What an AI starter kit for a small business should include, and what is not a kit

Rounds of AI starter kits for small businesses keep opening, and with them comes a question that reaches our WhatsApp almost verbatim: “what does one of those kits actually include?” It is a fair question, because the word “kit” means nothing on its own.

The short answer: a kit is not a bundle of tools. It is a closed scope.

Why the distinction matters

An MIT study that circulated widely in 2025 reviewed more than three hundred enterprise deployments of generative AI and found that the vast majority had produced no measurable return. The finding that got quoted most was not about the models but about how they get adopted: the projects were not failing for lack of technology.

We see the small-business version of that every week. A company buys a platform, switches it on, and three months later nobody can say what is running. It is not that the tool did not work. It is that there was never a sentence saying “when this happens, we are done.”

A kit is exactly that sentence, written before you start.

The four things a kit does include

1. One channel, not all of them

The channel is wherever people already write to you. If ninety percent of your conversations arrive on WhatsApp, the kit is a WhatsApp kit. Adding Instagram, email and web chat in month one multiplies what has to be tested and pushes back the day anything genuinely works.

And watch one detail that reorders the whole project: for a system to read and answer WhatsApp, the line has to be connected through the official API, not the phone app. That paperwork is the one step you cannot speed up by working harder, so it goes first.

2. One job that gets finished end to end

Not “customer support.” That is not a job, it is a department.

A job is this: someone asks about availability, the Met checks the calendar, offers the slots that are genuinely open, books the one the person picks and schedules the day-before reminder. It starts with a message and ends with a record that exists in a system.

That shape is what makes it reviewable. If nothing at the end of the job can be opened in another system, the job did not finish. It chatted.

3. The systems the answers come from

A Met with nothing connected answers with whatever you wrote into it and nothing more. That covers opening hours and return policies, and goes quiet at the exact moment the answer depends on live data: what is in stock in a medium, what it costs to ship three kilos, who owes money and since when.

The order we recommend never changes:

1. The CRM first. It is native and needs nobody’s credentials, so it connects on day one. Everyone who writes becomes a contact with their history, and without that there is no follow-up later.
2. Whatever answers your most frequent question. If you sell product, inventory. If you sell time, the calendar. If you collect, receivables.
3. Everything else, once the first one has run for weeks without anyone pushing it.

4. A way to know whether it worked

This is the one almost no kit carries, and it is the one that decides whether the rest was worth anything.

You do not need a dashboard. You need three numbers you can review in fifteen minutes: how many conversations closed without a person stepping in, how many ended without a useful answer, and how many answers sounded right and were wrong.

The third one does not come from any dashboard. You have to read conversations by hand, and that is where what the Met is missing is written down.

What is not a kit

A license with training. Access to a platform and two onboarding sessions are not a scope. All of that can stay switched on for months without a single job getting finished.

A list of available integrations. Thirty connectors existing tells you nothing about your case. What matters is which ones get connected in your kit and which questions they will be able to answer. And it is worth looking at what each one does not do: querying an ERP is not modifying it, and recording a sales invoice is not filing it with a tax authority.

A pilot with no end date. A pilot that never finishes never fails, which is also why it never teaches anything. If the kit does not say when it is declared done, there is no kit.

How to ask for one, in three questions

When someone offers you an AI kit, these three tell you almost everything:

1. Which job will be running on its own? If the answer is a category (“support,” “sales”) instead of a procedure, there is no scope yet.
2. Which systems will it connect to, and which concrete tools will it be able to call in each one? Names either exist or they do not. A list of names is verifiable; “it integrates with your CRM” is not.
3. What are we going to look at to say it worked, and when?

If all three have written answers, the kit is a kit. If not, it is a license with a different name.

Where we start

Almost always with booking or quoting, and not by accident: those are the two jobs where the result shows up the same week and where the system the answer comes from is usually already in order.

Both are published step by step, with the tool that runs at each one, in the automations. If your case looks like none of them, that is the best moment to write to us: the conversation starts with which questions reach you, not with which plan we sell you.