How peachdeck scores pitch decks, what happens to your documents, and why a purpose-built evaluation engine gives you something a general-purpose chat assistant cannot.
Last updated: 14 September 2026
peachdeck is a pitch deck evaluation engine. You upload a deck and it returns a structured score across more than 70 dimensions, covering how the deck is built and what it actually says, along with the evidence behind each score and concrete recommendations for improving it.
It is built and operated by VERS.AI, a French company, and is used by founders preparing a raise, by investors screening dealflow, and by the juries of pitch competitions and accelerator selections.
A general assistant gives you an opinion shaped by how you phrased the question. peachdeck gives you a measurement. The rubric is fixed, the output structure is guaranteed, and the same deck produces the same evaluation rather than a different one each session.
The difference is architectural, not a better prompt. A chat assistant is a general-purpose conversation that happens to be about your deck; peachdeck is a system built for one job, which is turning a deck into a reliable, structured, comparable evaluation.
Because comparability requires it. If the criteria shift between two evaluations, the two scores cannot be placed side by side, which makes the output useless for ranking a batch of decks or for tracking whether your own deck improved between drafts.
peachdeck runs a system prompt derived directly from written scoring specification documents. Every deck is measured against the same multi-category rubric, so the evaluation logic lives in the system rather than in how well a user phrased a question.
Scoring is split into form and content. Form covers how the deck is constructed: structure, clarity, typography, visual consistency and language quality. Content covers what it argues: the narrative, the strength of the evidence, the market reasoning and the credibility of the claims.
On the plans that include it, a third layer scores investor fit, assessing how well a specific company matches a specific investor thesis on sector, stage, geography and amount sought.
No. Uploaded documents are never used to train any model, neither peachdeck's own model nor any third party's. They are processed to produce your analysis and for nothing else.
They are also never shared with the investors listed in the matching database, and never sold or passed to any commercial third party.
In France. Documents are hosted at OVH in France and analysed by peachdeck's own model running on Scaleway infrastructure in France. They do not leave the European Union and are never transmitted to a third-party AI model provider for processing.
This matters for pre-fundraise material, which is typically confidential and often contains unprotected intellectual property.
Uploaded pitch decks are not retained. They are deleted after analysis, with a retention period of zero days. Internal access during processing is limited to the people who need it to operate and support the service, and that access is logged.
No. The evaluation engine is a purpose-built model that peachdeck self-hosts on its own European infrastructure, rather than a general-purpose model rented through a third-party API. No third-party AI model provider receives your deck at any point.
Because structured-output modes on general-purpose APIs are best-effort. The model is guided toward a schema but can still emit malformed fields, drop required keys or invent extra structure, which breaks parsing and forces defensive error handling on every call.
peachdeck constrains generation with a custom grammar at the decoding level, restricting every token the model may emit to what the target format permits. Invalid output becomes structurally impossible, so the score is parseable by construction rather than by luck.
Yes, that is the point of the design. The rubric is fixed, the system prompt is derived from a written specification rather than written fresh by the user, and the output structure is constrained at decoding time. There is no drift between one evaluation and the next.
This is what makes it possible to compare two different companies fairly, or to measure whether the second draft of your own deck is genuinely better than the first.
No, and it is not designed to. peachdeck is a deterministic, expert-assisted tool: it uses AI to guide and accelerate analysis while making the evaluation consistent, transparent, reproducible and auditable. The final decision stays with the founder, the investor or the jury.
peachdeck scores the fit between your company and investors in its database using sector, stage of development, geography and the amount you are raising, then ranks the investors whose stated thesis aligns most closely with your profile.
On higher tiers it also produces targeting guidance and an approach strategy for the investors on your shortlist.
No. No investor pays to be included in the matching results. Rankings reflect thesis alignment with your company, not a commercial relationship with peachdeck.
Yes. Investors can batch-screen inbound decks against their own custom evaluation criteria and force-rank opportunities by sector, stage and geography, which turns a pile of inbound decks into a prioritised list.
Pitch competitions, accelerators and summits use the same engine to score and rank applicants consistently and to give juries a shared basis for comparison.
Founder plans start at €29 per month for form and content review, €59 per month to add investor fit scoring, and €159 per month for proactive investor targeting and investment-committee simulation. Each plan includes a monthly allowance of credits, and unused credits roll over for active subscribers.
Investor, pitch competition and summit pricing is quoted per organisation based on volume.
Three groups. Founders and CEOs preparing a fundraise who want to know how their deck reads to an investor before sending it. Investors who need to screen dealflow consistently and at volume. Juries and organisers of pitch competitions and accelerator selections who need a shared, defensible basis for ranking applicants.
The same pitch deck, evaluated two ways. The difference is not the quality of the underlying model, it is what the system around the model guarantees.
| Aspect | General-purpose AI | peachdeck |
|---|---|---|
| Instructions | Ad hoc, written by the user in each conversation | Fixed system prompt derived from a written scoring specification |
| Output format | Structured output or JSON schema, best-effort and can still fail to parse | Custom decoding grammar, output is parseable by construction |
| Consistency | Varies by phrasing, model version and session | Same rubric, same 70+ dimensions, every deck |
| Result | A paragraph of commentary | A structured score you can rank and compare |
| Model | Large general-purpose model rented through a third-party API | Purpose-built evaluation engine, self-hosted |
| Data residency | Processed on third-party infrastructure, typically in the United States | French servers, never leaves the European Union |
| Training on your deck | Depends on the provider and your account settings | Never, on any model |
Anything else, write to [email protected].