In short: before hiring an artificial intelligence consultant, a small business should get a clear answer to eight questions: how results will be measured, what is not worth automating, what happens if the consultant becomes unavailable, who ends up owning what gets built, what personal data is processed and under which contract, what it costs to keep running, how the team will be trained, and how customers will know they are talking to an AI.
Why I'm writing this
According to the survey on ICT use in companies by Spain's National Statistics Institute (INE), in the first quarter of 2025, 21.1% of Spanish companies with 10 or more employees were using artificial intelligence. Among those with fewer than 10 employees, the figure was 13.4%.
Read the other way round, the figure says something useful: for most small businesses, commissioning an AI project will be a first. And the first time, you have nothing to compare against. You don't know what a normal answer sounds like, what an inflated promise looks like, or which questions someone should have asked you and nobody did.
I have a stake in this, because it's how I earn a living. That is exactly why it feels honest to put in writing the questions I would ask if I were on the other side of the table. Some of them don't make me look good. I'm including them anyway.
1. How will we know it has worked?
It's the question most often skipped and the one that matters most. If nobody measures how long a task takes today, how many times a month it happens and how many errors it produces, in six months nobody will be able to say whether the automation has improved anything. There will be impressions, not results.
A good answer: proposes measuring the starting point before touching anything, even with something as simple as a week of manual logging.
A red flag: specific savings figures before they understand how you work. Any number given before measuring your case is an estimate. That applies to the indicative ranges you see on consultancy websites too, mine included.
2. What would you advise me not to do?
A consultant who knows the field knows where AI doesn't pay off: processes that happen twice a year, tasks where human judgement is the product, or cases where the real problem is organisational and no tool is going to fix it.
A good answer: rules out something you had in mind and explains why.
A red flag: everything you suggest sounds like a good idea to them.
3. What happens if you're not around tomorrow?
This is the one that makes me most uncomfortable, because I work alone. Having a single point of contact has clear advantages, but also an obvious risk: if that person falls ill, changes direction or disappears, your business cannot be left with a system nobody else understands.
A good answer: documentation of what is built, access and accounts in your name, training for someone on your team, and a structure another professional could pick up without starting from scratch.
A red flag: "don't worry about that".
4. Whose name is everything in?
Automation tool accounts, API keys for AI services, domains, databases, code. If they are in the consultant's name, you depend on them for something as basic as cancelling a subscription.
It pays to be just as explicit about ownership of what gets developed. In Spain, the Intellectual Property Act only assigns a program's exploitation rights to the company automatically when it is created by a salaried employee in the course of their duties or following the employer's instructions, unless otherwise agreed (Article 97.4). With an external provider there is no such automatic assignment, so anything not agreed in writing is open to interpretation. Other jurisdictions have their own rules, but the practical advice is the same.
A good answer: distinguishes between what is built for you, which should be yours, and the consultant's own templates or methods that they reuse, and puts it in writing.
A red flag: the question takes them by surprise.
5. What data about my customers will pass through your tools, and where?
If the consultant or their tools will process personal data on behalf of your business, the General Data Protection Regulation requires that relationship to be set out in a data processing agreement (Article 28), with minimum content: subject matter, duration, types of data, an obligation to follow your documented instructions, conditions for subcontracting, and assistance with data subject rights. The Spanish Data Protection Agency (AEPD) publishes guidelines with model clauses for drafting one.
There's a detail that often goes unnoticed. AI model providers and automation platforms may also be involved in the processing. The GDPR requires your prior written authorisation before the processor engages other processors (Article 28(2)), and if the data leaves the European Economic Area, the rules on international transfers come into play.
A good answer: shows you what data will be processed, by which providers and in which region, and proposes the agreement before work starts.
A red flag: "your accountant handles that" or "the tool is already compliant".
6. What does it cost once it's running?
The implementation quote is the visible part. Then come the tool licences, AI model usage (often billed per use) and maintenance whenever a provider changes something, a form is modified or an integration stops responding. An automation is not a piece of furniture you buy once.
A good answer: a written estimate of the recurring cost, separating what you pay third parties from what you pay the consultant.
A red flag: maintenance doesn't appear anywhere.
7. Will my team understand what they're using?
There's a legal obligation here that many small businesses are unaware of. Article 4 of the EU AI Act requires organisations that use AI systems in their professional activity to take AI literacy measures for their staff and for anyone using those systems on their behalf. It has applied since 2 February 2025.
Regulation (EU) 2026/1744, known as the Digital Omnibus and in force since 27 July 2026, reworded that article: it no longer requires ensuring a sufficient level of AI literacy, but taking measures to support it. The obligation has been softened, but it has neither disappeared nor been postponed: it has applied continuously since February 2025. What the Omnibus did postpone are the obligations for high-risk systems, to December 2027 and August 2028, which most small businesses will not come across.
A good answer: training is part of the project, tailored to what each person will use, and documented.
A red flag: "it's very intuitive, there's no need".
8. If something is going to talk to my customers, will it say it's an AI?
Article 50 of the AI Act has applied since 2 August 2026. Among other things, it requires systems that interact directly with people, such as a website chatbot, a WhatsApp assistant or a voice agent, to inform them that they are dealing with an AI, unless this is obvious. The Digital Omnibus did not postpone this obligation. The European Commission published its guidelines on Article 50 on 20 July 2026.
A good answer: shows you at what point and with what wording the customer will be told, and how the conversation is handed over to a person.
A red flag: suggesting the agent should "seem human" as a selling point.
What these questions don't solve
They are not a substitute for legal advice. The points on data protection, intellectual property and the AI Act describe the general framework; how they apply to your particular business is a conversation to have with your lawyer or data protection officer.
Nor do they guarantee a good outcome. What they do is quickly separate the people who have thought about your business from the people who have thought about their tool.
If you ask these questions and get clear answers, you are probably dealing with someone serious. If any of them gets dodged, you've already learnt something important. And yes, you can ask me too.
Rafa Vidal, Isenzo-IA
Sources
- INE (Spain), Survey on ICT and e-commerce use in companies 2024-2025, press release of 22 October 2025 (in Spanish): Source
- Regulation (EU) 2024/1689 (AI Act), Articles 4 and 50: Source
- Regulation (EU) 2026/1744 (Digital Omnibus on AI): Source
- European Commission, Guidelines on transparency obligations for providers and deployers of AI systems, 20 July 2026: Source
- Regulation (EU) 2016/679 (GDPR), Article 28: Source
- AEPD, Guidelines for drafting contracts between controllers and processors (in Spanish): Source
- Spanish Intellectual Property Act (Royal Legislative Decree 1/1996), Article 97 (in Spanish): Source
