AI for your small business, one outcome at a time

By August 2025, 58% of small businesses in the United States said they used generative AI, more than double the 23% of two years earlier. The figure comes from the U.S. Chamber of Commerce, which has tracked it since 2023. If your team drafts emails or summarizes documents with a chat tool, you are already in that majority.

That kind of AI helps whoever is holding it, and only while they hold it. A subscription answers when somebody asks, and in a team of five nobody’s job is to sit and ask all day.

A system is a different purchase. It watches the channel where the work arrives, does the task, checks its own output and leaves a record you can audit. A system does not wait to be prompted, and it is measured against a number agreed before it was built.

Most AI software for small business is the first kind, and it is fine at what it does. What we build is the second kind, cut to the size of one process and one result.

Each package is scoped to one concrete outcome and quoted at a fixed price before anything starts. These are the four we offer.

Stanton manages rental properties. The electricity, gas and water bills their tenants sent in were typed into a spreadsheet by hand, a minute each. Today a chat the team already used receives them, a reading layer with a language model turns each one into a checked row in that same spreadsheet and 98% of the bills go through with nobody touching them. The client has kept commissioning more processes since.

Barceloneta Premium rents out apartments in Barcelona and receives dozens of WhatsApp inquiries a day. Checking each one used to take five to ten minutes of somebody’s attention. Now an agent holds the conversation, gathers what the agency needs to know and hands the team an email summary with the reasons already written. The agency got more than three hours a day back.

The third system is our own product. Wazzy books, reschedules and cancels appointments on WhatsApp for clinics, and it handles hundreds of appointments a month. In the entire history of the product, no appointment has ever been double-booked. We run it ourselves, with our own money on the line, which is where the maintenance discipline this page describes was learned.

Some projects are born from a pain point and some are born from a demo. We wrote about the second kind and called it the shiny toy, the project nobody misses when it is not there, whose gain never had a number and where the AI sits in the headline instead of in the result.

There is also the case where the right answer is simpler than an agent. If your process follows clear rules and nothing in it needs interpreting, conventional automation without a model does the job for less money and with no model to keep an eye on. We build that too, and we say so when it is the better buy.

A fixed-scope build falls between €2,500 and €10,000, and keeping it watched and maintained costs between €150 and €500 a month. They are the same ranges we publish for every project, broken down in the pricing guide.

What the model consumes and what the infrastructure costs are not inside that fee. Both run on accounts opened in your company’s name, so those bills arrive at your desk, visible line by line, and never pass through us.

The fee has no lock-in. You pay for the month you get and stop when you decide, and what was built does not switch off that day, because the repository and the infrastructure were yours from the first day.

With one process and one number. Before anything is built we agree what the system should save and how we will measure it, so the decision to extend rests on evidence instead of on enthusiasm.

The first step is a pilot, paid and small, and its price counts toward the final project if you go ahead. With the data, the access and the task well defined, two weeks for that pilot is realistic. When one of the three is missing, we say which one and what it takes to get there, instead of promising a date we do not control.