MANIFESTOCARABELA AI · MADRID · 2026

No company should skip a tender for lack of hours.

Carabela is built on an uncomfortable idea: most companies do not lose public contracts because their offer is worse. They lose them because they never get it submitted.

Companies do not need another tool for bidding. They need to win more contracts without bidding becoming the next bottleneck.

THIS DOCUMENT EXPLAINS WHY, AND WHAT WE BUILT TO MAKE IT TRUE

THE STARTING POINT

What is broken about the way bidding works.

None of these five is anyone's fault. They are what happens when an enormous job is done in human hours.

01

The market is vast and almost nobody reads it.

Spain publishes more than 180,000 tenders a year on the Plataforma de Contratación del Sector Público. A mid-sized company gets through a few dozen. Most opportunities are not rejected for being a poor fit — they are never seen.

02

The most expensive decision is made on instinct.

Bidding costs weeks of work, so go or no-go carries more consequence than anything else in the process. And it is almost always made on instinct, against the clock, without knowing who else is competing or what the last comparable contract was awarded at.

03

Every bid starts on a blank page.

The same company explains again who it is, what it has delivered and why it is financially sound. Again. The knowledge exists, but it is spread across folders, email threads and the heads of two people who are already busy.

04

Capacity is linear in headcount.

One person prepares a handful of bids a month. Doubling capacity means doubling the team. For decades that has been the only lever available: expensive, slow, and hard to reverse when the market cools.

05

Results never come back into the process.

The score is published, the file is archived, and everyone moves on. The most valuable information in the sector — who won, at what price, with what score and why — is lost exactly where the next bid should begin.

THE CEILING

A human department runs out of hours, not of judgement.

The ceiling on a company that bids is not its technical quality or its commercial ambition. It is how many offers fit into a month. Everything else — what gets published, the competition, the volume of tender documents — is another order of magnitude entirely.

PUBLISHED IN SPAIN
180,000+tenders a year
INDEXED BY CARABELA
1M+public tenders
WHAT ONE TEAM CAN PREPARE
Dozensof bids a year

MARKET FIGURE FROM PUBLIC PLATAFORMA DE CONTRATACIÓN DEL SECTOR PÚBLICO DATA · THE THIRD ROW IS AN ORDER OF MAGNITUDE, NOT A MEASUREMENT

WHY NOT ANOTHER SAAS

Software you have to operate does not remove the work. It moves it.

A search tool gives you a longer list. A bid manager gives you a better place to keep the work. A writing assistant gives you a draft somebody still has to direct, correct and assemble. All three are legitimate products and all three help. And all three leave the expensive part untouched: a person still has to push the process from one end to the other.

A SEARCH TOOLGives you a list. The work starts afterwards.
A BID MANAGERGives you somewhere to keep it. The work is still yours.
A WRITING ASSISTANTGives you a draft. Directing, reviewing and assembling it is still yours.
AN AUTONOMOUS SYSTEMGives you the prepared submission. Deciding is yours.

We do not want your team to be faster at operating a tool. We want them not to have to operate one.

WHY NOW

What changed is not the writing. It is the reading.

Almost the entire conversation about AI and bidding is about writing faster. Writing was never the expensive part. The expensive part was reading: reading everything that gets published, reading 142 pages of specifications and annexes, reading a contracting authority's award history, reading how the competitor who took the last contract actually won it.

A system of agents reads without tiring, without triaging by whatever is most urgent, and without forgetting what it read last month. And once reading stops being the constraint, the decision about which tenders to pursue — which is where bids are really won and lost — stops being an instinct and becomes a calculation.

Seven things that shape the product.

WHAT WE BELIEVE
  1. 01

    The work, not the tool.

    A bidding product should hand over a finished offer, not an interface for preparing one. If the customer has to learn to operate us, we have moved the problem rather than solved it.

  2. 02

    Autonomy is a measurement, not an adjective.

    Autonomous is not a marketing word: it is a number. How many decisions the system takes alone, and how many need a person. That number should fall from release to release, and we should be able to show it.

  3. 03

    Humans decide, they do not execute.

    Strategy, final price and signature belong to the company. Everything between those decisions is operational work, and operational work is ours. People step in where they bring what the system does not have: their own knowledge, strategic judgement, exceptions and approvals.

  4. 04

    Learn at a scale no team can reach.

    A good team learns from the twenty bids it prepared last year. A system can learn from every bid that gets published, scored and awarded. Not because it is cleverer — because it has seen vastly more.

  5. 05

    Every result feeds the next one.

    An award is not the end of a file. It is the input to the next bid. A system that does not close that loop is not improving: it is just repeating faster.

  6. 06

    No traceability, no trust.

    Every requirement has to point back to the line of the specification it came from, and every claim to the company evidence that supports it. Convincing text nobody can verify is worthless in a public contract.

  7. 07

    Say only what can be checked.

    Every figure on this site comes from a real case or a public source, and every statement about another product is quoted as its maker publishes it. In a market where everyone says autonomous, being able to show it is the only thing that makes the word mean anything.

THE LOOP

A system that improves is a system that closes the circle.

Every bid produces signals: a score per criterion, an award, a winning price, a discount, a competitor you were not expecting. In most companies those signals end up in a folder. At Carabela they return to the same system that will decide the next tender.

Tender
Bid
Evaluation
Award
Learning
A SHARPER NEXT DECISION
1M+
PUBLIC TENDERS INDEXED
41
PUBLIC SOURCES MONITORED
200K+
SPANISH COMPANIES MAPPED FROM AWARD HISTORY
WHERE THIS IS GOING

Bidding should stop being a department and become a capability.

We think public procurement is heading towards companies competing for far more contracts than they can work today, with teams that do not grow at the same rate. Not because the work disappears, but because it stops having to pass through a person to move forward.

That is Carabela's ambition: that a company can multiply its presence in the public market without multiplying its headcount, and that the human part concentrates where a human is genuinely needed. It is what orders every product decision we take.

WHO WE ARE

One company, one product.

Carabela AI is a Spanish company, based in Madrid, building exactly one thing: the system that takes over other companies' public procurement. There is no second business line and no consultancy behind it that scales by hiring.

We are a team of engineers with years of experience in artificial intelligence and in preparing and winning public tenders. That combination is what building this takes: the people writing the system are the people who have read the specifications, fought the deadlines and seen up close why a contract is won or lost.

WHY CARABELA

The caravel changed navigation because it could keep going where earlier ships could not. It sailed close to the wind, held its course in difficult conditions and opened routes that had been out of reach. Your company sets the destination; Carabela sails the route.

LEGAL NAME
CARABELA AI SL
BASED IN
Madrid, Spain
MARKET
Spanish and European public procurement
CONTACT
general@carabela.ai

Bring a real tender.

The shortest way to find out whether any of this is true is to put a contract you are actually considering in front of it and watch what it prepares.