Your AI isn't failing on technology.
It's failing on operations.
Pilots get demoed, applauded and quietly switched off. What survives is the operation underneath them: data people can actually reach, processes with an owner, a team that can maintain it. That's what I build — and what I leave running when I go.
If three of these sound familiar, you don't have a model problem.
You have an operations problem. And no technology vendor is going to tell you that, because it isn't what they sell.
The pilot shone, then went dark
It worked in the demo, the steering committee loved it, and it never reached production. Nobody can say exactly why.
The data exists, but nobody can reach it
It lives in three systems, in three formats, with no owner. Every query turns into a project of its own.
Everything depends on the consultant
When the vendor leaves, the knowledge leaves with them. What was built freezes at the version of their last day.
Nobody signs the decision
Five departments have opinions, none has authority. The project moves at the speed of the slowest meeting.
The team quietly avoids it
The tool exists and people still use the spreadsheet. Adoption was assumed rather than designed.
You don't know what it saved you
Plenty of enthusiasm, no baseline. Without a measurement taken beforehand, any result is a matter of opinion.
Six levers. Pulled in sequence, not in parallel.
This isn't a service catalogue. It's the order in which an operation becomes capable of sustaining automation without outside dependency.
Execution diagnostic
Two weeks inside your real operation: where time is lost, which data actually exists, and which decision has no owner. The output is a map, not a slide deck.
Data your systems can use
Not a three-year data lake. The minimum layer that gives your critical processes clean, reachable data with a named owner.
Process automation (BPM/RPA)
Flows that run themselves under auditable rules. Pega, RPA and orchestration where they earn their keep; a macro where a macro is enough.
AI applied with judgement
Models and agents only where the return is demonstrable and the risk is containable. Point AI at a process that isn't under control and it amplifies the mess.
Governance and compliance
Who decides, who audits, what gets logged. Traceability from day one — not a layer bolted on when the regulator turns up.
Capability transfer
The lever almost nobody sells: leaving your team able to run it without me. Living documentation, training, and a stretch where they drive and I correct.
Four phases. The last one is leaving.
An engagement done properly ends with your team doing it without me. If you still depend on the consultant six months later, the project failed even if the technology works.
X-ray
I work inside the operation, not in the boardroom. Short interviews, observing the real process, and a review of what you already tried.
Weeks 1–2Order of attack
We prioritise by return and by risk of failure. We pick a case that can be won quickly and that clears the path for the ones after it.
Week 3Build in production
Built against the real process, with real users, measured against the baseline we set before starting.
Weeks 4–14Clean exit
Your team drives, I correct. Living documentation, named owners, and a measurement mechanism still alive after I close the door.
Final 4 weeksExecution Index simulator
Six axes, ninety seconds, one verdict. You see the full result without leaving your email — because if the diagnostic isn't worth anything on its own, it isn't worth your data.
Would your operation survive the automation you're about to buy?
Six questions on the six axes where AI projects die. At the end you get a radar of your operation, the band you fall into, and the exact axis to start from.
Who's behind this and why it matters
An operator, not a deck salesman.
I started my first business at 17. Since then I've built operations rather than presentations — the difference between someone who explains how it should work and someone who has already had to make it work with a real team, a real budget and a real deadline.
I built the LATAM operation of RulesCube — an official Pega partner consultancy — from zero to $4.5M. Before that, I took a hotel in Iceland from €150K to €3M. Neither was done with brilliant technology: both were done by ordering processes, naming owners and measuring what nobody had been measuring.
Background in Computer Science, with a heavily mathematical curriculum. I work across Barcelona, Lima and Miami in Spanish, English and Icelandic. Which means I can sit with your steering committee and your engineers on the same day, with no translator in between.
built from zero
Iceland asset
in parallel
(+FR/IT/PT)
Field notes: four declassified case files
Real engagements, anonymised under confidentiality agreements. No client names and no invented percentages: just what I found and what was done.
Automation nobody was actually using
The scope that grew without anyone saying so
A regional practice built from nothing
A €150K asset that ended up worth €3M
What people ask before signing
I lead the engagement personally and assemble the team the work requires: Pega, RPA or data specialists I've already led before. I don't subcontract accountability — your point of contact is the same person from start to finish.
The two-week diagnostic has a fixed price and is credited in full if we continue into execution. The first conversation costs nothing and is usually enough to tell whether it's worth continuing.
Better. I'm not there to replace them: I'm there to make what they're building land in your operation. On several engagements my job has been precisely to get the incumbent vendor to deliver what they promised.
There's no pyramid here. You aren't sold a partner and then delivered someone six months into their career. And the stated goal of the engagement is that you stop needing me — something a billable-hours model can't afford to promise.
Both. The x-ray phase is on site whenever possible — some things are only visible from inside. The rest runs remotely with presence at the milestones. I operate across Barcelona, Lima and Miami.
I'll tell you, and I'll put it in writing. It has happened. In more than one case what was needed was to fix a process and name an owner, not to buy technology. Charging for a deployment you didn't need costs far more in the long run — for you and for me.
Open a channel
One thirty-minute conversation and you'll know.
Tell me what you tried, where it stalled and what you have to solve this year. If I'm not the right person I'll say so on that same call, and I'll probably tell you who is.