Five Questions Before Buying Any AI Tool: How SMBs Avoid Licenses Gathering Dust

The past twelve months have created an industry that floods SMBs with AI tools. Every vendor promises "80 % less manual work". The reality in most mid-sized companies: three to five licenses sitting unused, because either the prerequisites weren't there or the use case was unclear.

The problem is not the market — the problem is the order of questions. Whoever buys a tool before the process is clear almost always buys the wrong one.

These five questions help avoid that.

Question 1: Which specific process should change?

The most obvious, most often skipped question. "We want to use AI" is not a goal. "We want to automate our monthly sales reports" is.

Concretely:

Which manual step today should go away?
Who does it today, how long, how often?
What output is produced — and who reads it?

If those three questions have no clear answer, the tool purchase is premature. Two weeks of process observation is then the better investment than any new license.

Question 2: Which data does the tool need — and is it available?

The second usual loser-question. AI tools are only as good as their input data. Buying an AI tool for CRM workflows without clean CRM data buys an expensive script with no effect.

Practical self-check:

Do the input data exist in structured form?
Is data access (API, export, webhook) technically usable?
Who maintains the data? Who is responsible when it breaks?
Are there data hygiene standards that don't depend on a single person?

If more than one of these is open, every euro spent on the AI tool drains through a sieve.

Question 3: Who in the team carries implementation and maintenance?

Tools need maintenance. Every integration eventually breaks via an API update, UI update, or changed auth flow. AI tools with model dependence carry extra risk: models get sunset, retrained, or change their behaviour.

Concretely:

Who is internally responsible for keeping the tool productive?
Does that person have the technical prerequisites?
How much time per month is realistically allocated to maintenance?
What happens during a longer absence of that person?

If no maintenance contact can be named, couple the tool purchase to a service provider who takes maintenance with them. Otherwise the tool gathers dust within six months.

Question 4: What is the concrete measurable output — and who checks it?

AI tools are not deterministic machines. They produce output with variation. Without a clear validation concept, you get either blind trust or constant skepticism. Both are bad for adoption.

What must be clarified:

Which concrete KPI measures the tool's success?
Who checks at what rhythm whether the output is right?
What happens on deviation — who is informed, what is adjusted?
Where is the human in the loop, and where deliberately not?

Example: an AI-supported reporting tool can create reports automatically. But the first version always goes through a human before reaching the MD. That's not distrust — that's clean quality assurance.

Question 5: What if the vendor disappears tomorrow?

This question is systematically skipped — and the most expensive when ignored. AI tool startups vanish on a yearly rhythm. Larger vendors pivot or restructure pricing. Models change behaviour or get sunset.

What must be checked:

Which data stays inside the tool? Which can be exported?
Is there a data portability clause in the contract?
How mission-critical is the tool? What happens during a 4-week outage?
Are there market alternatives — and how costly would a switch be?

For mission-critical tools: never buy without a data-export option. If the data is trapped inside the tool, a switch later is a complete rebuild, not a tool swap.

What these five questions usually surface

In nine out of ten cases, the five questions reveal: the tool isn't the problem. The problem is one of three underlying structures:

The process is not cleanly described. Then no tool helps.
The data is not in usable form. Then no tool helps.
There is no owner for implementation and maintenance. Then no tool helps.

Whoever clarifies one of those three foundations before the tool buys better. Whoever ignores them buys more often — and the money lands on licenses nobody opens.

Common follow-up questions

Which AI tool should I pick now?

The right answer: it depends on the answers to the five questions. For reporting automation, n8n + a language model API is often enough. For more complex workflows, Make or a small custom app. "Which tool" is always the second question. The first is "which process".

Shouldn't we just unlock ChatGPT for everyone?

Careful. ChatGPT licenses for everyone without an application context are exactly the license-gathering-dust phenomenon from the intro. Better: define a clear pilot use case, run it productively with 3–5 people, check success criteria, then expand.

What's a realistic first AI tool cost in an SMB?

Highly variable. For a single productive workflow automation (e.g. n8n + OpenAI for reporting) realistically 2,500–6,000 € one-off and 80–200 € / month running. Licenses for standard AI suites (Microsoft 365 Copilot, ChatGPT Enterprise) rather 20–60 € / user / month. Custom solutions more.

What about data protection?

GDPR-compliant use is possible when vendors offer clean DPA / SCCs and processing is limited to concrete purposes. Caution with vendors that don't document data flows clearly — and with "training" clauses that use your data for model improvement.

How Motainment supports this

At Motainment, AI tool selection is never a standalone topic. It is embedded in a workshop where we first look at processes and data flows — and only then recommend concrete tools and workflows. That avoids the typical license-dust trap.

We bring our own standard building blocks (n8n, OpenAI, Anthropic, Pipedrive, Supabase) and complement them with tools already living in the client stack. No ideological tool religion — pragmatic selection per use case.

Workshop day rate: 1,200 € net. Output: a concrete concept document with prioritized levers, effort estimates, and tool recommendations. You can use it without us afterwards if you like.

What to check now

Write down the three AI tool ideas you've mentioned most often.
Answer the five questions above for each idea.
If even one idea has more than one open question, the tool buy is too early.
Instead, evaluate which of the three foundations (process, data, owner) needs clarification first.

If the evaluation stays ambiguous: an intro call clarifies in 30 minutes whether the next step is a workshop or a concrete pilot workflow.