How Flaynn reads a pre-seed dossier

In brief

• Five pillars scored out of 20, one score out of 100.
• Published weights for each stage.
• A separate counter-analysis, with no access to the score.
• Information blocks carry their provenance, verified, declared or inferred, as defined by the Flaynn provenance grid.
• The documents to request and the questions to ask, block by block, are in the pre-seed pre-diligence checklist.
• A reading signal and an analysis priority, never a recommendation to invest.

The steps of a reading

1. Receipt of the documents and the answers. What is missing is flagged, not guessed.
2. External research on the market and the competition, frozen for the dossier.
3. Reconciliation with a register source when the company is identified by the company number entered in the form (SIREN in France) and its name in the register contains the declared name. Otherwise, its data remain to be confirmed.
4. Comparison of the document's figures against the register and the BODACC, line by line, by a fixed rule.
5. Scoring of the five pillars on a banded scoring scale, justified pillar by pillar.
6. Separate counter-analysis.
7. Calculation of the score and the reading level.
8. Base rates and provenance shown alongside, with no effect on the score.

The five pillars

Each pillar is scored out of 20 on the basis of fixed criteria.

How a pillar score is chosen

• A pillar's score starts from a band, 17 to 20, 13 to 16, 9 to 12, 5 to 8 or 0 to 4, set by the criteria demonstrated, and is then placed within that band.
• Each criterion is rated: not demonstrated, partially demonstrated, or demonstrated by verifiable evidence. Demonstrated by verifiable evidence describes a criterion in the scoring scale; it is not the Verified status of the provenance grid.
• Penalties remove no more than 12 points from a pillar, except for unreadable documents or a major contradiction between the form and the deck.

The weighting by stage

Each row adds up to 100. The reading thresholds do not change with the stage. Series B and beyond are read as Series A. An unrecognised stage is read as pre-seed.

The counter-analysis

After scoring, a separate counter-analysis re-reads the dossier to look for what weakens it. It receives neither the score nor the pillar scores: it reads the uploaded document when it is readable, together with the lines comparing the document with the register, the form answers, the external research and the stage. Its effect goes in one direction only: it can lower the score or leave it unchanged, never raise it, and the more severe it is, the larger the reduction. That reduction is capped by stage, and the cap widens with the stage: tightest at idea stage, widest at Series A. Its exact value is not published. A marked disagreement is shown next to the score, without changing it.

The confidence level

Each pillar carries a confidence level, high, medium or low, depending on the coverage and consistency of the documents. When three or more pillars have low confidence, the score loses 3 points and the reason is written into the reading.

Thresholds and reading labels

The score out of 100 gives the reading level. The thresholds are the same for every stage. The label says what can be read from the dossier, not the value of the company.

These four levels are a reading signal for the dossier, not an investment verdict.

Provenance

Three provenance labels say where a piece of information comes from. A status never moves up: declared or inferred information does not become verified.

Verified

Consistent with a register source. Never a banking, legal, accounting or human audit.

Declared

Entered in the submission form or written in the uploaded document, page cited, and taken as is, without cross-checking.

Inferred

Produced by the analysis engine, including a figure it spots in the deck without a cited page: a reading to confirm, never a fact.

Missing and inconsistent are not provenance labels but reading signals: data that is expected and not provided, or two elements of the dossier that contradict each other. The detail is set out in the Flaynn provenance grid.

The document against the register

Three figures read in the uploaded document are set against the official register and the capital notices published in the BODACC, line by line. The comparison follows the fixed rule described here: the analysis engine does not perform it. Each line quotes the document's value in quotation marks, with its page, then what the register publishes and the date it was consulted. What is in quotation marks is written by the company, never a finding by Flaynn. The lines appear on the dossier record and in the memo.

• Revenue. Compared with the accounts published in the register for the same fiscal year, of twelve months' duration. It is consistent within 5%; beyond that, the difference is quantified as a percentage.
• Headcount. Compared with the employee headcount band published by INSEE (Sirene directory), with its year: within the band, it is consistent; outside the band, it is a difference. This band describes employees at 31 December of a given year and is published up to two years later (Insee); a recent company may appear there at zero. A difference is therefore first of all a question of date and scope.
• Declared funding round. Capital notices are searched in the BODACC from 4 months before to 6 months after the period given by the document. The line says whether notices appeared in that window, with their date and link, and never says that a funding round is consistent: published capital does not measure an amount raised. The absence of a notice is not a contradiction either: a funding round can go through a holding company, BSA AIR or convertible bonds without changing the published capital.

For revenue and headcount, the result of the comparison is one of three: consistent, difference, not comparable. Without a confirmed link, there is no comparison result: the line is marked To reconcile.

When no conclusion is drawn

• To reconcile. The company's link to the register is not confirmed. It is confirmed only with a company number entered in the form (SIREN in France), a name in the register that contains the declared name, and the register of the company's country. A company found by its name alone, a SIREN picked up from the text of the dossier or an unavailable register leave the line marked To reconcile: it cites the document, says why, and says nothing about the company. The BODACC is then not read.
• Not comparable. The two figures have no common basis: the document does not date the fiscal year or does not state it in euros, the register does not publish the accounts for that fiscal year or keeps them confidential, the published fiscal year does not last twelve months, the document counts the founders whereas the register counts only employees, or the register does not publish a headcount.
• Funding round with no conclusion. The document does not date the funding round, the window is not yet closed or starts before the available notices, the BODACC is unavailable or its reading incomplete: the line says so, and no conclusion is drawn.
• A figure that the document presents as a forecast is never compared. Revenue is compared with the register of the company's country, France, Belgium or Luxembourg, when the register gives its amount for the fiscal year of the document; outside France, a missing amount is not commented on. Headcount and capital notices are read only in France: for a company registered in another country, no conclusion is drawn on headcount or the funding round.

A difference is a question for the call

A difference is neither a contradiction, nor a proven error, nor an opinion on the company. The scope may differ: group, subsidiary, consolidated accounts. The document may count founders, interns or contractors; the register counts employees. And the register contains only what has been filed there. A line that is consistent does not make the document's figure verified: it remains declared. The counter-analysis receives these same lines, with two rules: a difference is a question, never a contradiction; a line marked To reconcile says nothing about the company.

Base rates as context

Frequency observed in a class of companies comparable by stage and sector, calculated offline from a third-party funding-round database. Frozen at the time of the reading, after the score, they do not enter into it. A class that is too small is widened; this level of widening is displayed and is read before the figure. A base rate describes a population, never the future of a dossier.

Limitations

• No investment advice. No recommendation to invest; the decision remains yours.
• No audit. The reading replaces neither an accounting, legal or banking audit, nor your own checks.
• Two readings may differ slightly. Two readings of the same dossier do not always give an identical result; the fingerprint says whether they are comparable.
• No comparison with an internal database of dossiers.
• No founder dossier shown to an investor.

Version

Each reading records the fingerprint of the method that produced it: scoring scale, weightings, engine instructions, code version. Two readings can be compared only if their fingerprints are identical. The version shown at the top of the page, 1.2, is that of this document.

Questions about the method

Flaynn produces an informative reading of a dossier. It constitutes neither investment advice, nor a recommendation, nor legal, accounting, tax or financial advice. The decision remains that of the reader and their advisers.