AI process automation for the work your team repeats every day
We build WhatsApp AI assistants, but WhatsApp is only one channel. The same approach works for email arriving in several inboxes, for documents someone checks one by one, and for the internal tools your team fakes with spreadsheets.
The system reads, sorts and prepares the work. The decision stays with a person on your team, and your rules and documents stay yours.
Do you only do WhatsApp? No.
We lead with WhatsApp because that is where most customers already are, and because the problem shows up fast there: the same questions, at any hour, answered in between other tasks. Our WhatsApp AI assistant for customer service has been live at Fixlab since June 2026.
But what makes that assistant work is not WhatsApp. It is writing the business rules down, testing it against real cases before launch, and deciding what always stays with a person. The same applies to email, documents and the internal work your team repeats every week.
Which processes we automate with AI
Four kinds of work, with what the system does and what stays with your team.
Email across several inboxes: sort it and draft the reply
- The problem
- Every morning someone goes through what arrived in several inboxes and decides what is a new request, what is a question about a file in progress, and what can wait.
- What the system does
- Reads each email, sorts it by your rules, routes it to the right person and drafts a reply with the details of the case.
- What a person decides
- Approves or corrects the draft before it goes out, and keeps anything that fits no rule.
Document and case-file review
- The problem
- Your expert checks the same points on every file and chases what is missing in several rounds.
- What the system does
- Checks each document and photo against your checklist and gathers everything missing into one request. We explain it in AI document review.
- What a person decides
- The expert judgement, the exceptions and the signature.
Internal assistants for your team
- The problem
- Procedures and internal rules live in documents nobody can find, and the questions always end up with the same person.
- What the system does
- Answers your team from your written procedures, inside the tools they already use.
- What a person decides
- Whoever owns the procedures decides what goes into the knowledge base and updates it when they change.
Custom internal tools
- The problem
- A spreadsheet or a manual process that has outgrown itself.
- What we have built
- A mortgage pre-qualification tool built to a mortgage advisory firm's criteria: prospects assess themselves and reach an adviser as an already-scored lead. And our own outbound prospecting system, which went from zero to finding 600+ companies and sending 135 personalised emails in under two weeks.
- What a person decides
- In prospecting, a person reviews and approves the proposed companies before anyone is contacted. In the mortgage tool, the decision stays with the adviser.
Fixlab's assistant has been in production since June 2026.
Fixlab is a vehicle certification lab in Spain. Its assistant answers from a real knowledge base, gathers the paperwork, opens the request in Fixlab's platform and hands the case to a person when judgement is needed. From 26 June to 24 September 2026 it handled 279 conversations from 218 people.
AI agents or automation? What you actually need
Search for AI agents for business and you will find teams of autonomous agents that talk to each other and split the work. It looks good in a demo. In a small business it is almost never what is needed.
Most processes worth automating are a few well-defined steps: read what comes in, compare it with a written rule, prepare the result and let a person approve it. That is software with an AI model in the steps where text, photos or voice have to be understood, not a network of agents deciding on their own. It is easier to test, cheaper to maintain and, above all, you know what it did and why.
A useful AI agent is exactly that: a program that uses a model to decide the next step within limits you set. More in what an AI agent is and why to think software.
What is worth automating (and what is not)
Most of what is possible with AI is not worth doing. Before we build anything we look at four things:
- Volume. It happens many times a week, not twice a year.
- A written rule, or one that can be written. If two people on your team would do it differently, first agree how it is done.
- An owner. One person who decides what correct looks like and will review what the system prepares.
- A cost of error you can measure. Knowing what happens when it gets something wrong tells you how much human review it needs.
What we do not automate: an expert's final decision, the rare cases that come up once a year, and anything where a mistake cannot be undone. To check your own project, the AI Project Scorecard scores it in two minutes, and is an AI project worth building? explains the five questions behind it. To put numbers on it, see the ROI of AI workflow automation.
How we work
Your rules, written down
We sit with whoever does the work today and write down how they do it: what they look at, what they decide and when they pass it to someone else.
A test built from real cases
We collect real cases from your day-to-day and agree with you, in writing, what counts as a correct result.
A small pilot
The system works on real cases and a person reviews and approves every result before it takes effect.
Production and measurement
Once it passes the test, it goes live. We measure how much work it removes and where it fails, and tune it.
That is the difference between a demo and a system your team can rely on; we explain it in AI proof of concept vs production. At Fixlab, the assistant passed a test of 50 real scenarios with zero critical failures before it launched.
It connects to what you already use
We do not ask you to change tools. The first thing we ask is which software you use and what it allows: an API, database access or, at the very least, exports. We start by reading, and only write back into your system once that has been tested.
If your software can be connected, the system leaves the result there; if not, it reaches you by email or the channel you already use, with the data and documents in order. At Fixlab, the assistant opens requests directly in their platform through its API: 20 reform requests created end to end between 26 June and 24 September 2026.
What is yours when we finish
The system is built on your criteria and your documents, which are yours. We do not share your rules or data with other clients: what we reuse across projects is the underlying platform, which is why a second project costs less than the first.
At handover we train your team and document everything. The documentation is fully yours, so your team can change the rules without us.