The guidance tells you what to submit. It does not tell you how good your data has to be. We measured that bar from 239 published opinions — and a handful of deficiencies double the odds of failure.
Ask any experienced applicant what the hardest part of a novel food dossier is, and few will say "knowing what to include." The mandatory sections are public. The 2024 EFSA guidance lists them. The real question — the one that keeps consultants up at night — is different: is my data good enough?
That bar is mostly unwritten. The guidance tells you that you need compositional data, a production process description, a genotoxicity battery. It does not tell you that batch-to-batch consistency is one of the most reliable predictors of a rejected dossier, or that "we characterised the contaminants" is a sentence EFSA disbelieves more often than almost any other. Those weightings do not appear in any guidance document. They only appear in outcomes.
So we measured them.
We took 239 published EFSA opinions — the full set we could extract under the current and previous novel food regimes — and tagged each one for the categories of deficiency the panel actually raised. Then we looked at what happened to the application.
The baseline is the reference point for everything that follows. Across all 239 opinions, 18% ended non-favourably — meaning the panel either could not establish safety (unfavourable) or could not reach a conclusion on the data provided (inconclusive). The other 82% were favourable, most of them with conditions of use attached.
18% is the coin flip you start with. The interesting question is which deficiencies bend that number — and by how much.
These are the deficiency categories that, when present, are most strongly associated with a non-favourable outcome. The multiplier is relative to the 18% baseline.
Batch consistency — non-favourable in 37% of cases where it was flagged (2.1× the baseline, N=54). The single strongest signal in the data. When a panel doubts that your production is reproducible, it doubts everything downstream.
Contaminant characterisation — 36% non-favourable (2.0×, N=80). Not whether you mentioned contaminants, but whether you characterised them completely enough for the panel to rule out a concern.
Exposure modelling — 34% (1.9×, N=77). Intake estimates that are too broad, too optimistic, or that push past an established safe level.
Stability — 32% (1.8×, N=25). Whether the novel food, and its relevant constituents, hold up over the proposed shelf life and processing conditions.
Specification justification — 31% (1.7×, N=74). Specifications that are incomplete, unjustified, or not anchored to the batches actually tested.
Genotoxicity battery — 30% (1.7×, N=90). The most frequently flagged category overall — an incomplete or inadequately designed battery, or a test material that does not match the product going to market.
Process representativeness — 29% (1.6×, N=66). Whether the batches and studies in the dossier actually represent the commercial product and process.
Allergenicity assessment — 28% (1.6×, N=32). Particularly for protein-rich novel foods from new sources.
None of these is obscure. They are the foundations: what the product is, how it is made, what is in it, and whether the safety data rests on material that represents the real thing.
The pattern becomes concrete when you read what the panels actually wrote.
On batch consistency, one plant-extract opinion (ON-6196) records the panel's suspicion of data that was too clean: the results of all tested parameters across five batches produced at different time points were "identical ... which is considered unrealistic." The applicant submitted consistency; the panel read it as evidence the batches were not independently produced.
On contaminant characterisation, a chemical-nutrient opinion (ON-6134) concluded plainly that "the information provided does not allow a complete characterisation of the composition of the NF and is therefore considered insufficient." Not absent — insufficient. The applicant had submitted data; it did not clear the bar.
On exposure, a microalgae opinion (ON-3757) found that the estimated intake "exceed[ed] the ADI by approximately two- and three-fold." The science was fine; the proposed conditions of use were not.
On specifications, an iron-compound opinion (ON-5369) noted that the proposed specifications "did not contain any information on the particle size" — a single missing parameter, treated as a gap in the identity of the product itself.
In each case the applicant believed the section was done. The panel disagreed about whether the data was enough. That gap — between included and sufficient — is the bar.
A non-favourable opinion is the worst case. The far more common cost is a clock stop — a formal pause while the applicant produces missing data, often adding months to a timeline that already runs well past the nine-month statutory deadline.
Here the same deficiencies cluster at the top. Under the current regulation (opinions published from 2018 onward), roughly 80% of all dossiers receive at least one clock stop. But among dossiers flagged for batch consistency, that rises to 91%; for contaminant characterisation, around 82%; for stability, 80%; for specification justification, 77%. These deficiencies do not only raise the odds of rejection — they almost guarantee a delay.
EFSA's guidance is categorical by design. It tells you that a section must be addressed, not how heavily a panel weighs failure in that section relative to others. That weighting cannot be written down in advance, because it emerges from hundreds of individual judgements about whether specific data, for a specific product, was sufficient.
This is the gap that EFSA's own pre-submission advice does not fill. It will point you to the relevant guidance; it will not tell you whether your data is enough. The only honest answer to "is my data enough?" lives in the accumulated record of what panels accepted and rejected — and that record has to be read at scale to be useful.
That is the bar Borgh is built to surface: not a restatement of the guidance, but the empirical weight behind each requirement, drawn from the opinions where the same gap decided the outcome.
The practical implication is narrow and useful. The sections that most often sink a dossier are not the exotic ones — they are identity, composition, contaminants, specifications, and the question of whether your test material represents your product. Front-load them.
Treat consistency across batches as a result you have to demonstrate, not assert. Characterise contaminants to the point where a sceptical reader could rule out a concern, not just acknowledge it. Anchor your specifications to the batches you actually tested. Check that your exposure estimate survives the conditions of use you are proposing.
The bar is "is your data enough," not "did you include a section." Most applicants pass the second test long before they pass the first.
The outcome of each opinion — favourable, conditional, inconclusive, or unfavourable — is EFSA's own published conclusion. The deficiency tagging is Borgh's structured analysis of 239 opinions, so the per-category counts are our reading of which gaps each panel raised, not an official EFSA classification. Rates are expressed within our dataset and as multiples of its 18% baseline.
Two caveats apply. The 239 opinions span both the current novel food regulation and the earlier regime, including older opinions where clock-stop activity is sparsely recorded — which is why we report the clock-stop figures against the post-2018 cohort. And because only published opinions are analysed, the dataset is weighted toward applications that survived to a conclusion; dossiers withdrawn or invalidated earlier are not represented. The direction of every signal above is robust; treat the exact percentages as well-grounded estimates, not decimals.
- Own structured analysis of 239 published EFSA opinions (2003–2026), extracted via Europe PMC and EFSA opinion PDFs. Deficiency categories and outcomes tagged per opinion; baseline non-favourable rate 18% (N=239). - Verbatim panel statements quoted from EFSA opinions ON-6196, ON-6134, ON-3757, and ON-5369. - EFSA (2024). Scientific Guidance for the preparation of applications for authorisation of novel foods. *EFSA Journal*. - Neytinck, J. et al. (2025). Analysis of novel food application timelines and outcomes. *npj Science of Food* — clock-stop and timeline context for the post-2018 cohort.
Paste a section and Borgh will flag the gaps that matter — with references to the relevant EFSA guidance.
Try a section →