An online shop owner talks to three consultants. The first says, “Get a chatbot, everyone has one.” The second quotes a price over the phone with no detail. The third first asks, “How many messages do you get a day, and which ones eat your time?” The right choice is usually the third. We offer this service ourselves, so apply the same criteria to us.
What makes a good AI consultant?
You can check the seven criteria below in one or two meetings. You don’t need to be technical; you only need clear, written answers.
| Criterion | What to ask | A good answer |
|---|---|---|
| Pain first, tools second | “Which of our tasks do you think is worth automating?” | They ask about your daily work, how often things repeat and what they cost |
| Real work | “Can I try one of your projects myself?” | A link to an app or a sample that works right now |
| Written proposal | “Will you put the price, timeline and acceptance criteria in writing?” | A written proposal with a fixed price and acceptance criteria |
| Live sample before payment | “What will I see before I pay?” | A working sample, at least for the first part |
| Data protection | “Where does our customers’ data go?” | Only the data needed, limited access, and running on your own server if necessary |
| Ownership | “Whose name are the code and accounts in?” | After payment, the code, domain and accounts are in your name |
| Honesty about limits | “Where is AI not a good fit for us?” | A clear example of where they would not recommend it |
- List your pains
Before the meeting, write down three repetitive, costly tasks.
- Short meetings
Talk to two or three consultants and ask the questions in the table.
- Written proposal
Ask for price, timeline, deliverables and acceptance criteria in writing.
- Small pilot
Start with a limited sample project.
- Measure
After a month, check the result against the agreed measure.
Which signs mean you should look for another consultant?
Take these five signs seriously:
- Promising a guaranteed result or a big number before seeing your work, such as “your sales will double”.
- Recommending a particular tool or subscription before hearing the problem.
- A vague or verbal price, with no clear criteria for delivery.
- Asking for full access to all your data and accounts from day one.
- No work at all that you can try for yourself.
Anthropic’s guide to building AI systems also says to find the simplest solution that works and add complexity only when you need it. So if a consultant proposes a big, expensive system on day one, they should explain why clearly.
What should you ask about data and security?
If AI is going to handle customer messages or information, ask these questions. NIST’s AI Risk Management Framework also treats identifying and managing risk as part of every AI project.
- Which data is needed and which isn’t? A good answer: only what is needed.
- Where is the data stored, and who can access it?
- Is the system tested against prompt injection (a message that tries to trick the AI)? OWASP lists it among the main risks for applications built on language models.
- If the work ends, how are the data and accounts handed over or deleted?
For a fuller list, see the responsible AI checklist for business.
How do you judge the cost of AI consulting?
The price depends on the scope, so a fixed price quoted before anyone has heard your needs is not a good sign. Instead of comparing numbers, compare what you get, when, against which success measure, and whether you see a sample before paying. To estimate build costs, see the chatbot and AI model cost guide.
What should the contract and handover include?
- The scope and deliverables, with clear acceptance criteria
- A timeline with staged delivery
- A fixed price and payment terms, ideally after you see each stage
- Ownership of the code, domain and accounts
- The support period and how the engagement ends
Why does seeing a sample before paying matter so much? Read live sample first, then payment. If you are not sure where to begin, AI for Iranian businesses: where should you start? will help.
What should you do in the first meeting with a consultant?
- Prepare your three main pains with their real symptoms: how often they happen and how long they take.
- Ask the seven questions in the table above and write down the answers.
- Ask to try one of the consultant’s previous projects yourself.
- Ask for a written proposal with a fixed price and acceptance criteria.
- Start with a small pilot. OpenAI’s guide to building agents also recommends starting with a simple use case and planning for human intervention on high-risk actions.
Frequently asked questions
How is an AI consultant different from a developer?
A consultant first finds the pains and priorities and tells you which work is worth building; a developer builds it. In a small business it is often best when one person or a small team does both, under one written proposal.
How can I tell a consultant’s work is real?
Ask to try it yourself: an app link in a store, a demo version or a live sample. Screenshots and claims are not enough.
Is a fixed price before the first meeting a good sign?
Usually not. A sound price comes after the need is clear, and it should be written down with defined deliverables.
Does the consultant need access to our data?
Only as much as the work requires, with limited and logged access. It is best to start with a small pilot and little data.