Skip to content
DT Nova

How to choose an AI development company or AI consulting firm

How to choose an AI development company matters more than which model it uses: the partner decides whether your AI project ends as a demo or as a system people trust and use every day. This guide covers the kinds of AI partner, the selection criteria, a scoring table, the running cost logic, the AI Act and GDPR checks, the risks, when an off-the-shelf AI tool is the better answer, and the questions to ask before you sign.

Use is growing fast. Eurostat reports that 19.95% of EU enterprises with ten or more staff used at least one AI technology in 2025, up from 13.48% in 2024 and 8.06% in 2023.

The short version

  • Choose a partner that measures answers, not one that shows the best demo.
  • Insist on a test set built from your own cases before the build starts.
  • Ask for running costs per use, not only the build price.
  • Keep code, prompts, test sets and provider accounts in your company's name.
  • Start with a small paid pilot on one real use case.

Kinds of AI partner

KindSuitsWatch out for
AI consulting firmStrategy, use-case selection, governance, vendor choiceAdvice that stops before anything works
AI development companyBuilding and running assistants, agents, document AI, searchDemos that are not tested against real cases
Platform vendorA ready AI product for a common taskLock-in, data use terms, limited fit
Large consultancyBig programmes across many unitsJunior staffing, heavy overhead for small projects

Selection criteria

  1. Evaluation method. How they measure accuracy, with test sets and thresholds agreed with you.
  2. Working references. Systems in daily use, and clients you can talk to.
  3. Data handling. Where data is processed, whether it is used for training, how access is logged.
  4. Model neutrality. Willingness to choose among models and switch later.
  5. Operations. Monitoring, cost tracking, updates when models change.
  6. Regulation. A clear view of your role under the AI Act and what it requires.
  7. Ownership. Code, prompts, configuration and accounts in your name.
  8. Integration skill. AI is only useful when connected to the systems where work happens.

A scoring table for AI partners

CriterionWeightGood evidence
Evaluation method3A sample test report from an earlier project
Working references3A live system shown, a client you can talk to
Data handling3Processing agreement, EU processing options, no training on your data
Operations2Monitoring dashboards and a cost-per-use estimate
Regulation2A short written view of your AI Act role and duties
Ownership2Contract clause transferring code, prompts and test sets
Integration skill2Past integrations with systems like yours
Model neutrality1Experience with more than one model provider

Score from 1 to 5 per line, multiply by the weight and compare totals.

Total cost: build, run and oversight

AI systems have a running cost ordinary software does not: each use of a model is paid per token or request, and volume grows as people adopt the tool. Compare offers on these lines over three to five years:

  • Discovery, data preparation and the build.
  • Model usage at today's and expected volume.
  • Hosting, vector search or other infrastructure.
  • Monitoring, testing after model updates, and improvements.
  • Your own staff time to review answers and keep source material current.

Our guides to artificial intelligence cost and AI agent development cost break these lines down.

AI Act and GDPR checks

Article 4 of the AI Act has applied since 2 February 2025. As replaced by Regulation (EU) 2026/1744, it requires providers and deployers of AI systems to take measures to support the development of AI literacy of their staff and other persons operating and using AI systems on their behalf, and it does not require them to guarantee any specific level for any individual. The AI Act applies in general from 2 August 2026; obligations for high-risk uses listed in Annex III apply from 2 December 2027, and for Annex I products from 2 August 2028. Article 50 sets transparency duties, such as telling people when they interact with an AI system.

Under the GDPR, a partner that processes personal data for your AI system is your processor (Article 28), and decisions with legal or similarly significant effects based solely on automated processing are restricted by Article 22. Ask how the design keeps a human in the loop where it matters. Our guide to AI governance covers the internal side.

Risks and red flags

  • A polished demo, but no test results on your own cases.
  • No answer to "what does it cost per use at our volume?"
  • Your data sent to services outside your control without a processing agreement.
  • One model hard-wired everywhere, with no way to switch.
  • Provider accounts in the partner's name.
  • Promises of full automation for decisions that need human review.

When an off-the-shelf AI tool is the right answer

  • General writing, summarising and search help for staff.
  • AI features already inside the software you use, such as your CRM or office suite.
  • Tasks where errors are easy to spot and fix.

Custom work pays off when AI must act on your own data and processes, give answers you can check, and stay under your control. Our build vs buy software guide covers the method, and AI strategy helps choose the first use case.

Questions to ask before you sign

  1. Which of your AI systems are in daily use, and can we talk to that client?
  2. How will you measure whether answers are right, and what threshold counts as done?
  3. Where will our data be processed, and is it used to train any model?
  4. What will each use cost at our expected volume?
  5. What happens when the model provider changes or retires a model?
  6. What is our role under the AI Act, and what must we do?
  7. Will code, prompts, test sets and accounts be ours?

Our AI consulting service starts with one use case and a measured pilot, and AI agent development covers systems that act inside your tools. Every guide is in the guides library.

Questions people ask

What should we look for in an AI development company?

Working systems you can see, a clear method for measuring whether answers are right, a written view on data protection and hosting, knowledge of the EU AI Act, a plan for monitoring after launch, an estimate of running costs per use, and a contract that gives your company the code, prompts and accounts.

What is the difference between an AI consulting firm and an AI development company?

An AI consulting firm mostly advises: strategy, use cases, governance and vendor choice. An AI development company builds and runs the systems. Many firms do both. If you need a working system, choose a partner that builds, or make sure the adviser stays involved through delivery.

How do we test an AI partner before a big project?

Start with a small paid pilot on one real use case with your own data and a test set of questions with known answers. Agree beforehand what accuracy or time saving counts as success. The result shows how the partner works and gives you a measured basis for the next step.

Who owns an AI system a partner builds for us?

It depends on the contract. Ask for the rights in the code, prompts, configuration and test sets to transfer to your company, for the accounts with model and cloud providers to be in your name, and for a clear statement that your data is not used to train anyone else's models.

Does the EU AI Act apply to a small AI project?

The AI literacy duty in Article 4 applies to providers and deployers of any AI system, and transparency duties under Article 50 apply to certain systems such as chatbots. The heavier rules apply to high-risk uses listed in Annex III, such as recruitment and credit scoring, from 2 December 2027. A good partner can tell you which applies.

Sources