Search for AI for public tenders and the market hands you a list of products that look remarkably alike: they find opportunities, read tender documents, summarise and draft. Almost all of them solve the same problem the same way, and none answers the question a COO actually has to settle: once this works, whose capability is it?
This piece is not a tool comparison — it is the decision framework that comes first. If what you need is the drafting method, it is in how to write the technical proposal for a public tender. If your case is a private client's request for proposal — free format, negotiation — the piece is automating RFP responses with AI. Here we are talking about public tendering: a tender document, a legal deadline and published award criteria.
The market already decided for you, and it decided to sell a box
Half an hour of searching is enough to see the pattern. Gober presents itself as an assistant that knows every tender in Spain and generates tasks for your team. LICAI automates document analysis and opportunity detection. TendersTool provides real-time data on IT tenders and awards. AIverso orchestrates workflows so a small company can reach what only large consultancies used to have. Everglow, Tendios and Licita-ia play variations on the same board.
These are legitimate products and some are well built. But notice what they share: they all sell access to a capability that lives outside your company. You pay a subscription to use an engine you do not control, trained on a corpus that is not yours, with a drafting standard you did not define.
That is not a moral criticism — it is a description of the model. And it has very concrete implications as tender volume grows or as the requirements get harder. The question none of those products asks on its website is the one you should ask before signing: which part of this do I want my team to know how to do in two years' time?
What a tool genuinely automates, and what none of them can
It helps to split the real work of a tender response into three layers, because they automate very differently.
Layer 1 · Find and filter. Spotting what has been published, whether it fits your standing and whether it is worth bidding. This is bulk reading of bulletins and tender documents. It automates well and carries little risk: a tool's mistake here costs you a missed opportunity, not a disqualification.
Layer 2 · Retrieve and adapt your own material. Finding, in your own history, the methodology description, the org chart, the certifications, the similar project you already delivered — and adapting it to this particular tender. This is where most of the time goes and where AI contributes most. It is also where the quality of the output depends entirely on your documentation, not on the model.
Layer 3 · Decide what to commit to, and sign it. What improvements you promise, with what resources, at what price and at what risk. That is business judgement and legal liability. It is not delegated to a system, and no serious tool claims otherwise.
The usual commercial misunderstanding is selling layer 2 as if it were layer 3. The usual technical misunderstanding is believing the model solves layer 2. Context solves it: which of your documents are reachable, how they are structured, what counts as a reliable source and what may never be claimed without evidence. An excellent model on top of disorganised documentation produces proposals that read well and cannot be evidenced. That is precisely the mistake that loses points.
The three questions that settle the decision
There is no universal answer. There are three questions that, answered honestly, almost always tip the balance.
1. How often do you bid each year? Below roughly ten tenders a year, the cost of building internal capability rarely pays back: buying is reasonable. Above a steady flow — or if tendering is your main revenue channel — every month of subscription buys time, not learning, and the maths changes.
2. Is your edge in the tender document or in how you answer it? If you compete for contracts where the technical section carries weight — and in more than half of all tenders quality scores above price — the way you argue your methodology is part of the product. Outsourcing it means outsourcing your commercial argument.
3. What happens the day you stop paying? The uncomfortable one. If the answer is "we go back to doing it the old way", you did not buy a capability: you rented an outcome.
A table to take to your committee
| Situation | Signal | Reasonable decision |
|---|---|---|
| Fewer than 10 tenders/year, no proposals team | Low volume | Buy. Start with a tool and measure |
| Steady flow, proposals always written by third parties | Recurring cost, zero learning | Build the capability, with the tool as support |
| Contracts where quality outweighs price | The technical argument is your edge | Build: judgement is not subcontracted |
| Internal documentation scattered or out of date | Layer 2 will fail on any engine | Organise the context first, decide afterwards |
| Urgent need for a tender closing soon | Legal deadline | Buy now, settle the structure later |
That last row matters: a tender deadline is not negotiable. If you have one on the table, buy whatever solves this week and make the structural decision when there is no clock running.
The cost that is not in the budget
Every tool evaluation compares monthly fees. Almost none calculates the exit cost, which is where you find out whether the investment built anything.
Exit cost answers a simple question: if you cancel the subscription tomorrow, what stays inside your company? With a pure subscription model, usually the documents you already generated — and not much else. What does not stay is the judgement about which of your own material to use for each type of award criterion. Nor the structure of your documentation. Nor the knowledge of which claims can be evidenced, and with what.
That knowledge exists. It is generated every time you answer a tender. The only question is whether it settles in your organisation or in your supplier's.
Our position on this is well known and it is the same for tendering as for any other process: the goal of an AI partner is to become unnecessary in the area it has transformed. Not because the relationship ends, but because a relationship that depends on the client never learning is not a healthy one. When the judgement lives in your team, you can change tool, model or supplier without starting over.
Your history is not an archive: it is the raw material of your intelligence
There is a difference between a tool that drafts and a system that learns, and almost every sales conversation ignores it. Every tender you bid for leaves material behind: what you offered, how you argued each criterion, what score you got and where you lost it. That trail is not filed paperwork: it is the only historical record that exists of how your company competes.
And in public procurement there is a second source almost nobody exploits: the award is public information. Who won, at what price, with what score per criterion. It is a corpus about your own market — which competitors take which contract types, at what price level, where they score better than you — that nobody has to sell you, because it is already published.
When a system draws on both — your entire history plus the public award record — it stops being a drafting assistant and starts being corporate intelligence: which criteria you systematically lose points on, which contract types suit you, who you are really competing against in each. That is no longer time saved: it is the judgement to decide what to bid for and what to skip, a far more valuable decision than drafting quickly.
It is also why an asset like this is worth more in year two than in year one: the proposal bank grows with every file and the lessons stay inside. A generic subscription does not build that, because it is not built on your history: it drafts. That is the difference between renting a capability and compounding an advantage.
It is exactly what Licia, our tender assistant, does: it works on your full history and on the public award record —including the analysis of what your competitors are winning and with what score— and the postmortem of every file stays inside your organisation. The same principle underpins onext Enterprise AI.
When buying is the right call
It would be dishonest to write this without saying when the answer is to buy. It is so in more cases than would suit us to admit.
- When the volume does not justify it. Building capability has a fixed start-up cost. With few tenders a year it does not pay back, and standing up an internal structure for it is over-engineering.
- When the layer you need is layer 1. If your problem is knowing what has been published and filtering it, there are tools that do it well and no reason to rebuild them.
- When a deadline is on top of you. As said: the tender does not wait for your strategic plan.
- When you do not yet know what you need. A few months of subscription is a cheap way to discover where your real bottleneck is. Many companies believe their problem is drafting and discover their problem is that they cannot find their own documentation.
Buying is the wrong call in one specific case: when it happens by default, without anyone asking where the capability should end up. That is where companies sign subscriptions that three years later still cost the same and have left nothing behind.
What "building the capability" actually means
"Building capability" sounds like a long, expensive project. In this area it is more contained than it seems, because it comes down to four concrete assets — and all of them stay in-house.
1. Your documentation, organised and reachable. Past proposals, methodologies, certifications, team CVs, delivered projects, financial standing. This is not a two-year document management programme: it is making that material retrievable by a system, with the source identified. It is the asset with the highest return and the one almost nobody has ready.
2. The rules on what may be claimed. What can be said with what evidence behind it, and what is never said without a document to support it. It is the difference between a proposal that reads well and one that survives evaluation.
3. A workflow with the human in the right place. Not reviewing everything just in case, but deciding what only a person can decide: what is offered, what is committed and what is signed. Automation removes the blank page; it does not remove accountability.
4. The method, written down. How to read a tender through its award criteria, how to distribute length according to the points, which mistakes disqualify. This is method, and method is the only thing that survives a change of tool or model.
All four assets are supplier-agnostic. They work with the tool you bought, with the next one, or with none at all.
A path that does not force a day-one choice
The decision does not have to be taken in one go. In practice it works better in phases.
- Measure before buying. How many tenders a year, how many hours per proposal, how many you drop for lack of time. Without that number, any decision is a hunch.
- Organise one type, not everything. Pick the contract type you bid for most and organise only that material.
- Use a tool for layer 1 right away. Finding and filtering opportunities is a solved problem; do not rebuild it.
- Work layer 2 on your own documentation. This is where quality is decided, and where the result depends on your material rather than the engine.
- Write the evidence rules while answering the next tender, not in a theoretical document beforehand.
- Put human review where it decides, not where it keeps busy.
- Review at six months using the same number from step 1. If time per proposal has not fallen, the problem was not the tool.
None of this requires picking a side between buying and building. It requires knowing what you are buying when you buy.
Frequently asked questions
Is it legal to use artificial intelligence to prepare a public tender?
Yes. What you submit is your company's bid, signed by your company, and responsibility for its content is yours regardless of the tools used to prepare it. What does not change is the requirement that everything you state be true and evidenced.
Can an AI tool write the entire technical proposal?
It can produce a full draft. Finding, selecting and adapting your own material is the most time-consuming part and the one that automates best. Deciding what to commit to, reviewing and signing remain human: automation removes the blank page, not the judgement.
Which costs more, the subscription or building the capability?
In year one the subscription is almost always cheaper. The honest comparison is not annual cost but exit cost: what stays inside your company the day you stop paying. With a steady flow of tenders, the maths changes quickly.
Is AI any use if my documentation is a mess?
Very little, and it is the most expensive mistake. With scattered documentation any engine produces plausible but unevidenced proposals, which is exactly what loses points. Organising retrievable material pays off more than switching models.
Does this apply to a private client RFP?
Not entirely. Public tendering is governed by the tender document, with legal deadlines, published award criteria and formal grounds for exclusion. A private RFP allows free format and negotiation. The private case is covered in automating RFP responses with AI.
Conclusion
The AI for public tenders market is settled on the supply side: the tools exist, they work and some are good. What is not settled is the underlying question, and it is the only one a committee should debate before signing anything: where do you want the ability to answer a tender to live?
If your volume is low or you have a deadline on top of you, buy. If tendering is part of how you make money and technical quality is your edge, organise your context and keep the judgement. In that case the tool becomes what it should always have been: an interchangeable instrument, not the place where your knowledge lives.
Jordi García is Tech Lead at onext. He works on bringing AI into governed production across development and product teams —with Spec-Driven Development, context engineering and human verification at every step— and authors onext's technical insights on the method, quality and cost of applied AI.
LinkedIn →