To choose an AI (artificial intelligence) consulting company, ask whether it has worked on a decision like yours and how it measured the result. Data Department, an Istanbul-based AI and machine learning consultancy founded in 2013, suggests these 10 questions for comparing proposals: references, who does the work, data ownership, data protection, ERP (enterprise resource planning) integration, success metrics and what happens after the pilot.
Why does the choice of consultant matter so much?
Most companies hire an AI consultant because they do not yet have the expertise in-house. In Türkiye, 14% of enterprises with 10 or more employees used AI in 2026, up from 7.5% a year earlier, and 72.3% of those that considered AI but did not use it named lack of expertise as the main barrier (TurkStat data reported by Bloomberg HT, 2026).
When the expertise is missing inside the company, it is also hard to judge it from outside. Pitch decks look alike and demos are impressive, but far fewer projects end up in daily use.
The questions below help you tell whether real experience sits behind a proposal. For each one we describe what a good answer looks like and which answer is a warning sign. They apply whether you are searching for an AI consulting company in Turkey, a machine learning consultancy in Istanbul or a firm anywhere else.
Which 10 questions should you ask?
1. Can you describe work you have done on a decision similar to ours?
Experience in your sector is not enough; you need experience with a similar decision, such as how much stock to hold, what price to set, which route to drive or which part to reject. A good answer states clearly which decision was addressed and which business metric measured the result. NDAs (non-disclosure agreements) may prevent the consultant from sharing client names and project details; that is normal. For example, our spare-parts work for an automotive distribution group gave about 10% lower inventory holding cost (case study). Warning sign: only a demo or an "AI platform".
2. Who will actually do the work?
Ask how much time the senior people in the sales meeting will spend on the project. Request the names, roles and time allocation of the team in writing. Warning sign: the team in the proposal changes after signature.
3. Who will own our data and the models you build?
Your data should always remain yours. Ownership of the code, models and documentation built for the project should be written into the contract. Warning sign: the model runs only on the consultant's closed platform and cannot be moved.
4. How will you handle data protection, and where will the data be processed?
If the data includes personal data, ask where it will be processed, in which country, and whether it will be transferred abroad. In Türkiye this falls under the KVKK (Kişisel Verilerin Korunması Kanunu, the Turkish Personal Data Protection Law); for data about people in the European Union, the GDPR (General Data Protection Regulation) applies. A good answer proposes working on your own infrastructure, such as AWS (Amazon Web Services), Google Cloud, Azure or on-premises servers, and anonymising data where needed. Warning sign: "Just send us the data and we will take care of it."
5. How will the results be integrated with our ERP and MES?
If results stay in a separate dashboard, nobody uses them in daily work. Ask how the model output will be written into the ERP or the MES (manufacturing execution system), how often it will be updated and who will build the integration. Warning sign: integration is moved out of the proposal as "a later phase".
6. How will we measure success?
Success should be measured with a metric the finance team already tracks, such as inventory cost, waste, margin, revenue or hours saved, not with model accuracy alone. Agree on that metric on day one and measure the current baseline before the project starts. Warning sign: a promise of "95% accuracy" and nothing else.
7. What happens after the pilot?
Ask about deployment, monitoring, retraining when model performance drifts over time, and the handover plan to your team. A good answer makes clear who will run the system and what training your team will receive. Warning sign: no plan after the pilot, or an open-ended dependency on the consultant.
8. Is hardware in scope?
In computer vision or sensor projects, the contract should state who is responsible for cameras, sensors and installation. At Data Department we keep hardware out of scope: we cover the models, software and integration, and work with your hardware supplier. Whichever model a consultant uses, the boundary should be drawn at the start. Warning sign: hardware and software priced as one line with no detail.
9. When would you tell us AI is the wrong tool?
Some problems are solved with rules, better data collection or a simple spreadsheet. An honest consultant says so. Ask for an example of when they told a client "you do not need AI for this". Warning sign: the same solution for every problem.
10. How does the discovery phase work, and can we stop at the end of it?
A good process starts with a short, fixed-scope discovery: ranked use cases, a data check and a go or no-go decision. Ours takes 2 weeks and is followed by a pilot of 6 to 10 weeks. "Do not continue" should be an acceptable outcome of discovery. Warning sign: a large implementation project proposed without any discovery.
How can you use the questions as a checklist?
Use the table below to compare proposals side by side. Score each question 0 (no answer), 1 (general answer) or 2 (specific, written answer).
| # | Question | Sign of a good answer | Score (0–2) |
|---|---|---|---|
| 1 | Work on a similar decision | Decision named, business result measured | |
| 2 | Who does the work | Named team, time allocated | |
| 3 | Data and model ownership | Yours, in the contract | |
| 4 | Data protection and location | Your infrastructure, clear transfer terms | |
| 5 | ERP and MES integration | In scope, owner named | |
| 6 | Measuring success | Business metric agreed on day one | |
| 7 | After the pilot | Deployment, monitoring, handover plan | |
| 8 | Hardware scope | Boundary in writing | |
| 9 | When AI is not needed | A concrete example | |
| 10 | Discovery and stop decision | Fixed scope, clear decision point |
A total of 16 or more out of 20 points to a strong proposal. Below 10, ask for the missing answers in writing.
What should you check when hiring an AI consultancy in Turkey?
International buyers and Turkish companies searching in English usually ask three extra things. First, working language: we work in English and Turkish, and documentation can be delivered in either. Second, time zone: Türkiye is on UTC+3 (Coordinated Universal Time plus three hours) all year, which overlaps with most of the European working day and part of the US East Coast day. Third, contracts and data: ask which law governs the contract and confirm that data stays on your own cloud account or servers, so it does not need to leave your control. Data Department works with companies in Türkiye, Europe and the US.
What if you do not know which project to start with?
It is usually easier to choose the project first and the consultant second. You can rank candidates with the scoring table in how to choose your first AI project. We can also go through these questions together in the free 30-minute AI opportunity review. If you need a longer-term roadmap, our AI strategy and fractional Chief AI Officer page describes part-time senior AI leadership, and what a fractional Chief AI Officer does explains the role in more detail.
Frequently asked questions
What is the difference between an AI consultant and a software company?
A software company usually builds a system that has already been defined. An AI consultant also decides which decision is worth modelling, whether the data is good enough and how success will be measured; often it builds the model as well.
What should be in the contract before a pilot starts?
Scope, the business metric and how the baseline will be measured, the team, data and model ownership, where the data will be processed and the options after the pilot. If these are not written down, disagreements tend to surface at the end of the pilot.
What should we do if the consultant wants to move our data to its own servers?
Ask in writing why it is needed and how the data will be protected under KVKK or GDPR. Most projects can run in your own cloud account or on your own servers.
Can we speak to a reference client?
You can ask, but many AI projects run under NDAs. In that case, an anonymised case study with the decision addressed and a concrete business metric is a reasonable alternative.