Almost everything written about the technical proposal for a public tender is advice: read the tender document, answer the criteria, back what you claim. What is usually missing is seeing it work. This piece does that with one example: a criterion, an answer that loses points and the same answer rewritten, with every sentence annotated.
The general method is in how to write the technical proposal for a public tender: read the tender document by its criteria, size the length by points and back every claim with a document. We do not repeat it here; we apply it.
The thesis: style does not score; what scores are the decisions that are written down and can be checked. The two versions in the example describe the same service. The one that wins is not better written: it is the one that left none of the decisions the criterion asked for unmade.
Invented example Everything that follows is illustrative. The city council, the tender document, the Annex III figures and the scores are invented: they come from no real tender and no client. A real committee scores in its own way. The example shows where scoring is lost, not how many points any proposal will get.
The tender document in the example
Imagine the support and maintenance contract for the e-government portal of a fictional city council. The tender document splits one hundred points across three criteria. Only the first is assessed by judgement, and it is the one we will answer.
| Criterion | Points | How it is assessed |
|---|---|---|
| 1. Incident handling plan | 12 | Judgement, in the technical proposal |
| 2. Maximum response time for critical incidents | 18 | Formula, in the automatic-criteria documentation |
| 3. Financial offer | 70 | Formula, in the automatic-criteria documentation |
Tender document invented for this example. The twelve points of criterion 1 are split into three sub-criteria: 1.1 (5 points), 1.2 (4) and 1.3 (3)
We stay with sub-criterion 1.1, the heaviest. The tender document words it like this:
“The proposal shall describe how incidents are classified and how their priority is decided, reasoning the proposal with the types and volumes in Annex III. It will be assessed whether the criteria can be applied without ambiguity and are accompanied by the evidence that backs them.”
Annex III, also invented, lists 312 incidents from the last year: 141 certificate-login problems, 96 errors in application forms and 75 of other kinds.
The version that loses points
This is the text that comes out when you write with the material already at hand and without rereading the sub-criterion. Almost every sentence in it is true. That is the problem: nothing it says can be scored.
We have extensive experience managing support services for public administrations and a highly qualified team. Our incident management methodology, based on industry best practices, guarantees agile, high-quality service. All incidents are logged and classified according to their severity and are resolved with the utmost diligence. We commit to responding to critical incidents in under 30 minutes. We have state-of-the-art tools for tracking them and a team committed to the council's satisfaction.
| Sentence | What the evaluator finds | Why it does not score |
|---|---|---|
| “Extensive experience… highly qualified team” | A claim with no project, date or profile | There is nothing to check. Besides, the sub-criterion does not ask about the company |
| “Methodology… based on industry best practices” | An adjective instead of a method | It would fit any contract. The sub-criterion asks for something applicable to these incidents |
| “Logged and classified according to their severity” | The word “classified” with no levels, no criterion and no owner | It repeats the tender document's question instead of answering it |
| “Responding to critical incidents in under 30 minutes” | A time commitment that the tender document scores by formula (criterion 2) | It earns nothing here and it is the costly mistake: see the next notice |
| “State-of-the-art tools… committed team” | Filler | It takes up space and answers no sub-criterion |
onext's own reading of an invented text
Risk The fourth sentence can cost more than a few points. Law 9/2017 provides that criteria resolved by formula are evaluated after those that depend on a judgement, and that this prior evaluation is made public when the documentation for the formulas is opened (article 146.2). A proposal that anticipates a formula datum breaks that sequence, which is why it can end in exclusion. We develop it in the method piece, and it is the reason someone other than the author reviews it.
The version that wins points
Same company and same service. Three things change: what opens each paragraph, where the data comes from and what was decided before writing.
1.1 Incident classification and priority criterion. We propose four levels defined by the effect on citizens, not by the technical symptom: critical (nobody can submit an application), high (it can be submitted, but one step is blocked), medium (there is a workaround) and low (query or improvement).
Annex III lists 312 incidents in the last year: 141 certificate logins, 96 application forms and 75 of other kinds. The first two groups are 76 % of the volume (237 of 312), so for them we prepare our own diagnostic guides and review recurring incidents every week.
Decision rule. Any incident that prevents submitting an application whose deadline falls within the next 48 hours is treated as critical, whatever its symptom. The on-call technician classifies at first contact; the service manager may reclassify it and records the reason in the log.
Evidence. The four levels and their application to three Annex III incidents are in Document 1.1-A (page 4). The response time per level is offered under criterion 2.
What changed, in the order in which it is best done:
- Open with the sub-criterion's question. The heading repeats its name. The evaluator finds the answer where they look for it and does not have to reconstruct it.
- Work with the tender document's own data. The 312 incidents in Annex III size the proposal. It is the proof that this contract was read and not another one.
- Turn every adjective into a decision. “According to severity” becomes four levels, a 48-hour rule, who classifies and who can change it.
- Point to the proof. “Document 1.1-A, page 4”. The tender document asks for evidence, and article 145.5 itself requires criteria that allow the information supplied by bidders to be checked.
- Leave out what another criterion scores. The response time is referred to criterion 2 and not anticipated here.
How a committee might score them
A hypothetical score helps to see the order of magnitude. We insist: the values are invented for the example.
| Version | Sub-criterion 1.1 (out of 5) | Why |
|---|---|---|
| A | 1 | No levels, criterion or owner, and no use of Annex III. It also anticipates a criterion 2 datum |
| B | 4 | Defines levels, rule and owner, uses Annex III and points to the proof. A demanding committee may deduct a point if it misses, for example, how a downgrade is handled |
Score invented to illustrate the example; it is not a measurement
The same move in the other two sub-criteria
There is no need to repeat the whole exercise. It is enough to see how the sentence that opens each answer changes:
| Sub-criterion | Opening that loses | Opening that wins |
|---|---|---|
| 1.2 Escalation and communication (4 points) | “We will maintain fluid and transparent communication with the council.” | “For a critical incident, the on-call technician notifies the designated council contact as soon as it is classified and issues an hourly report until closure, with known cause, pending cause and next action.” |
| 1.3 Continuity during absences (3 points) | “We have enough staff to cover any absence.” | “Each service has two people with access and proven knowledge of the environment; the person who covers has first completed an accompanied on-call shift, recorded in the service log.” |
Sentences invented for this example
What the law puts underneath
Version B does not answer a whim of style. It answers what Law 9/2017 demands of the tender document. Award criteria must be formulated objectively and must not give the contracting authority unlimited freedom of decision, and they must come with specifications that allow the information supplied by bidders to be effectively checked (article 145.5). The relative weighting of each criterion must appear in the tender document (article 146.3), so the split of points is not the bidder's opinion: it is the first thing that is read.
It also applies to improvements. If the tender document scores them, they must be sufficiently specified, with their requirements, limits, modalities and characteristics (article 145.7). An improvement the tender document does not specify is just another adjective.
The disagreement: what if the committee does not read like this?
The whole example assumes a committee that scores against the sub-criterion, sentence by sentence. We have not measured how many points each version earns before a real committee, and it would be wrong to present it as if we had. What can be said is that version A leaves all the work to the evaluator and version B saves them that work.
There is also a cost that B does not hide: it is more expensive to write. It forces you to have decided the levels, the 48-hour rule and who classifies, and to be able to meet them, because what the proposal states becomes a contractual obligation. That is why version A is tempting: it commits to nothing.
Limit Copying the shape of version B without the decisions or Document 1.1-A behind it is worse than version A: it promises evidence that does not exist. The method's rule still applies: if a claim cannot point to a document in your company that backs it, either it becomes something that can, or it goes.
A test with your latest proposal
Before reaching any tool, the example can be applied to what you already have. Take your latest proposal, pick one sub-criterion and run these five checks:
- Does the section begin with the name of the sub-criterion as the tender document writes it?
- Does at least one datum from the tender document or its annexes appear?
- Does every adjective have a level, an owner or a rule behind it?
- Does every claim point to a document and a page?
- Is there any datum that the tender document scores by formula under another criterion?
Where AI can help, and where it cannot
The laborious part of version B is not typing. It is locating Annex III, finding among your earlier proposals the document that backs each rule and gathering everything before the deadline. That is where an assistant with your company's context can help.
In the case of Licia, the tender assistant in onext Enterprise AI, it searches the Spanish Public Sector Procurement Platform (PLACSP) for the tenders that fit your company, based on what you have told it about the company. Analysing the tender document, assessing feasibility and preparing the technical proposal with evidence from your documentation is configured during set-up: it is not a ready-made product. The assistant consults collections of your company's documents, organised during that set-up, and in processes with risk it proposes and a person resolves from a pending-tasks inbox, with an audit log.
What no tool changes: it cannot create a decision you have not made or evidence your company does not have. That is the part that decides the points, and it stays yours. On the earlier decision of buying a tool or building the capability, we write in AI for public tenders: buy or build?; and on how to organise your documentation so the assistant can find it, in Agentic RAG over your internal documents.
If you are on the other side of the tender document, writing it, it makes sense to ask what evidence a software-development tender with AI should require: we explain it in what to ask for in a software-development tender with AI. And on what the law says and where a person must remain when a public administration uses AI, in human oversight in public administration.
Frequently asked questions
Where can I see an example of a technical proposal for a public tender?
This piece has an annotated one. It is an invented example, with a city council, a tender document and figures that do not exist, so that it can show the version that loses points and the one that wins them without pointing at anyone. For your own case, the best model is the tender document you are about to answer, together with the proposals you submitted before and the scores they received.
How do I write a section of the technical proposal so that it scores?
Start with the name of the sub-criterion exactly as the tender document writes it, answer with data from the tender document itself, turn every adjective into a decision that can be checked (which levels, who decides, under what rule) and point to the document that backs it. What the evaluator cannot find or check, the evaluator cannot score.
What is a criterion assessed by judgement?
It is an award criterion whose quantification depends on the evaluator's assessment, as opposed to those calculated with a formula, such as price. Article 146.2 of Law 9/2017 covers both types and requires those not resolved by formula to be evaluated before those that are, with a documented record.
Does the response time I commit to belong in the technical proposal?
It depends on how the tender document scores it. If it is scored by formula, it belongs in the documentation for the formula criteria, not in the judgement-based proposal: anticipating it in the proposal breaks the evaluation order set by article 146.2 and can end in exclusion. If it is unclear where it goes, ask the contracting authority during the question period.
Can I copy the example and adapt it to my tender?
Not as text. The figures, the levels and the 48-hour rule belong to the example and only hold for its invented tender document. What is reusable is the move: open with the sub-criterion, use the tender document's data, turn adjectives into decisions and point to the proof. The decisions, and the evidence that backs them, have to be your company's.
Can AI write the version that wins points?
It can help with locating and gathering your company's material and proposing a draft, but it cannot create a decision or evidence that does not exist: if the document backing a rule is not there, the problem is not the writing. In Licia's case, analysing the tender document, assessing feasibility and preparing the proposal with evidence from your documentation is configured during set-up. And in processes with risk, the assistant proposes and a person resolves.
Sources
- Law 9/2017, of 8 November, on Public Sector Contracts (Ley de Contratos del Sector Público), articles 145 (paragraphs 5 and 7) and 146 (paragraphs 2 and 3): consolidated text on the BOE (in Spanish) (consulted on 10 October 2026).
- The tender document, Annex III, the criteria, the documents and the scores in the example are onext's invention; they correspond to no real tender.
Bernat López is founder and CEO of onext, an AI boutique. He helps development and product teams work with AI with method —specification before building, a person who decides where there is risk and Spec-Driven Development— and applies to his own company what he proposes: onext runs on its own agentic system.
LinkedIn →