AI consultancy at INDOLES starts with two fixed-price pieces of work: the three-week Digital Transformation Audit at €5,500 and the six-week AI Pilot at €15,000 ($6,000 and $16,500), both list prices excluding VAT. A first engagement that starts with one package therefore falls between €5,500 and €15,000; take both back to back and the list prices add up to €20,500. Two items sit outside that price and cannot be fixed: consumption-based costs such as model usage and cloud fees depend on your usage volume, and the move to production after the pilot depends on what the pilot measures — it becomes a written proposal built on the roadmap in the pilot report.
Kerem is director of operations at a packaging manufacturer, and the board wants a single budget figure for AI from him. The two proposals on his desk do not describe the same work: one offers an open-ended "AI transformation" for a fixed monthly fee, the other quotes a one-off figure for setting up a chatbot, and neither says how much the model will run up in usage fees each month. I invented Kerem for this article; I did not invent his question — "how much does AI consultancy cost?" is one of the first questions we hear in a buying conversation.
The honest answer comes in two parts: the part that can be fixed and the part that cannot. Below I open up our own price list first, then the five variables that grow the price, the lines that sit outside it, and how the budget after the pilot gets written. How to test a consultant in the first meeting is covered in 12 questions to ask an AI consultant; how a pilot runs day by day is in the 90-day pilot framework; and the method behind the work is on the AI advisory service page.
What is the price band for AI consultancy at INDOLES?
For a first engagement that starts with one package, the price band runs from €5,500 to €15,000 excluding VAT, and it is made up of two fixed-scope packages. Depending on where you start, it plays out in one of three ways:
- Audit only: €5,500, three weeks. When it is not yet clear which process is a candidate for AI and which for classic automation.
- Pilot only: €15,000, six weeks. When the problem to solve and its data are already clear.
- Audit, then pilot: €20,500 in total at list prices, nine weeks of work. When one of the candidates the audit ranks is to be tried in the field.
The Digital Transformation Audit spends three weeks observing 3-5 processes on site, running 5-10 structured interviews with process owners and drawing the current-state maps. Its output is 3-5 pilot recommendations ranked by ROI projection, a separate technical spec for each one and a six-month implementation roadmap. The audit does not look at AI alone; a finding that "this process does not need AI" is a valid output too, and it keeps budget from being tied to the wrong place.
The AI Pilot gives six weeks to a single use case. Its scope has five items: use case selection and value validation, a data inventory and quality check, model selection — a large language model, classical machine learning or a mix of the two — a prototype that operators or end users can work with directly, and a two-week field test with real users. What gets delivered is the working prototype with its source code, a pilot report covering the metric impact and the cost analysis, and a roadmap setting out the technical steps, estimated budget and timeline for moving to production. The source code stays with the client in full ownership.
Should you start with the audit or go straight to the pilot?
If the problem is clear, start with the pilot; if it is not, start with the audit. The pilot puts six weeks into one use case, and the wrong use case spends all six; the audit ranks the candidates from data in three weeks. The €5,500 paid for the audit is the price of opening a €15,000 pilot on the right problem.
The clarity test is simple: can you write one sentence? "We want to cut the time it takes to prepare quotes; quote requests are on record in email, and the process owner is the sales manager." If that sentence can be written — process, record and owner all known — the pilot can start directly. If the sentence stays at "how do we benefit from AI", the right purchase is the audit rather than the pilot. When the audit recommends a candidate on the AI side, the pilot starts straight from that candidate's spec; the two packages were designed to run back to back for exactly that reason.
What makes AI consulting prices vary?
Five variables set AI consulting prices: how ready the data is, the number of use cases, the integration surface, the scope of the field test and its users, and usage volume. The first four grow the cost of the build; the last grows the running cost that comes back every month. When two proposals sold under the same heading differ several times over, the reason is usually that one of the five was assumed differently.
A fixed-price package does not remove these variables; it bounds them. In the AI Pilot there is one use case, at most two candidates and a two-week field test; in the audit the scope is 3-5 processes. Inside those bounds the price stays fixed. If a variable crosses its bound — a second use case, a sixth process, a second site — duration and price are recalculated in writing; the boundary is never widened quietly. The four sections below take the five variables in turn, and each ends with the line to look for in a proposal.
What does data preparation add to the budget?
Data preparation usually adds to the budget not on the invoice but in the calendar and in your team's hours. Collecting records, finding empty fields, opening access rights and drawing up a cleaning plan can take longer than building the model; if this line is missing from a proposal, its cost quietly lands on you.
In the AI Pilot, the data inventory and quality check are inside the price and happen in the first weeks. If the data falls short, that is said while the use case is being chosen rather than halfway through the pilot: either the use case changes or measurement and record-keeping get set up first. We don't start a pilot on incomplete data, because a pilot measured on incomplete data gives no decision anything to stand on. That is also why the pilot starts with at most two candidates; whichever has its data ready gets picked.
Two data jobs sit outside the price. The first is creating a record where none exists; that belongs to a digitisation step, not to a pilot. The second is an ongoing data labelling operation; labelling is not a one-off but a monthly job, and it sits outside our AI consultancy scope. The line to look for in the proposal: who prepares the data, how many weeks it takes, and what happens if the data falls short.
How do the number of use cases and the integration surface grow the price?
Every new use case and every new connection arrives with its own data preparation, its own testing and its own failure cases, so the price grows with the number of use cases and systems. The AI Pilot is built on a single use case and a single flow: one customer segment, one channel or one order flow.
On the use case side the rule is clear. The use case stays fixed for the whole pilot; switching it after the field test starts resets the pilot. Minor corrections inside the prototype are in scope, a new problem definition is not. A second use case is priced as a separate pilot and needs its own six weeks: six weeks is planned for one problem, and split in two, both come out half-finished.
On the integration side, the real question is how many places the system connects to, and whether it only reads from them or also writes back. A prototype that reads data from the ERP and a production system that writes orders into the ERP are not the same job; the second needs permissions, error handling and a way to roll back. The pilot prototype is built on a core architecture that can move to production, but the full set of production integrations belongs to the production roadmap rather than to the pilot. The line to look for in the proposal: which systems get connected, whether the connection reads or writes, and who writes the integration.
What do the field test and the number of users change?
The field test changes not the price of the pilot but the cost around it. In the AI Pilot the field test lasts two weeks with real users, and those two weeks are inside the price. As more users join the test, two lines grow: the usage fees the model consumes and the time your team puts in.
The field test is not simulated, because what gets measured is not laboratory accuracy but the unit time of the process on real work. That has a cost: operators, the marketing team or customer representatives work with the new system for two weeks, the process may run slower than usual in the first days, and the process owner runs the measurement. That time never appears in a proposal, but it is a real cost; leave it out of the budget and the pilot looks cheaper than it is.
In production the number of users matters even more. A system used by five operators and one used by five hundred dealers can run on the same model, but their running costs, support load and training needs are not the same. The line to look for in the proposal: how many people take part in the field test, in which roles and for how long, and which assumption the number of production users was calculated on.
Why don't model usage fees go into a fixed price?
Model usage fees depend on consumption: the more requests the system handles, the longer those requests are and the model chosen, the more the bill moves. That is why they are not folded into a fixed package price; as usage rises the figure rises with it, and presenting it as fixed would mislead.
The pilot does not remove this uncertainty, but it measures it. The two-week field test produces the first data on real usage: how many requests came in, what each one consumed, which tasks fell to human review. The pilot report turns that data into an estimate of the annual running cost, and the production roadmap writes that estimate down as a separate budget line. You see the running budget with the report, not on the first invoice after the pilot.
The running cost has four components: model usage fees, cloud infrastructure, integration maintenance and human review. Where sensitive data is involved the model can run on your own infrastructure; server cost then takes the place of usage fees, while GPU server procurement and hardware operations sit outside the consultancy scope. Because the model layer is the fastest-moving and fastest-cheapening part of this field, a well-built system keeps the model as a replaceable component; when a better-value model appears, it should be possible to swap it in without rewriting the system. The line to look for in the proposal: the monthly running cost estimate, the usage assumption behind it, and whose name the accounts are opened in.
Which lines sit outside the package price?
The package price covers the five items on the scope list; everything outside it is named up front. In the Digital Transformation Audit, software licences, tool subscriptions, hardware and implementation labour sit outside the price; in the AI Pilot, model usage fees, cloud infrastructure and tool licences do. Four more pieces of work sit outside our AI consultancy scope as a whole:
- Training models from scratch and academic research: for most tasks, ready-made services or rule-based automation give the same result far more cheaply.
- Running an ongoing data labelling operation.
- GPU server procurement and hardware operations.
- Monthly usage fees for third-party AI services.
What sits outside is not invisible. Every pilot recommendation in the audit report comes with an estimated cost, and the pilot report writes down the production and running budget as estimates; those figures sit in the report, not on the invoice. If scope changes, repricing is put in writing, and no surprise line appears along the way. INDOLES is a reseller of the İKAS e-commerce platform, which is its only commercial tie on the software side; it has no partnership or commission relationship with any model, cloud or AI software provider it recommends in AI consulting. Because no share is taken from a recommended tool or model, the recommendation follows the requirement rather than the price.
How is the price of moving to production set after the pilot?
Moving to production has no fixed price; it is set by written proposal, based on what the pilot measured and on the production roadmap in the pilot report. The roadmap sets out the technical steps, the estimated budget and the timeline; the decision to go ahead belongs entirely to the client.
There are three reasons the price cannot be fixed in advance. The first is the outcome: a pilot closes with one of four decisions — scale, extend, fix and rerun, or stop — and each of the four carries a different budget. The second is scope: production takes in integrations, users and failure cases the pilot never touched, and their size is only known after the field test. The third is running cost: usage in production differs from usage in the field test and can only be estimated from the test's data.
I see that not as a gap but as an order of operations. The pilot's job is to move the production budget from guesswork to measurement; a buyer who asks for that figure before the pilot is asking for the price of a system nobody has measured. Because the source code is yours, the production work can be run by your internal team, an existing vendor or INDOLES. If what you are after is a finished product ready to go live, what you need is not a pilot but product development, and that work belongs to the MVP Build package.
Who is a fixed-scope package for, and who is it not for?
A fixed-scope package suits a company that wants a measured answer to one question: which process is a candidate for AI, or does this use case work in the field? It does not suit a company looking for ongoing AI capacity. We have no published monthly fee for AI consultancy; the work runs through two fixed-scope packages, and if the pilot pays off, the system is handed over to your in-house team. If it is not yet settled who you will work with — a large consultancy, a boutique team or an in-house team — we gathered the criteria for that decision in a separate article.
There are four situations where we don't recommend the packages. With no accumulated data, the model has no history to learn from and the pilot turns into guesswork; the record has to be built first. If no concrete operational problem can be named, a pilot is premature, and the audit comes first. If management has already settled on a particular tool and is only looking for endorsement, an independent audit will be uncomfortable; the right work is the installation. And if access to process owners and the operations team will be withheld, the audit falls short, because the map is drawn from the real flow the team describes.
One more warning: not every piece of AI work needs a six-week pilot. Connecting a chatbot or a ready-made service to a single screen can be done in a few days today, and the price should say so. Paying €15,000 for that would be the wrong purchase; a pilot is for a system whose result has to be measured and which has to connect to an existing process.
Which line is missing from a cheap-looking AI proposal?
A cheap-looking proposal usually gets cheap by deleting a line, and the deleted line is billed to you later — either in your team's hours or on the first usage invoice. Look in five places.
- No baseline: if today's value is not recorded before the work starts, the pilot's result can only be asserted, never proven.
- No data preparation: "you supply the data" hands you the longest part of the job.
- No running cost estimate: usage fees sitting outside the price is normal; never being given an estimate of them is not.
- Unclear ownership: if the source code, the rule and prompt sets and the accounts don't stay with you, you are leasing the system, not buying it.
- No stopping criterion: if nobody has written down below which figure the project closes, the pilot doesn't end; only its budget does.
Work out the real price of a cheap proposal in one line: the proposal amount, plus your team's hours on data and the field test, plus usage and infrastructure invoiced separately, plus the cost of getting the system back when the contract ends. The last item looks like zero, because nobody invoices it at signing. How to ask about these five in the first meeting is set out in the 12-question list.
What does it take for the pilot fee to pay for itself?
Rather than estimating the gain the pilot fee has to produce, calculate it; the calculation needs only two numbers: how many times a month the process runs, and how many months you expect the fee to take to come back. The result is the value each run of the process has to produce.
The figures below are hypothetical; they belong to none of our clients and were chosen only to show how the calculation is built. There are four assumptions:
- Process: preparing quotes. It runs 30 times a working day; at 20 working days a month, that is 600 runs a month.
- Cost: published list prices — €15,000 for the AI Pilot alone, €20,500 with the audit.
- Payback period: 12 months.
- Left out: the move to production and the monthly running cost are not in this calculation; both get added from the estimates in the pilot report.
Divide €15,000 by twelve months: €1,250 a month. Divide that by 600 runs: for the pilot fee to pay for itself within a year, each run has to produce about €2.08 of value. For €20,500 with the audit, the same calculation gives about €1,708 a month and about €2.85 per run. The value comes from three places: time saved, errors avoided and requests that no longer slip away. Only you know how many minutes of work €2.08 per run buys in your business; run the calculation with your own fully loaded cost per staff hour.
If the same process runs 60 times a month instead of 600 — three times a day — the value needed per run rises tenfold, to about €20.83. How often a process repeats matters more than the price itself, which is why the first pilot goes to the process that repeats most. The threshold for scaling a pilot sits one step higher than this calculation, because the gain also has to cover the running cost. I set out how we write that threshold in the success and stopping criteria section of the 90-day pilot framework.
Conclusion: which three figures should you ask for before approving the budget?
The price of AI consultancy is not one figure but three: the fixed build fee, the estimated monthly running cost and the estimated budget for moving to production. The first is written in the proposal; the second and third often are not, yet a budget approved without them gets reopened at the first usage invoice.
Here is a test you can run today: open the proposal on your desk and look for three lines. What does the fixed fee cover, by name? On which usage assumption was the monthly running cost estimated? When, and in which document, will the production budget be visible? If any of the three has no answer, that price is not yet a price; it is an opening figure.
Ask us the same three questions. The first figure is written in this article and on the package pages: the Digital Transformation Audit at €5,500 and the AI Pilot at €15,000. The second and third arrive with the pilot report; how we work is set out on the AI advisory service page.