OC
Oriol Cervantes · Decision Intelligence · Sitges & London

Twenty years inside UN-scale institutions, restoring trust in the data behind decisions worth billions. Now with a council of AI models I direct.

Hard decisions, turned into models you can interrogate.

Three cases below, every one of them live. Change the criteria and reach your own verdict. Then decide whether we should talk.

See the three cases → Book 30 minutes

Three ways to work with me: a conversation, a decision stress-test, or the CV.

Scale · verified
160+ country offices · USD 2.5B a year
4 AI models · one human verdict
01

Three decisions, rebuilt as models you can play with.

Each one is a decision somebody had to make. The problem, what I did, and what changed. Open any of them: move the criteria, open the evidence behind a number, and see whether the recommendation still holds.

150 country offices · USD 3.25B · 160,550 purchase orders · synthetic data
Institutional procurement for oversight · procurement leadership · country teams

Paretum

The problem

A network of 150 offices orders USD 3.25B a year through 160,550 purchase orders. Standard BI shows the transactions. Nobody can say where to look first, or which number can be trusted.

What I did

The spend-mapping and office-clustering method I ran by hand inside a global institution from 2012, rebuilt with AI as an agency-neutral engine: a five-mark scorecard (fit, efficiency, price, resilience, local economy) against 150 peers, with every finding opening its chart, its table and its method, in that order. Everything is a signal to review, never a verdict; the controller decides.

The original Excel heatmap, countries by categories
The hand-made original · ~2017 · Excel30 million cells a year (500,000 purchase-order rows × 60 fields), summarised into 30,000 on one heatmap: 150 countries × 200 categories. 350 reports and 150 risk matrices a year, before any AI.
you are the health minister · set the budget · watch the system respond
Public health & financing for ministries · funders · foundation boards

Health Systems Are Decision Systems

The problem

Where a child is born sets a thirty-year gap in life expectancy before a single choice is made. Global health has no shortage of actors or money; they flow in different directions, and only about a third of the signal in raw data reaches a decision intact.

What I did

Public-health, financing and outcome data reconciled into one model, then challenged by four AI models for gaps and overclaiming before any chart was drawn. Built as six scroll-driven chapters where the reader holds a real-sized budget and moves the money, and the ecosystem of funders, ministries and communities responds live.

4 parameters · 100 scenarios · holds or breaks
Decisions under uncertainty for boards · funders · executive committees

The Fragility Lab

The problem

A 4.8× recommendation looks robust on paper. Then reality intervenes, and the biggest number is usually the one whose model is most wrong in your favour.

What I did

Each of four parameters is estimated from imperfect evidence. The tool runs a hundred scenarios and maps the whole parameter landscape, so the gap between the optimistic number and the realistic spread is visible. Knowing where a model breaks in the field is twenty years of institutional work, not a statistic.

02

One promise, verified twice.

Most analysis asks you to trust two things you can't see: that the data is sound, and that the reasoning is sound. I make both inspectable, before a decision is made on either.

Most decisions don't fail at the strategy. They fail at data nobody could trust, and reasoning nobody could check.

At a major UN agency, reconciling 500,000 procurement lines by hand took two months. The same first reconstruction pass now takes a day, complex domains then iterate from there. That speed isn't the product; it's what makes the rest possible: once the foundation is sound, I rebuild it into something a board, or a regulator, can defend.

2 months → 1 day*
first-pass data reconstruction, 500k rows, 60 fields, by hand then with method · *complex domains iterate further evidenceSourceUN agency procurement remediation · ERP reconstruction (first-hand)TierOfficial
Reasoning trust

Council of Models

Several AI models as a challenge panel, not one source of truth. They disagree, stress-test each other, and surface the blind spots. Human judgement is the synthesis layer.

Data trust

Turbo Reset

Chaotic, unreliable data rebuilt into a clean, structured, auditable foundation in a day, with every value traceable to source, tier and confidence.

Decision intelligence · live

Not one model. A council I direct.

Four models advise. I set the brief, weigh the disagreement, and make the call.
Oriol · the brief
Claude · synthesiser

ChatGPT · model 02

Gemini · model 03

Grok · model 04

Decision · synthesis layer
synthesising

Oriol Cervantes
03 · Travel Shockwaves · live engine

Live proof. Not a screenshot.

A live preview of the engine itself, built to the same brand, running inside the page. Hover to bring it to colour, drag the controls, navigate four views.It comes to colour and moves on its own, use the arrows to change views. The full instrument opens in one click.

LIVE PREVIEWTravel Shockwaves · interactive engine Open full engine ↗
An interactive data essay

TRAVEL
SHOCK
WAVES

Tourism does not simply bounce back. It mutates. An engine for how shocks stop, compress, reroute and reshape global travel.

1.52BArrivals 2025
−72%2020 Collapse
60MChina Gap
138%M.East vs 2019
35MDigital Nomads
Global arrivals · 2000–2035 1.52B 2025
0005101520253035
Shock signatures · 4 types · 5 dimensions
DSpDuReLe
STOP · Health
COMPRESS · Economic
REROUTE · Geopolitical
MUTATE · Climate
Atlas · global shockwave map
Watch shocks reshape the map.

Health stops flows · economic dims them · geopolitical bends them · climate erodes them. Pick a lens, the globe responds.

Shock lens
Shock signature · the DNA of disruption
Depth
0
Speed
0
Duration
0
Rebound
0
Legacy
0
Recovery Race
Who bounced back first?

A race is easier to understand than a dense table. Press play and the regional divergence becomes immediately visible.

2019

All regions collapsed in 2020 (−72% globally). Asia-Pacific deepest at −84%.

✈ Data Film
12 seconds that explain 25 years.

An animated data narrative, no commentary needed. Press play and let the numbers tell the story.

Screen 01 / 04 · Overview
hover to bring it to life · everything moves on its own · take control anytime
04

Twenty years of decisions at institutional scale.

The rigour the largest international institutions demand, applied to the data their senior decisions actually depend on. This is the track record the AI work stands on.

Major UN Agency10 years
Procurement & expenditure analytics across 160+ countries and ~USD 2.5B a year evidenceSourceInstitutional financial records · 160+ country officesTierOfficial. Remediated hundreds of millions in misclassified data, restoring audit and decision confidence.
UN Agency. SupplyDirector's Office
Global procurement strategy; an automated dashboard methodology deployed across 145 countries and 27 categories. ~USD 6M annual savings.
UN Agency. HealthBI Project Lead
A standardised scoring and benchmarking framework across 53 country offices, restoring cross-country comparability during organisational change.
The LEGO GroupSenior Data Analyst
Forensic scrap analysis across five global factories, separating signal from noise in high-volume production data for manufacturing leadership.
Global consultancyPresent
Senior individual contributor in decision intelligence, data strategy and AI readiness, restoring decision clarity where ownership and reporting were fragmented.
Claude“Robust on average, fragile at the tail.”
ChatGPT“Survives the base case, not the stress case.”
Gemini“Stress-test the worst five percent first.”
Grok“‘Average’ is the comfortable lie.”

Oriol · synthesisHold.

Work with me. Three ways in.

Pick the one that matches your situation. All three start with the same thing: a decision, and the evidence behind it.

01 · A conversation

You are appointing strategy, data or AI leadership.

Thirty minutes, no deck. Bring the decision your organisation is circling and I will tell you how I would open it up.

Book 30 minutes →
02 · A decision stress-test

You have one decision and data you do not fully trust.

FormatWritten brief, no meetings required
DeliverableInteractive stress-test + one-page verdict
Turnaround48 hours for the first read, five days for the model
Send the decision →
03 · The CV

You are hiring, and need the record on one page.

Twenty years across five institutions, decision intelligence, data strategy and AI readiness. The full record is on LinkedIn; the one-page CV comes by email.

LinkedIn ↗ Ask for the CV →
Direct oriol.cervantes@gmail.com calendly.com/oriol-cervantes linkedin.com/in/oriol-cervantes

Based in Sitges, working with London and international institutions. Independent work, shared in a personal capacity.

05

More from the shelf.

The same method, proven on other subjects. Every one is a live engine; none of them is a slide.

Travel ShockwavesTravel Shockwaveshow shocks reshape global travel, 2000 to 2025 The Civilisation LabThe Civilisation Lab193 nations, 25 indicators, 1900 to 2050 World Cup Memory MachineWorld Cup Memory Machineninety-six years, every figure labelled Atles dels CastellsAtles dels Castells1,453 castles, a thousand years, in 3D
Carta Blanca Labsgames, atlases, languages and what is still in the ovenbrowse ↗ Studio & the generative archivevisual direction, Prism, and a long generative practice that predates the current AI wavein the Labs ↗

Disclaimers

Method, not real data. The interactive engines on this page are demonstrations of method. Their figures, datasets and scenarios are stylised or illustrative, they do not reproduce confidential client data, nor any third party's proprietary estimates.

Imagery. Every generative image is original work, art-directed in-house. No stock.

Track record. Entries summarise work delivered under confidentiality; institutions, figures and outcomes are described at a level that preserves it, and are shared in a personal capacity.

Sources

Per-claim provenance. Each line in the track record carries its own evidence marker, open it to see the source and its tier.

Tiers  ·  Official, institutional records   /   Verified, corroborated   /   Indicative, estimated.

Interactive demos. Run on synthetic or public-shaped data generated purely for illustration; no underlying confidential figures are exposed.

Contact. Specific methodology notes available on request.