Independent advisor · payments · data · AI
Payments, data & AI — delivered hands-on.
I turn payment data into decisions, operations into automation, and AI ambition into shipped tools — independent, senior-only, fixed scope.
“Are we overpaying our payment provider?” · “Why is my checkout approval stuck?” · “Where would AI actually pay off?”
Not sure yet? See the proof first.
The mission
"I turn data into decisions, operations into automation, and AI ambition into shipped tools — working alongside your teams to give your customers the product they deserve and your finance team the costs they dream of."
01 — three ways I pay for myself
Three value streams. One pair of hands.
Most engagements start in one stream and end up using two. Pick the one that hurts most today.
Stream № 1 — payment optimisation
Pay less per transaction. Turn away fewer customers.
Effective cost per transaction against quoted rates · checkout approval & involuntary churn · settlement and reconciliation that runs without humans.
Where payment stacks leak →
Stream № 2 — applied AI
An AI advisor who ships, not one who slides.
Honest AI strategy, tools in production, each measured against a number we agree on — the pattern I shipped to 1,000+ merchants.
What an AI advisor actually does →
Stream № 3 — data & visualisation
Data people act on. Dashboards with a Monday.
Platforms that fit your budget — 20TB and 10-billion-row tables under a strict cost envelope — and KPIs tied to decisions, not decoration.
See the platform case →02 — stream № 1 · where it leaks
Payment stacks leak money in quiet places.
Technology has never been more powerful — yet customers are more let down than ever. The businesses I work with are reversing that, because customer-driven operations win twice: more revenue, and often less cost.
Not in one dramatic failure — in dozens of small ones nobody owns. Two places the money goes — and one habit that makes both worse:
What each transaction costs you
Your reports show volume and fees. They almost never show the gap between quoted and effective rates — the drift hiding in scheme fees, cross-border structures, routing and FX. Since interchange got capped in Europe, that's where the real levers moved. Most teams haven't moved with them.
Fill in the numbers — then add your email and I'll send the verdict.
This opens a pre-filled draft in your own mail app — nothing is sent from this site and nothing is stored. I reply with your verdict, usually within one business day. Directional benchmarks (consumer cards); commercial cards differ.
The customers your checkout turns away
A chunk of legitimately good orders get declined, most soft declines are recoverable, and the churn they cause gets mislabeled as a marketing problem. Nobody owns the number, so nobody fixes it. Involuntary churn is the retention problem hiding inside payments.
Fill in the numbers — then add your email and I'll send the estimate.
This opens a pre-filled draft in your own mail app — nothing is sent from this site and nothing is stored. I reply with your estimate, usually within one business day. Industry checkout abandonment averages ~70% (Baymard); the payment-caused share is estimated at a conservative 10–20% of drop-offs.
Settlement and reconciliation that still touch humans
Every settlement step that needs a person costs money, adds risk, and quietly rehearses your next audit finding. Meanwhile the data you'd need to see all of this clearly is already in your systems — it's just spread across extracts, spreadsheets and tribal knowledge.
The fix starts with a number. Once you can see the leak, the project stops being a belief and becomes a business case.
03 — stream № 2 · applied AI
I don't sell AI strategy. I help you ship it.
"AI strategy" has become the most expensive deck in the building — and six months later, nothing in the company works differently. The trade I do instead starts with three uncomfortable questions: which decision does this tool change? Who uses it on Monday? What number proves it worked?
One workflow that eats real hours. One tool scoped in weeks, deployed in your own environment. A baseline we agree at kickoff. And because nobody pays me commissions, I'll also tell you which project you should not build.
Plain-language answers over your own data
Your merchants and teams ask in their own words and get their own numbers — no analyst, no SQL, no queue. The pattern is proven in production: a natural-language SQL assistant shipped to 1,000+ merchants.
An AI data-driven organization
Employees become independent: AI-assisted self-service lets them write their own data requests instead of waiting days for someone else. Decisions speed up — and the data team gets its calendar back.
Internal tools that eat the busywork
Ops screeners, triage automations and follow-up agents on the queues that eat your team's week. Small tools, in your environment — each measured against the hours it returns.
The test for anyone selling you AI, including me: ask what ships in ninety days, and for whom. Vague answers are the deliverable. Mine isn't — read the full note →
04 — stream № 3 · data & visualisation
Your data exists. The access and the interface usually don't.
Decisions made on incomplete extracts. Analysts as gatekeepers. Merchants blind to their own numbers. I build the chain end to end — platform, self-service, dashboards — so the people who need the data can actually use it.
Big data shouldn't mean big bills.
It's not because your data is big that it should be expensive to use. I architect scalable platforms that keep full-depth access at competitive cost — inside the cost envelope your CFO actually set.
In practice: 20TB+ with 10-billion-row fact tables — 160+ internal users in self-service.full case →
A data strategy that makes the organization smarter.
Not a slide deck — a working one: the data that matters, the KPIs that drive which decisions, and the self-service that puts them in every team's hands. Your people stop debating the numbers and start acting on them.
Dashboards with a Monday.
Monitoring and reporting where every chart has a decision and a person attached. Reports nobody acts on are tracking wearing a suit.
In practice: a top European football club — near-real-time sales per stand, steering match-day crowds.full case →That chain — platform, product, self-service — is what I build under real budget constraints. See the full platform case →
05 — proof, not promises
Done it from the engine room.
Not as a consultant studying payment companies — as the operator who ran the systems. Three stories, anonymized because the numbers carry the weight anyway.
06 — how we start
A number first. Then the fix.
Senior-only, fixed-scope — no juniors learning on your budget. The first step is deliberately small.
The payment performance & data diagnostic
Two to four weeks, fixed fee. I quantify what your stack is leaking — effective cost per transaction against benchmarks, approval and decline analysis, settlement flow review, data & AI readiness. You get a number you can take to the board, and a 90-day roadmap.
fixed fee · credited against whatever we build next
Build & fix programmes
Approval-rate uplift, payment cost reduction, settlement and pricing engine work, data platforms that fit your budget. Fixed-scope engagements of four to twelve weeks — designed by someone who has operated these systems, not studied them.
AI that actually ships
One tool, weeks to production, measured against a number we agree on. If it doesn't reach your team, it doesn't count. Concretely:
- Natural-language assistants over your own data — the pattern I shipped to 1,000+ merchants.
- Alerting & monitoring that catches the anomaly before Monday's meeting does.
- MCP tooling on the APIs and systems you already run — AI that acts where the work happens.
- Ops screeners & automations — triage and follow-ups on the queues that eat your team's week.
Scoped small, deployed in your environment. Your data stays yours.
Ongoing advisory
A few days a month: roadmap, delivery oversight, vendor decisions, regulator-facing data work. The product here is continuity — an advisor who stays until the value is real, then keeps it real.
Most client relationships start at i. — and the diagnostic usually pays for itself in what it finds. If it finds nothing, you've bought a benchmark cheaply.
The 25-minute call
You describe the stack. I'll tell you what I'd measure first.
- No deck, no pitch — your stack, your numbers, your doubts.
- You leave with the two or three numbers worth pulling first.
- If there's nothing to find, I'll say so — you keep the benchmark.
Bring a question like: are we paying our payment provider too much? Is our checkout approval rate where it should be? Where would AI actually earn its keep? And is this vendor proposal any good?
07 — how I work
I stay until the value is proven.
Alongside, not above
I work inside your teams — engineering, product, finance, risk. No report-and-vanish. The fix gets designed where the work happens, in your systems' language.
Value proven, not promised
At kickoff we agree the metric and the baseline. At the end we measure. No fixed promises on proposal slides — demonstrated results against a number we both signed.
Capability that stays
Documentation, tooling, and people who understand what was built and why. The improvement should outlast the engagement — a fix that needs me forever isn't a fix.
One thing sits behind all three: independence. No vendor commissions, no partner targets, no internal politics to serve. A fresh view is only worth paying for if it's conflict-free.
08 — what you won't get from me
Six things I turn down.
What an advisor refuses says more than what they offer. These are standing decisions, not negotiation openers.
Analytics that change no decision
If we can't name the decision new data would change, the tracking is decoration. Decoration is someone else's billable hour.
Cost cuts paid for by the customer
The goal is more revenue and less cost. If a saving exists only because customers get a worse deal, I'll say so — and walk.
Projects that turn three systems into four
Complexity is the tax small companies can least afford. New tooling has to retire more than it adds, or it's a no — even when it ships.
Engagements with no artifact and no metric
Every mandate ships something and proves something. A project that ends in a deck ended in the wrong place.
"AI strategy" with nothing shipping
I'll help you pick the tool, scope it honestly, and put it in production. What I won't do is produce forty slides of AI futures.
Dashboards without a Monday
Every chart earns its place with a person and a decision attached. Reports nobody acts on are tracking wearing a suit.
Sometimes the right advice is "don't build this." You'll get that advice too — it's cheaper than the project.
09 — about
Eight years in the engine room.
I'm Thibault Mambour — Finnura is my one-person practice. Eight years in banking and payments, most recently leading data, analytics and settlement products at a Swiss payment provider. There: scaled processing 3×, took settlement to 99.99% automation, built the analytics platform behind it all, shipped a natural-language SQL assistant to production — and sat across from a national regulator during a settlement-stack audit.
Before that, a Belgian bank and a Big Four consultancy, where I learned why enterprise data is hard before learning how to make it easy.
What I'm actually good at is bridging rooms. Engineering, product, finance, risk, regulators — each speaks its own language, and the expensive failures happen in translation. I speak all of them well enough to be dangerous, and translate both ways.
One conviction underneath it all: technology has never been more powerful, yet customers are more let down than ever. I work with the businesses that want to reverse that — because customer-driven operations are the rare kind where revenue goes up and cost comes down.
- Focus
- Payments · data platforms · applied AI
- Clients
- PSPs, PayFacs & online merchants
- Markets
- Switzerland, Benelux, France — remote-first, on site when the assignment needs it
- Base
- Portugal — travelling across Europe when it helps the work
- Languages
- EN & FR fluent · PT & DE working on it
- Mode
- Senior-only · fixed-scope · hands-on
10 — still here?
Skip the form. Buy me breakfast.
The Sparring Breakfast — Porto or Minho
30 to 90 minutes, in person, your agenda. Bring a pricing doubt, a checkout that converts too low, an AI idea that needs a sanity check, a vendor proposal you're unsure about, a dashboard nobody acts on — or a leak you can smell but can't name. I bring the questions — and, when it's useful, the numbers to chase next. No deck, no pitch, no obligation.
11 — questions, answered
The short answers.
What merchants usually ask before the first call.
What does Finnura do?
Finnura is the independent payments data & AI practice of Thibault Mambour. It works across three streams: payment optimisation (cost per transaction, checkout approval, settlement automation), applied AI shipped to production, and data & visualisation — platforms and dashboards people actually use.
Who is behind Finnura?
Thibault Mambour — eight years in banking and payments, most recently leading data, analytics and settlement products at a European payment provider. Track record: 3× processing volume on the same team, 99.99% settlement automation, a 20TB platform with 10-billion-row tables, a natural-language SQL assistant used by 1,000+ merchants, and a national regulator settlement audit passed.
Do you sell AI strategies?
No. AI strategy decks are the most expensive deliverable in the building. Finnura ships instead: one workflow that eats real hours, one tool scoped in weeks, deployed in your own environment, measured against a baseline agreed at kickoff — with a clear opinion on which AI project you should not build.
How does an engagement start?
With a 25-minute call: you describe the stack, Finnura tells you what to measure first. The usual next step is the payment performance & data diagnostic — two to four weeks, fixed fee, credited against whatever gets built next.
Where is Finnura based, and what languages?
Based in Portugal, working remote-first across Europe — Switzerland, Benelux and France in particular — with on-site possibilities when the assignment needs it. Working languages: English and French; Portuguese and German in progress.