How to write a Horizon Europe proposal: what it looks like from the other side of the table

The number that should change how you write

Almost every guide to Horizon Europe is organised around the thresholds. Score at least 3 out of 5 on Excellence, Impact and Implementation, reach 10 out of 15 overall, and you have written a fundable proposal. This is true, and it is close to useless as a writing strategy.

Here is the number that matters instead. Across Horizon Europe, roughly 70% of submitted proposals are assessed as excellent and worthy of funding. Around 16% receive a grant. The oversubscription rate across evaluated calls has run at about 4.7, meaning the budget would need to be roughly five times larger to fund everything applicants asked for. In some national contexts the picture is starker still: a large majority of proposals rated excellent are never funded.

Sit with the implication for a moment. If you write a proposal that is clear, compliant, well structured and technically sound, you have most likely joined the 70%. You have not meaningfully improved your odds of being in the 16%. The threshold is an entry requirement, not a target. The real competition happens in the half point increments above it, and those half points are not awarded for compliance. They are awarded for conviction.

What actually happens to your document

Understanding the mechanics changes how you write, so it is worth being concrete about the process rather than the paperwork.

Your proposal is assigned to at least three independent experts. They read and score it remotely and individually, without discussing it, each producing scores and written comments against the three criteria. Only afterwards does a consensus discussion take place, moderated to produce one agreed score per criterion and a single Evaluation Summary Report. Where a call attracts many proposals, a panel review then ranks everything that passed.

Three consequences follow, and they are the practical reason this article exists.

First, your proposal is read under time pressure by someone with a stack of others. Evaluators are competent and conscientious, and they are also human beings working through a queue. A proposal that requires the reader to hold six things in their head before the argument resolves will be scored by someone who did not hold all six. Front load. State the claim, then support it. Never build to a reveal.

Second, your proposal must survive being described by someone else. In the consensus discussion, one evaluator will often summarise your project to colleagues who read it less recently or less closely. If your project cannot be restated accurately in two sentences by a tired person, the version that gets discussed is not the version you wrote. This is the single most underrated constraint in proposal writing. Write a project that paraphrases well.

Third, disagreement gets resolved toward the middle. When one evaluator scores a criterion 4.5 and another 3, the consensus rarely lands at 4.5. Ambiguity in your text is not neutral, it is a downward force, because the sceptical reading is the one that has to be argued away. Every sentence that could be read two ways will be read the less generous way by at least one person in the room.

The rules, and what each one means in practice

These are the operational specifics for the 2026-2027 Work Programmes, verified against the revised Standard Application Form published in December 2025. The interpretation next to each is mine.

Verified against the revised Standard Application Form for RIA and IA, version dated 15 December 2025, applicable to 2026-2027 call topics. These limits change between Work Programme cycles. The only authoritative version is the template attached to your specific topic in the Funding and Tenders Portal, and you should download it there rather than trusting any article, including this one.

What changed for 2026-2027, and why a lot of advice online is now wrong

The Commission issued a revised Standard Application Form in December 2025. The evaluation architecture is unchanged, but three edits matter enough that guidance written against the older template will now push you in the wrong direction.

The scale and significance requirement is gone. Section 2.1 previously asked applicants to describe the scale and significance of expected contributions. That instruction has been removed. The Commission's reasoning, which matches what the reading experience was like, is that the requirement was difficult to quantify meaningfully and generated elaborate narratives that did not improve anyone's understanding of what the project would actually change.

This is worth dwelling on, because you will still find recent guides instructing you to quantify every impact with target values and estimate the magnitude and timeframe of each one. Against the current template, that advice spends scarce pages on material the form no longer asks for. What Section 2.1 wants now is a clear account of who benefits, how they benefit, and why your approach produces that result. Fewer invented numbers, more defensible causal logic.

I want to be precise here, because the opposite error is also available. This is not a licence for vagueness. Quantification still helps wherever you have a real basis for it, and an unquantified claim that could have been quantified still reads as evasive. The change is that manufactured precision is no longer rewarded by the form itself.

Section 3.1 got lighter. The tables on other cost categories and on in kind contributions were removed, and the purchase costs table now only needs justification when equipment depreciation exceeds 15% of personnel costs. The suggested internal allocation for the work plan section dropped from 14 pages to 12, or 17 with lump sum funding.

Everything got shorter. RIA and IA went from 45 pages to 40, CSA from 30 to 25. The suggested allocations inside Section 2 fell too, with pathways to impact at 3 pages and measures to maximise impact at 3. Section 2.3, the summary, became optional.

Fewer pages with an unchanged scoring rubric means the cost of a weak paragraph went up. There is now less room to be comprehensive, which in practice means the proposals that suffer most are the ones that treat every section as an obligation to be discharged at equal length rather than an argument with a load bearing core.

The three criteria, read from the scorer's chair

Excellence

Excellence is where evaluators form the impression that colours everything after it. The formal content is objectives, ambition beyond the state of the art, and methodology.

What separates a 3 from a 4.5 is almost never the quality of the science. It is whether the state of the art section is honest. A comparison that presents your approach as superior on every dimension reads as marketing, and marketing makes an evaluator sceptical about the rest of the document. A comparison that concedes where existing approaches are genuinely better, and then explains why your trade off is the right one for this problem, reads as expertise. The second version is more persuasive precisely because it demonstrates you understand the field well enough to know its strengths.

The other reliable separator is whether objectives are falsifiable. "Improve the efficiency of X" cannot be assessed, so it gets assessed as weak. "Achieve Y throughput at Z cost, measured by W, against a baseline of V" can be assessed, and can also be defended in the consensus discussion by an evaluator who liked your proposal. Give your advocate ammunition.

Impact

Impact is where most proposals lose the funding decision, and the reason is structural. Researchers are trained to be careful about causal claims, so they write impact sections in a register of appropriate scientific hedging. Evaluators read hedging as absence of conviction.

The distinction that reliably moves scores is between outputs, outcomes and impacts. Outputs are what you produce: deliverables, publications, a prototype. Outcomes are the change in behaviour of someone outside the consortium: an end user adopting the method, a standards body incorporating the result, a company changing a process. Impacts are the longer term effects that follow. Proposals that score 3 describe outputs in detail and then assert impacts. Proposals that score well describe the outcome, the specific actor who has to do something differently, and why that actor would.

Name the actor. Not "industry stakeholders" but the category of organisation, ideally with evidence that you have talked to some of them. The most convincing impact sections I have read contain a sentence a marketing team would consider a weakness, something like "adoption depends on certification, which typically takes eighteen months, so our exploitation plan begins that process in month twelve". That sentence tells an evaluator you have thought about the world rather than the proposal.

Quality and efficiency of the implementation

Implementation is the criterion applicants underestimate most, because it looks administrative. It is where the panel decides whether it believes you.

The work plan, the Gantt chart, the person month allocation and the risk table are read as evidence about whether the consortium has actually thought through delivery, not as forms to complete. The give away is coherence across sections. If Section 1 promises validation in three industrial settings and the work plan contains a single validation task with four person months, you have not been caught in an inconsistency, you have demonstrated that the sections were written by different people who did not reconcile them. Evaluators notice this constantly and it is expensive.

On risk tables, the pattern is nearly universal: every risk listed is technical, every mitigation is "the consortium will monitor closely and adapt". A risk register containing only technical risks tells an evaluator you have not considered that a partner might withdraw, that a key hire might not materialise, or that your market assumption might be wrong. Include the risks that are uncomfortable. The willingness to name them is itself evidence of competence, and it is cheap to supply.

Why 3 out of 5 is the score to fear

If you are rejected, you will receive an Evaluation Summary Report with scores and written comments. How you read it determines whether your next attempt is better or merely different.

A 2 is bad news and good information. Something specific is broken and the comments will usually name it. You can fix a 2.

A 3 is worse news dressed as reassurance. It means the evaluators found nothing wrong. Your proposal was clear, complete, compliant, and it did not persuade anyone to argue for it. There is no single element to repair, which is why so many resubmissions of a 3 come back as another 3: the applicant tightened the prose, added detail, addressed the specific comments, and left the underlying argument untouched.

Recovering from a 3 usually requires deciding what the proposal is actually about and being willing to subordinate everything else to it. That is a harder and more uncomfortable revision than fixing a 2, and it is the one worth making. If you are resubmitting, the useful question is not "what did they criticise" but "what would have made someone in that room fight for this".

It also helps to remember the arithmetic from the top of this article. If about 70% of proposals clear the bar and roughly one in six is funded, then a competent rejection is the normal outcome rather than a verdict on your project. Many funded projects were funded on a second or third attempt. The ones that succeed on resubmission are generally the ones that changed the argument, not the ones that polished it.