All insights
ORBITRA INSIGHTS

How to Choose an AI Consulting Firm: 12 Questions to Ask Before You Sign

Twelve questions to ask before you hire an AI consulting firm, the red flags to watch for, how to structure the first engagement, and a scorecard you can copy.

By Orbitra AIPublished 9 min read
AI ConsultingEnterprise AIAI AutomationProcurement
How to Choose an AI Consulting Firm: 12 Questions to Ask Before You Sign

The number of firms offering "AI consulting" has grown fast in the last two years. Software agencies, digital marketing shops, large consultancies, and independent specialists now sell under the same label. The websites look alike, the demos are impressive, and everyone says they work "end to end". The cost of choosing badly is usually time rather than money: six months spent on a pilot that never reaches production.

This guide covers the twelve questions to ask before you choose an AI consulting firm, the warning signs to watch for, and how to structure the first engagement. As a consulting firm ourselves, these are the questions we want buyers to ask us too.

The short answer

  • Look at a live system, not a demo. Seeing a system that serves real users today tells you more than ten presentations.
  • Find out who will do the work. The expert in the sales meeting and the team on the project are not always the same people.
  • Agree on measurement first. A good firm proposes measuring the current state before the project starts, and explains how success will be proven.
  • Test their independence. A firm tied to one model or platform, or paid commissions by vendors, may not give neutral advice.
  • Put data and ownership in writing. Data protection, data residency, and who owns the code and prompts must be clear in the contract.
  • Start small. A short discovery phase followed by a pilot with measurable success criteria is a safer start than a long strategy project.

First, do you need an AI consultant at all?

Not every AI need requires a consulting project. Rolling out an enterprise chat assistant, switching on AI features in software you already use, or adopting a single off-the-shelf SaaS product can often be done in-house. In those cases, what you need is less consulting and more good training and a clear usage policy.

Consulting adds value when:

  • You want to automate processes that connect several systems and need human approvals and exception handling.
  • You will work with personal or sensitive data and need an architecture that complies with laws such as GDPR or KVKK.
  • You are starting your first AI programme and cannot decide which use case to start with.
  • You have an internal team, but no experience running AI systems in production.

Types of AI consulting firms

Type of firm Strength What to watch
Large strategy and consulting firms Board access, change management, broad scope High cost; implementation is often handed to other teams
Systems integrators Deep integration with enterprise systems, scale May be tied to specific platforms; AI depth varies by team
Specialist AI firms Current technical depth, fast pilots, production experience Capacity may be limited; check references and the team closely
Software and digital agencies Interface, web, and product development May have added AI recently; ask about evaluation and governance experience
Independent specialists Flexibility, lower cost Continuity and support risk; dependence on one person
Software vendors' professional services The deepest knowledge of their own product They will naturally recommend their own product

Which is right depends on whether your need is strategic or practical. For most mid-sized companies, the need is a team that can turn a strategy into a system running in production.

The 12 questions to ask

Proof

1. Can you show us a system you built that is running in production today? Not a demo or a prototype, but a system real users rely on every day. If possible, ask to speak to that client, and ask how much the system changed after it went live.

2. Who will work on our project? Ask for the names, roles, and experience of the team. Write into the contract how much time the senior expert from the sales process will spend on the project.

Approach

3. How do you decide whether a use case is worth automating? A good answer includes a method that weighs business value, technical feasibility, data readiness, and risk together. A firm that says "yes, AI can do that" to every idea is not prioritising.

4. What will we have at the end of the first 90 days? Ask for concrete outputs rather than a "strategy document": a pilot tested in production, measured results, and a decision point for scaling.

5. How will we measure success? The firm should propose measuring the current state before the project: handling time, error rate, cost, or customer waiting time. Without a baseline, you cannot prove the return on the investment. For a first estimate with your own numbers, try the AI ROI calculator.

Technology

6. Are you tied to a particular model, platform, or vendor? Ask whether the firm has vendor partnerships or commission arrangements. The right model should follow the requirements of the work; sometimes a frontier model is best, and sometimes a smaller model running on your own servers is.

7. How do you test the AI before it goes live? Expect to hear about an evaluation set built from real examples, measurement of wrong or invented answers, and tests against attacks such as prompt injection. "The model is very good, it will be fine" is not a method.

8. What happens after launch? Monitoring, maintenance, model updates, cost tracking, and who responds when something goes wrong must be clear. AI systems still need work after they go live.

Data and compliance

9. How will you handle our data? Ask which models and which countries personal data will go to, how cross-border transfers are handled under the laws that apply to you, and what data processing agreements are in place. A good firm raises these questions before you do. For Türkiye, see our guide to KVKK and AI; for the Gulf, see Saudi and UAE data protection for AI assistants.

10. Who will own the code, prompts, and data at the end? Make sure the source code, prompt templates, evaluation sets, and documentation are handed over to you. If you ever want to change firms, you should not have to rebuild everything.

Team and commercials

11. How will our team learn to run the system? A good engagement makes your own team stronger: handover, documentation, and role-based training should be part of the plan.

12. How is pricing structured, and what is not included? Fixed price or time and materials? Are model usage fees, hosting, licences, and support after launch included? To see the full cost, read our guide to what an AI assistant really costs.

Red flags

  • Percentage savings promised before seeing your data. A firm that says "we will cut your costs by 40%" has not yet measured your process.
  • Only demos. A firm that cannot show any system in production may not yet have met the hard parts of production.
  • The same tool for every problem. A firm that recommends the same platform or model whatever the problem is selling a product, not a solution.
  • No mention of evaluation or governance. If testing, monitoring, and human approval points are never discussed, the risks will surface later.
  • Long strategy projects with no pilot. If months of analysis end in nothing but a presentation, your learning slows down.
  • Unclear ownership. If the contract does not say who owns the code and data, you will be locked in.

How to structure the first engagement

The safest route is to split the work into small, measurable steps:

  1. Discovery (2–4 weeks): a review of the processes, scoring of use cases, a data and compliance assessment, and a pilot proposal. This can be done with a fixed scope and a fixed price.
  2. Pilot (6–12 weeks): one use case tested with real data and real users. Success criteria should be written down before the pilot starts.
  3. Scaling decision: continue, change, or stop, based on the pilot's measured results. A good firm is not afraid to recommend stopping when the results are weak.

To prepare your leadership team, an executive AI workshop can be a good step before discovery.

A scorecard for comparing firms

Copy the table below, score each criterion from 1 to 5, and multiply by the weight. Adjust the weights to your own priorities.

Criterion Weight What you are looking for
Production references 20% Verifiable systems in use today
Team and seniority 15% Experience of the people who will do the work
Method and measurement 15% Prioritisation, a baseline, success criteria
Technical depth and independence 15% Model-agnostic architecture, evaluation, security
Data and compliance 15% Data protection, data residency, contracts
Ownership and handover 10% Code, prompts, and documentation stay with you
Commercial terms 10% Transparent pricing, with exclusions stated clearly

Frequently asked questions

How much does AI consulting cost?

It varies widely with scope, integration depth, and how much preparation your data needs. That is why it makes sense to ask for a short, fixed-price discovery phase as a first step. When comparing proposals, look beyond the development fee to the model, hosting, and maintenance costs after launch.

How long does an AI project take?

For a single use case, discovery and a pilot can usually be completed within a few months. What extends the timeline most is rarely the technology; it is data access, approval processes, and integrations. Ask about these dependencies when you discuss the schedule.

Should we choose a local firm?

It is not essential, but it has advantages. A team that can judge quality in languages such as Turkish and Arabic, knows local data protection rules and business practice, and works in your time zone moves faster. If you have international customers, experience across several languages matters too.

Do we need a consultant if we already use ChatGPT Enterprise or Copilot?

Those tools raise individual productivity, but they do not, on their own, build process automation that connects several systems, includes approval steps, and produces measurable business results. Training and a policy are often enough to roll out the tools; automating processes needs architecture and integration.

Next steps

Ask these twelve questions of every candidate, Orbitra included. We show our work running in production (for example, the 100AI assistant), tell you up front who will do the work, put measurement at the start of the project, and write into the contract that the code, prompts, and documentation are yours. To book a discovery conversation, see our AI automation consulting page; to prepare your team, explore our corporate AI training programmes.

LET'S GET STARTED
PROJECT INTAKE

Let's put your AI automationinto orbit

Tell us the process, customer journey, or team capability you want to improve. We will map the right next step for AI automation consulting, an AI assistant, corporate training, or a measurable pilot.

Email Us[email protected]For inquiries, strategy, and demos
ORBITRA / PROJECT BRIEFSECURE CHANNEL
01
02
03
04