Zéro Écart
A fixed-price catch-up service for French small businesses that are months behind on bookkeeping: upload a bank export plus a pile of receipts, and receive a finished, importable accounting journal, renamed receipts, and an itemised list of exactly what could not be matched.
One product, one founder, thirty days. Everything below is built to be executed inside the contest window — the official rules still decide what counts.
Decide in 20 seconds
The facts that decide whether this is worth thirty days of yours.
Owner of a French TPE or micro-entreprise (1–10 people, no in-house accountant, consultant, agency or e-commerce seller) who is 6–18 months behind on bookkeeping and whose expert-comptable has just asked for the year's documents before the 31 December closing.
Operational simplicity 3.0/10
Remove regulated data, human review steps and custom infrastructure; keep the product self-serve and stateless on managed services.
Why this could work in 30 days: one founder can ship the MVP in 5 days; the plan targets a first sale on day 11; 26 sales at $39 clears $1,000 of collected revenue; strongest signal is speed to first revenue at 8.3/10.
Score and projections come from the generated report and are not guarantees. The $100K Hackathon is run by Ship or Die — read the official rules before you build.
Your first 24 hours
10 concrete actions. Tick them off as you go — nothing is saved or sent.
Product opportunity and positioning
Target customer
Owner of a French TPE or micro-entreprise (1–10 people, no in-house accountant, consultant, agency or e-commerce seller) who is 6–18 months behind on bookkeeping and whose expert-comptable has just asked for the year's documents before the 31 December closing.
Why now
The year-end closing is a dated, unavoidable trigger inside the contest window, and the deliverable is cheap to produce with managed OCR and no bank API, no licence and no procurement.
Problem
Bank transactions and receipts both exist but nobody has matched them; the accountant will not sort the shoebox without billing for it and the closing date is fixed.
Solution
One drag-and-drop upload of bank CSV/OFX plus receipt PDFs/photos; the pipeline extracts supplier, date, amount TTC and VAT, links receipts to bank lines with a confidence score, and returns a ZIP with journal.csv, a FEC file, renamed receipts, a Livre des Écarts PDF and a cover note for the accountant.
1. Product opportunity and positioning
The buyer is a French TPE owner who is nine months behind, not a finance department. She does not want a platform to learn, a chart of accounts to configure or a monthly subscription. She wants the pile gone before her expert-comptable's deadline, and she wants to stop paying someone to sort paper. Zéro Écart sells exactly that: one upload, one ZIP, one PDF that names the lines still unsupported.
The trigger is a dated request from the accountant — « il me faut les documents de l'année avant la clôture » — not a vague intention to be better organised. That matters commercially: the buyer is already under deadline, already has the bank export somewhere, and already has a receipt pile in a shoebox or a phone camera roll. There is no onboarding, no migration, no data import into a competing ledger. The output goes straight back to the accountant who asked for it.
Positioning is deliberately narrow and stays inside one deliverable. The pitch is a screenshot, not a deck: « 17 lignes sur 214 sans justificatif — voici lesquelles. » The Livre des Écarts is the product's proof of work and its marketing asset at the same time. It is also the honest boundary: it lists what could not be matched rather than pretending everything reconciled.
The competitive gap is real but thin, and the plan treats it as thin. Indy, Tiime, Pennylane and Dougs bundle receipt capture inside a subscription — but they require the owner to adopt the tool for the year, and they do not help someone who arrives in November with twelve months of backlog. A human accountant will do it, billed hourly, at ten to twenty times the price of a pack. Zéro Écart sits between: a fixed price for a fixed job, sold to the person who actually feels the pain.
The JEV scoring of this concept (band B, 65.7/100) is consistent with that reading. Speed to first revenue and market demand score well; monetization potential and — decisively — operational simplicity score badly (2.96/100). The plan below is shaped by that weakness rather than ignoring it: delivery is capped, manual QA is budgeted as the real constraint, and every step that could turn this into a services job is either automated or priced in.
The commercial risk is that the buyer simply hands the shoebox to the accountant anyway. The counter is timing and specificity: the offer arrives in the week the accountant asks, with a free one-month sample delivered in 24 hours, so the comparison is not "$299 versus free" but "$299 versus another three evenings of scanning".
Smallest sellable MVP
A page that accepts a bank CSV plus receipt files, runs extraction and matching, returns a redacted Livre des Écarts preview for free, and unlocks a ZIP (journal.csv, FEC, renamed receipts, full gap ledger, cover note) after a Stripe payment.
Must-have features
- Intake page with drag-and-drop upload for bank CSV/OFX plus receipts, and column-mapper presets for Qonto, Shine, Revolut Business, BNP and Crédit Agricole
- Receipt extraction via managed OCR returning supplier, date, amount_ttc, vat_amount, vat_rate and doc_type, with a retry queue that flags unreadable files instead of guessing
- Deterministic matcher: exact amount, then ±10-day window, then supplier token similarity, each link carrying a 0–1 confidence
- Journal export as journal.csv and a FEC-format file with editable keyword-based PCG account suggestions
- Livre des Écarts PDF: unmatched bank lines, orphan receipts, duplicates, personal-expense suspects, missing VAT lines, plus an accountant cover note
- Packaging and delivery: receipts renamed YYYY-MM-DD_SUPPLIER_AMOUNT_TVAxx.pdf, zipped, signed download URL plus email
- Paywall: redacted free preview showing counts and three masked rows, Stripe Checkout unlock, one-click refund
- Single admin page with job status, manual match override, re-run and refund
Explicitly out of scope
- No bank API aggregation (Plaid, Bridge, Powens) — upload only
- No accountant marketplace, matching or referral network
- No VAT filing to impots.gouv.fr and no tax or accounting advice
- No Factur-X or e-invoicing generation
- No multi-user workspace, client portal or accountant-facing role
- No mobile app
- No subscription billing
Demo flow (what the buyer sees)
- Owner drops a Qonto CSV and 40 receipt photos on the intake page
- Column mapper auto-detects date, label, debit and credit columns
- Extraction job returns structured fields per receipt and flags 3 unreadable ones
- Matcher links 37 receipts, leaves 17 bank lines unmatched with confidence scores
- Free preview shows counts and three redacted rows
- Owner pays $39/$179/$299 via Stripe Checkout
- Signed ZIP link and email arrive with journal.csv, FEC, renamed receipts, Livre des Écarts and cover note
Technical implementation plan
Stack
Architecture
The Next.js app writes uploaded files to a private Supabase Storage bucket and inserts a job row. A Vercel cron worker picks up queued jobs, calls the OCR API per document, normalises fields with one LLM call, then runs the deterministic matcher in plain TypeScript against the parsed bank lines. The matcher writes every link and every non-link to Postgres; pdf-lib renders the Livre des Écarts, a packaging step renames and zips the receipts, and a signed URL is emailed via Resend. Stripe Checkout flips the job from preview to full unlock; a webhook on the job row triggers the final packaging. Nothing is stateful on the server and no bank credentials are ever held.
Build steps
- Create Supabase project in EU region, storage bucket, jobs table and RLS policies
- Build intake page with dropzone, file-type validation and five bank column-mapper presets plus a generic fallback
- Implement the cron worker: claim job, run OCR per document, retry twice, mark unreadable
- Add the normalisation LLM call constrained to a JSON schema, with a low-confidence flag
- Write the matcher: exact amount, ±10-day window, supplier token similarity, confidence output, gap list
- Generate journal.csv and the FEC text file with editable keyword-to-PCG table
- Render the Livre des Écarts PDF and the accountant cover note with pdf-lib
- Build the renaming convention and ZIP packaging, store the ZIP privately, emit a 7-day signed URL
- Wire Stripe Checkout plus webhook unlock and a one-click refund endpoint
- Add the redacted preview (counts plus three masked rows) and the admin override page
- End-to-end test on a real anonymised 12-month export; measure field accuracy and time per pack
AI components
- Managed OCR (Mistral OCR or Mindee) for supplier, date, amount TTC, VAT amount and VAT rate
- One LLM call for field normalisation: canonical supplier names, doc_type classification, and PCG account suggestion from the keyword table
- Confidence flagging: anything below threshold is labelled « à vérifier » and pushed to the Livre des Écarts instead of being guessed
Pricing and monetization
Mois Unique
Entry offerFor Owner one month behind who wants to test output before committing
- One month journal.csv + FEC
- Renamed receipts
- Livre des Écarts for that month
- Cover note for the accountant
Rattrapage 6 mois
For Owner whose accountant just asked for the second half of the year
- Six months journal + FEC
- Renamed receipts
- One consolidated Livre des Écarts
- Personal-expense and duplicate report
Rattrapage 12 mois
For Owner facing a full year-end closing with nothing matched since January
- Twelve months journal + FEC
- Renamed receipts
- Consolidated Livre des Écarts
- Priority 24h turnaround
- One free re-run after corrections
Model: One-time fixed-price packs, paid upfront by card before the ZIP unlocks; no subscription.
First revenue offer: Free one-month pack, delivered in 24 hours, no card — then the $179 six-month or $299 twelve-month link in the same thread within 48 hours of delivery.
Expected ACV: $189
Upsell path: The free month converts to the 6-month pack as the default reply and the 12-month pack when the accountant explicitly asked for the full year; each paying customer earns a €50 credit for an introduction in the same group, which is the only referral mechanic.
First-customer acquisition
Direct replies and DMs in French micro-entrepreneur and TPE Facebook groups plus LinkedIn, triggered by public posts complaining about late bookkeeping or a document request from their accountant.
Where to find them
- French Facebook groups for micro-entrepreneurs and auto-entrepreneurs (5 groups, joined and read, not posted to on day 1)
- TPE/PME gestion and comptabilité Facebook groups
- Freelance and consultant Facebook groups (indépendants, freelances FR)
- LinkedIn hashtags #expertcomptable #TPE #microentreprise #bilan
- Public posts from the last 60 days containing retard de compta, bilan, relance de l'expert-comptable, liasse, clôture
Exact outreach motions
- Log 40+ dated complaint posts in a sheet with group, author, date and exact quote before writing any code
- Post the anonymised Livre des Écarts screenshot (counts visible, names redacted) with the line « 17 lignes sur 214 sans justificatif — voici lesquelles »
- Reply in-thread to every reaction with the free one-month offer; never pitch the paid pack in the first message
- DM the original poster with the same offer, referencing their own wording, one message, no attachment
- Deliver the free month within 24 hours, then send the paid link in the same thread within 48 hours
- Ask each paying customer for one introduction in the group and credit €50 to their next pack
- Repost a new gap-ledger screenshot every three days throughout the 30 days
Message template
Bonjour [prénom], j'ai vu votre post sur [groupe] au sujet du retard de compta. Je teste un service gratuitement cette semaine : vous m'envoyez l'export bancaire d'un mois (Qonto, Shine, BNP, Crédit Agricole — CSV ou OFX) plus les justificatifs, je vous renvoie sous 24h le journal, les justificatifs renommés et un PDF « Livre des Écarts » qui liste les lignes sans justificatif. Ce n'est pas un conseil comptable, c'est de la préparation de données que votre expert-comptable relit. Ça vous intéresse ?
30-day execution roadmap
Week 1 · Build & validate · 7 days
Infrastructure and demand evidence
- Create the Stripe live account, connect it to TrustMRR, complete identity verification, take no payments yet.
- Join 5 French Facebook groups and 2 LinkedIn hashtag feeds; log 40 recent complaint posts in a sheet.
- Register the domain and create Vercel plus Supabase EU projects.
Deliverable: Live Stripe/TrustMRR connection, domain, Supabase EU project, and a sheet of 40 dated pain posts.
Upload and bank parsing
- Build the dropzone page validating CSV/OFX and image/PDF receipts.
- Write column-mapper presets for Qonto, Shine, Revolut Business, BNP, Crédit Agricole plus generic fallback.
- Store files in a private bucket and insert a job row.
Deliverable: Working intake page that parses five bank formats into normalised lines.
Receipt fields
- Wire the managed OCR call per document with two retries.
- Constrain the normalisation LLM to a JSON schema and add the low-confidence flag.
- Run 20 real receipts through and record field accuracy.
Deliverable: Extraction job returning structured fields for 20 receipts with an accuracy log.
Links and gaps
- Implement exact-amount pass, ±10-day window pass, supplier token similarity pass.
- Emit a 0–1 confidence per link and a gap list for everything unlinked.
- Add duplicate and suspected-personal-expense detection.
Deliverable: Matcher producing links, confidences and a gap list on one real month.
Exports and packaging
- Generate journal.csv and the FEC text file with the editable keyword-to-PCG table.
- Render the Livre des Écarts PDF and the accountant cover note with pdf-lib.
- Rename receipts to the strict convention, zip everything, expose a signed URL.
Deliverable: End-to-end ZIP produced from the founder's own anonymised bank export.
Money path
- Add the redacted preview showing counts and three masked rows.
- Wire Stripe Checkout unlock, signed-URL email via Resend and the one-click refund.
- Publish privacy page, DPA and the 30-day deletion policy.
Deliverable: Live paid flow tested in Stripe test mode, then one live €1 self-purchase refunded.
First public artifact
- Produce a full Livre des Écarts from a real export and screenshot it with names redacted.
- Post it in 3 groups with the « 17 lignes sur 214 » line and no product pitch.
- Log every reaction and comment into the outreach sheet.
Deliverable: Two screenshots live and a ranked list of 20 warm contacts.
Week 2 · Launch & first sale · 7 days
Free samples
- DM the 20 warm contacts with the free one-month offer.
- Reply in-thread to every commenter within two hours.
- Qualify each reply: months behind, bank, accountant deadline.
Deliverable: 20 DMs sent, 8 qualified sample requests in the sheet.
Manual first packs
- Run 3 free one-month packs through the pipeline and hand-check every match.
- Note every failure mode in a fix list ranked by frequency.
- Deliver all 3 within 24 hours of receipt.
Deliverable: 3 free packs delivered and a ranked fix list.
Paid links
- Send the $179 or $299 link in each delivery thread within 48 hours.
- Fix the top three extraction failure modes.
- Post a second screenshot in two different groups.
Deliverable: 3 paid links sent and three fixes shipped.
Close and collect
- Follow up every thread once, at 24 hours, with one line only.
- Close the first paying customer and confirm the pack lands inside TrustMRR.
- Re-check that the delivered pack is importable by asking the buyer's accountant.
Deliverable: First verified payment collected and profit line recorded.
Refund prevention
- Audit the delivered pack against the accountant's feedback.
- Add the missing one-to-many matching rule for split payments.
- Tighten the « à vérifier » threshold.
Deliverable: Matcher patched and one documented accuracy improvement.
New groups
- Join 3 additional groups and log 40 more posts.
- Send 20 fresh DMs using the same template.
- Publish a third gap-ledger screenshot with different counts.
Deliverable: 40 new leads logged and 20 new DMs sent.
Week-two economics
- Compute real conversion: DMs to samples to paid.
- Kill the worst-performing group and double down on the best.
- Raise the 12-month pack price if two buyers paid without objection.
Deliverable: Written conversion table and a channel focus decision.
Week 3 · Repeat & systematize · 7 days
Six-month tier
- Add the 6-month recommendation to the delivery email automatically.
- Ship the re-run button so the free correction is self-serve.
- Deliver the day's packs.
Deliverable: Self-serve re-run shipped and upsell copy live in the delivery email.
Second surface
- Post the gap-ledger screenshot as a LinkedIn post with a plain caption.
- Comment on 15 posts using the hashtag feeds, no links.
- DM 10 commenters with the free-month offer.
Deliverable: LinkedIn live with 15 comments and 10 DMs.
Throughput
- Batch-process all pending jobs in one morning block.
- Cap intake at 3 new packs per day to protect turnaround.
- Publish the 24h turnaround cut-off date inside the product.
Deliverable: Backlog cleared and an explicit capacity cap published.
Introductions
- Email every paying customer asking for one introduction, with the €50 credit.
- Add the credit code mechanic to the admin page.
- Ship the duplicate-report fix.
Deliverable: Referral ask sent to all paying customers and credit codes issued.
Fresh numbers
- Post a new screenshot with real, current counts.
- Rewrite the post headline to the count that got most replies.
- Answer every comment within two hours.
Deliverable: One post published and a rewritten headline for reuse.
Unpaid previews
- Send a single follow-up to every unpaid preview, referencing their own gap count.
- Offer $39 Mois Unique as the lower-friction option.
- Log who declines and why.
Deliverable: Objection log with at least 5 real decline reasons.
Reality check
- Compare collected profit against the base projection.
- Decide whether to hold price or add a $99 mini-pack.
- Fix the single biggest delivery defect.
Deliverable: Documented midpoint decision on pricing and one defect fixed.
Week 4 · Scale what works · 7 days
Seasonal offer
- Publish a 12-month offer framed on the 31 December closing deadline.
- Email all previous free-sample recipients who did not convert.
- Post the offer in the three best-performing groups.
Deliverable: December-closing offer live and emailed to the full sample list.
Volume
- Send 25 DMs from the newest logged posts.
- Reply to every public comment on the offer post.
- Book delivery slots for the week.
Deliverable: 25 DMs sent and a filled delivery calendar.
Turnaround
- Deliver all queued packs inside 24 hours.
- Record time per pack to test the capacity assumption.
- Refund immediately on the first rejected pack.
Deliverable: All packs delivered and a time-per-pack number recorded.
Data handling
- Purge all source files from Storage for jobs older than 7 days.
- Confirm the deletion job runs from cron without manual action.
- Add the retention line to the confirmation email.
Deliverable: Automated deletion verified and retention statement in email.
Fourth screenshot
- Post a screenshot showing the share of lines successfully matched.
- Ask one paying customer for a one-line quote about the accountant's reaction.
- Publish the quote only if permission is written.
Deliverable: One post published and, if granted, one written quote.
Last full week
- Send 25 final DMs before the window narrows.
- Re-contact every unpaid preview with a deadline.
- Prepare a 48-hour fast lane for year-end buyers.
Deliverable: 50 touches logged and a fast-lane delivery option live.
Final volume
- Deliver every open pack.
- Close intake for anything that cannot be delivered before 16 December.
- Fix the top refund driver or stop selling that format.
Deliverable: All open packs delivered and intake limits published.
Days 29-30 · Consolidate · 2 days
Cash in
- Chase every unpaid invoice with one plain reminder.
- Confirm every payment is visible in TrustMRR.
- Refund any pack the buyer rejected, without argument.
Deliverable: Payment ledger reconciled against TrustMRR.
Reconcile and stop
- Export the final TrustMRR statement and compute net collected profit.
- Write the post-mortem: conversion by channel, defects, retention wins and losses.
- Publish a clear end-of-window notice on the site.
Deliverable: Final reconciliation, post-mortem document and a closing site notice.
Revenue and profit scenarios
How the numbers are defined
Hackathon profit — Collected revenue − refunds − chargebacks − payment processing fees − paid marketing spend.
Business profit — Hackathon profit − operating costs (hosting, domains, tooling, AI/API delivery costs).
The contest ranks entrants on hackathon profit only. Hosting, tooling and AI/API delivery costs are not deducted by the organisers, but each sale still has to cover them — so we show both. Definitions verified on 2026-11-15 against the official rules.
Stated assumptions
- Base case assumes 9 paid packs in 30 days from ~90 cold DMs and 4 public screenshot posts (Conservative 3, Upside 22).
- Blended ARPU reflects the mix of $39, $179 and $299 packs actually sold, discounted by any goodwill credits (Conservative $179, Base $189, Upside $214).
- Payment fees assumed at 4% of collected revenue, using Stripe Checkout in the EU with EUR/USD conversion included.
- Refunds modelled as whole packs: zero in the conservative case, one $179 pack in the base case, one $299 pack in the upside case.
- No chargebacks modelled; every case assumes card payments collected upfront and a human answering disputes.
- Ad spend is zero in conservative and base — the channel is organic DM and group posting. Upside includes $200 of paid boosts on the best-performing posts.
- Operating costs cover Vercel, Supabase EU, domain, Resend and OCR/LLM delivery at roughly $2–$3 per pack plus re-runs.
- 70 packs in 30 days from the original concept math is treated as upside-adjacent and not modelled here; delivery capacity is capped at 3 packs per day.
- All figures are projections built from the listed assumptions, not verified revenue, customers or traction.
Conservative
3 customers- Refunds
- $0
- Chargebacks
- $0
- Payment fees
- $21
- Paid marketing
- $0
- Hackathon profit
- $516
- Operating costs
- $45
- Business profit
- $471
Base case
9 customers- Refunds
- $179
- Chargebacks
- $0
- Payment fees
- $68
- Paid marketing
- $0
- Hackathon profit
- $1,454
- Operating costs
- $120
- Business profit
- $1,334
Optimistic
22 customers- Refunds
- $299
- Chargebacks
- $0
- Payment fees
- $188
- Paid marketing
- $200
- Hackathon profit
- $4,021
- Operating costs
- $280
- Business profit
- $3,741
Model your own numbers
Runs in your browser · nothing is saved- Collected revenue
- $351
- Payment fees (3.9%)
- $14
- Paid marketing
- $0
- Hackathon profit
- $337
- Business profit
- $217
Hypothetical arithmetic, not a forecast. The fee rate is a flat 3.9% assumption, refunds and chargebacks are left at zero here, and none of this is tracked data.
All three scenarios are projections derived from the stated assumptions. They are not verified revenue, existing customers, traction or a forecast of results.
Main risks and validation experiments
Risk 1Regulatory framing: tenue de comptabilité is open but présentation des comptes is reserved to experts-comptables, and the Ordre can complain.
Never use tenue de comptabilité or bilan in copy; always « préparation de données relue par votre expert-comptable »; no filing, no liasse, no signature, no PCG enforcement.
Post the exact sales message in two groups and count accountant or admin objections in the replies.
Two written objections from qualified accountants in the first 10 posts means the wording must be rebuilt before any further outreach.
Risk 2GDPR exposure: bank statements and receipts are financial personal data and a single leak destroys the brand.
EU-only Supabase region, private buckets, signed URLs expiring in 7 days, encryption at rest, deletion cron at 7 days, published DPA and retention policy.
Run the deletion cron in staging and confirm files are gone from Storage and metadata purged from Postgres.
If deletion cannot be automated before day 6, the launch slips — manual retention is not acceptable.
Risk 3OCR failure on thermal receipts produces a pack the accountant rejects, driving refunds and killing the referral loop.
Low-confidence receipts are flagged « à vérifier » and written to the Livre des Écarts rather than guessed, so the pack is honest and importable.
Run 20 real thermal receipts and measure field accuracy; anything under 90% on amount and date goes to manual review.
If more than 25% of documents in the first 5 packs are unreadable, restrict intake to PDF invoices only.
Risk 4Timing: the contest window closes 16 December, before the January closing crunch, so urgency may arrive too late.
Frame every message on the 31 December closing and the December accountant document request, not on January.
Post one December-deadline offer and one generic offer in different groups; compare reply rates.
If December framing gets no replies in 40 DMs, the demand thesis is unproven and outreach stops.
Risk 5Channel throttling: group admins mark repeated screenshots as spam and the only acquisition channel collapses.
Post once every three days per group, always with a different artifact and a real count; DM in-thread rather than opening cold.
Track post reach and removal events group by group for the first two weeks.
Two removals in the same group means that group is dropped permanently.
Risk 6Solo capacity: manual QA on each pack turns a product into a job and caps volume below the projection.
Cap intake at 3 packs per day, publish the turnaround cut-off, and keep the admin override limited to a single page.
Time every pack delivered in week two and compute minutes per pack.
If a pack takes over 90 minutes of manual work, cut the 12-month tier and sell only 1-month and 6-month packs.
Hour-by-hour: the first 24 hours
Hours 1-2
- Create and verify the Stripe live account, connect it to TrustMRR, take no payments.
- Register the domain and open Vercel and Supabase EU projects.
Output: Paid and hosting infrastructure ready, zero customers taken.
Hours 3-5
- Join 5 French Facebook groups and 2 LinkedIn hashtag feeds.
- Log 40 dated posts mentioning retard de compta, bilan, relance de l'expert-comptable or liasse.
Output: Leads sheet with 40 real posts, author, group and date.
Hours 6-8
- Build the intake dropzone and the bank column-mapper presets.
- Store uploads in the private Supabase bucket and insert a job row.
Output: Working upload page storing files and creating jobs.
Hours 9-12
- Wire the OCR call with a retry queue and a low-confidence flag.
- Run 10 sample receipts through and record field accuracy.
Output: Extraction job plus an accuracy log on 10 real documents.
Hours 13-16
- Write the exact-and-window matcher and emit confidence scores.
- Draft the French sales message and the « 17 lignes sur 214 » post caption.
Output: Matcher skeleton and ready-to-send French copy for day 7.
Full 30-day launch report
Read the full generated report
1. Product opportunity and positioning
The buyer is a French TPE owner who is nine months behind, not a finance department. She does not want a platform to learn or a chart of accounts to configure. She wants the pile gone before her expert-comptable's deadline, and she wants to stop paying an hourly rate for someone to sort paper.
The trigger is dated: the accountant asks for the year's documents before the 31 December closing. That is why this sells inside a 30-day window. The bank export already exists, the receipts already exist as photos, and the only missing step is the matching nobody has done.
Positioning stays inside one deliverable. The pitch is a screenshot, not a deck: « 17 lignes sur 214 sans justificatif — voici lesquelles. » The Livre des Écarts doubles as the product's proof of work and its marketing asset, and it is honest by construction — it lists what could not be matched rather than pretending everything reconciled.
The JEV score is band B at 65.7/100. Speed to first revenue and demand score well; operational simplicity scores 2.96/100, the clear weak point. This plan answers that weakness by capping intake at three packs a day and keeping every step machine-driven, rather than pretending the product is stateless when it is not.
2. Smallest sellable MVP
Upload a bank export and a set of receipts. Return a redacted preview of the gap ledger for free. Unlock a ZIP after payment containing journal.csv, a FEC-format file, receipts renamed to a strict convention, the full Livre des Écarts PDF and a one-page cover note addressed to the accountant. Nothing else ships before day 6.
3. Technical implementation plan
Next.js on Vercel holds the intake and admin pages. Files land in a private Supabase Storage bucket in the EU region, and a job row is inserted. A Vercel cron worker claims jobs, calls managed OCR per document, normalises fields through one constrained LLM call, then runs a deterministic matcher in plain TypeScript. The matcher writes links with confidence scores and writes everything unlinked to the gap list. pdf-lib renders the ledger and the cover note, a packaging step renames and zips the receipts, and Resend emails a seven-day signed URL. Stripe Checkout flips the job from preview to unlock.
The hardest technical risk is one-to-many matching: a single bank line paying for three receipts, or one receipt covering two lines. Naive amount matching produces false confidence there, and false confidence produces a rejected pack and a refund. The mitigation is to flag splits explicitly and to push them to the Livre des Écarts rather than guessing.
4. Pricing and monetization
Three one-time packs: Mois Unique at $39, Rattrapage 6 mois at $179, Rattrapage 12 mois at $299 with priority turnaround and one free re-run. No subscription, no seat pricing, no trial. Payment is taken before the ZIP unlocks, which keeps cash collection inside the contest window and avoids chasing invoices.
The first revenue offer is always the free one-month pack, delivered within 24 hours, no card. The paid link follows in the same thread within 48 hours, and the 6-month pack is the default recommendation. A €50 credit for an introduction in the same group is the only referral mechanic.
5. First-customer acquisition
One channel only: direct replies and DMs in French micro-entrepreneur and TPE Facebook groups, plus LinkedIn hashtag feeds. Every message is triggered by a public post that already says the owner is late, so there is no cold education.
A single sales message, in French, referencing the person's own words, offering one free month, stating explicitly that this is data preparation the accountant reviews. A new gap-ledger screenshot every three days. No paid ads in the base case; $200 of boosts only if the upside case materialises.
6. 30-day execution roadmap
Days 1–6 build the thing: infrastructure and lead logging on day 1, intake and bank presets on day 2, extraction on day 3, matching on day 4, exports and packaging on day 5, paywall and live money path on day 6. No selling before the pipeline produces a real ZIP.
Days 7–14 sell the first packs. Day 7 publishes the first screenshot, day 8 sends 20 DMs, day 9 delivers three free months by hand, day 10 sends paid links, day 11 targets the first collected payment. Days 12–14 fix the top extraction defect, open three new groups and compute real DM-to-paid conversion.
Days 15–21 convert and harden: self-serve re-run, LinkedIn as a second surface, an explicit three-per-day intake cap, referral credits, and a midpoint decision on price.
Days 22–30 are the December closing push and the close-out: a 12-month offer framed on the 31 December deadline, 50 more touches, automated file deletion verified, final collection, and a reconciliation against TrustMRR on day 30.
7. Revenue and profit scenarios
All three scenarios are projections from the stated assumptions — not verified revenue, not existing customers. Conservative: 3 packs at a $179 blended ARPU, $537 gross. Base: 9 packs at $189, $1,701 gross, one $179 refund, $68 in payment fees at a 4% Stripe assumption, $120 of operating cost. Upside: 22 packs at $214, $4,708 gross, one $299 refund, $188 in fees, $200 of paid boosts, $280 of operating cost.
Hackathon profit excludes hosting, tooling and AI delivery; business profit subtracts them. At the base case the difference is $120 — small, but it is the number that decides whether this is a business or a hustle. The concept's original 70-pack / $20,930 projection is not modelled here because three packs a day with any manual QA is the real ceiling for one person.
8. Main risks and validation experiments
Regulatory framing is the sharpest risk. The copy must never say tenue de comptabilité or bilan; it says data preparation, reviewed by your expert-comptable. Test: post the exact message and count qualified objections.
GDPR is the reputational risk. EU region, private buckets, seven-day signed URLs and an automated deletion cron that must be verified working before day 6.
OCR failure on thermal receipts is the refund risk. Anything below confidence threshold is flagged « à vérifier » and lands in the ledger, so the pack is honest rather than wrong.
Timing is the demand risk: the window closes 16 December, before the January crunch, so all urgency is framed on the December accountant request.
Channel throttling and solo capacity are the operational risks: three posts per three days per group, an intake cap of three packs a day, and a hard rule to drop the 12-month tier if a pack takes more than 90 minutes.
9. First 24-hour action plan
Hours 1–2: Stripe live account connected to TrustMRR, domain registered, Vercel and Supabase EU projects open. Hours 3–5: five groups joined, forty dated pain posts logged. Hours 6–8: intake page and bank presets live, files stored privately. Hours 9–12: OCR wired with retry and confidence flag, ten receipts processed and accuracy recorded. Hours 13–16: matcher skeleton and the French sales message drafted. Nothing is posted publicly on day 1 — the evidence log comes first.
This brief was generated from public data about established startups, then narrowed and scored for a solo 30-day launch. Revenue and profit figures are scenario models, not verified results, guarantees or evidence of existing customers. Validate demand yourself before you spend a month on it — and check the contest rules for what counts toward your score.
Related ideas
All ideas →Chase Ladder
Turns a freelancer or small agency's overdue invoice list into a dated, escalating chase schedule with ready-to-send email drafts and a calendar file they can run themselves.
Bridge Brief
Turns a founder's own LinkedIn connections export into a ranked warm-intro map against a specific target list, with a ready-to-send ask for every path.
Cleanup Scope Report
Upload a prospect's QuickBooks exports and get a white-labeled PDF that lists every books issue, its dollar exposure, estimated cleanup hours and a defensible fixed fee — in under five minutes.