It is also the cheapest way to find out whether the vendor is the right one.
How to measure your AI consultant once the solution is live
Everyone measures the proposal. Almost nobody measures the result. Three numbers are enough, and they must be agreed before the work starts, because otherwise they turn into a feeling.
The first is time saved per week in the process that was automated. Measure it before and after, the same way both times, and let the employee who does the work own the measurement. For our customers it typically comes to 15 hours per week, and at Markeity it equals 20,000 DKK/month on systems we built once.
The second is the share of cases the solution handles on its own. Measure it on the cases it was built for, not on everything that comes in. If it falls over time, it is almost always because reality has moved and nobody has updated the flow.
The third is the time from a failure occurring to it being fixed. That is the number that decides whether you really own the solution. If nobody on your side can answer it, you have a deliverable that depends on a phone number. If you want to put numbers on the whole equation, you can calculate the ROI of an AI project up front and use it as the basis for the agreement.
What an AI consultant costs, and what grants do to the math
The market is opaque, and hourly rates say little when you do not know how many hours it will take. Ask instead for a fixed price on a defined deliverable, or an hourly rate with a cap. A proposal without a cap is not a proposal.
Our own prices are on our website: a defined build starts at $1,500 one-time (the Starter Build), and an ongoing partnership starts at $750/month (the Growth Partnership). We have built 53 automations on that model. It is not the market average, but it is a number you can hold other proposals up against.
And count the grant. The SMV:Digital advisory grant covers 50% of your spend on external advisory, and the grant can be up to 150,000 DKK. The funding pools open in rounds and close when they are empty, so the date matters as much as the application. We have collected the schemes in our guide to grants for your AI project.
Key numbers
15 hours
saved per week for a typical customer
90 days
to measurable ROI, written into the contract
FAQ
What does an AI consultant do?
An AI consultant finds the processes in your business that cost time without creating value, designs and builds a solution where AI or automation takes over the work, and hands it over to your organization. In practice, most of the time goes into understanding the process, not writing code.
What does an AI consultant cost?
It depends more on scope than on the vendor, and hourly rates alone tell you nothing. Ask for a fixed price on a defined deliverable or an hourly rate with a cap, and have handover and support priced separately. Our own prices start at $1,500 one-time for a defined build (the Starter Build). Keep in mind that AI automation in a smaller company rarely requires a big project first.
Do I need an AI strategy before I hire an AI consultant?
No. You need one concrete problem, an estimate of what it costs you per month, and a person internally who owns the project. The strategy is the consultant's job, not your prerequisite, and a vendor who demands a strategy first is selling you an extra step.
How long does an AI project take in a smaller company?
A defined process can be live in about 3 weeks, if the decisions sit with one person and access to the systems is in place. Integrations with older finance or CRM systems stretch it out. If the proposal runs half a year, the scope is too big for a first project.
How do I know whether an AI consultant is good enough?
Ask to see a flow that is running, not a presentation. Ask for three references from companies your size that you may call. And ask what they would advise you not to automate. The answer to that last one separates the advisor from the salesperson faster than anything else.
You do not have to pick an AI consultant on instinct. Ask the six questions, demand the four deliverables, and get ownership in writing, and the rest is easy to judge.