These are the criteria we apply to any company. We publish them so that, before you talk to us, you know how we'll think about yours.
Fran Sacares, founderWhy in this order
AI amplifies whatever it finds. On a clear process and software that works, it multiplies what the team can do. On a muddled process, it multiplies the mistakes: faster, at greater scale and harder to see. So the question isn't where to put AI, but what needs fixing first.
Three decisions before automating
When it works.
If a program or a way of working does its job well, it stays as it is. We connect it to everything else so data is entered once and reaches everywhere.
Signs
When it's the problem.
If the process has steps that add nothing or the software forces workarounds, we say so. Redesigning the process or switching tools in time saves more than any automation.
Signs
When nothing fits.
If no software on the market fits the way you work without making you give up what sets you apart, we build it to measure, integrated with everything else.
Signs
Then
With a clear process and the tools in place, we choose the simplest technology that solves each step. Most of the time it's a rule. Sometimes, AI. And an agent only when decisions need chaining together.
Not everything needs AI
Putting a language model where a rule is enough costs more, runs slower and fails more often. We choose the technology based on what the process needs.
Clear rules
Traditional automation. No AI, no agents.
Structured data across several systems
Integration: data travels on its own between your tools.
Documents in different formats
AI to extract the data and validation against your systems.
Language and interpretation
Generative AI with defined criteria and a constrained output.
Several steps, tools and chained decisions
An agent, with permissions, limits and traceability.
A critical decision
AI and rules, with a person's oversight before it runs.
When an agent makes sense
When information first has to be interpreted, systems queried and conditional decisions made, an agent can add value. Always with permissions, limits, traceability and a person to escalate to.
We automate the work. Your team reviews the exceptions.
How we work
We don't install a tool: we design the system around your process.
First we understand the process. The technology is decided afterwards.
You tell us how you work and where your time goes. We tell you where we see room for improvement and whether it makes sense to go further.
We go through your processes with your team: what comes in, who decides, what is done by hand. The result is a prioritized map of opportunities.
What gets automated, what gets integrated, where AI comes in and where a person approves.
We build, connect to your systems and test with real data.
We keep it running, review exceptions and extend it to new processes.
Security and control
Automating means giving access to email, documents and management software. It is designed with the same precautions as any other access to company data.
Own credentials, revocable and separated by integration and environment. Each automation or agent only has the permissions it needs and explicit limits on what it can do.
Data is checked before it is written to your systems, not after. What doesn't fit is set aside with context instead of being forced through.
A record of what the system did, with which data and when, plus alerts when a flow fails or behaves outside what is expected.
On critical steps the system prepares the operation and a person confirms before it runs.
No system is free of failures, and regulatory compliance depends on each case. That's why we design them so that, when something doesn't add up, it is detected and stopped before it spreads.
What you can expect
It's more expensive, slower and fails more. If a step can be solved with rules, it's solved with rules.
The assessment may conclude that the process needs sorting out, the software needs replacing or nothing should be done yet. We write that up like any other conclusion.
We work with n8n, which can be hosted on your own servers and whose flows can be exported, and with documented in-house development. Every part can be reviewed.
Each integration uses its own credentials, separated by environment, with only the permissions it needs, and they can be withdrawn at any time.
In the first meeting, at no cost, we look at one of your processes with these criteria.
Analyze my company