A weekend ↓ architecture v1.0 · 2026-08-17

The warranty claim register: you honour the guarantee, and then you claim it back

Two clocks on one row - the customer's two-year legal guarantee, and the much shorter window your supplier gives you to recover what the faulty item cost you

Setup: A weekend (~8-14 h) Running cost: $0-20/mo (software included) Time saved: Medium After setup: — Data lives: —

BookkeepingDocuments & archiveDocument reading (OCR)Reminders

Built for small businesses in general — not a sector list.

Version history

v1.0 · 2026-08-17 — First release. Built around the gap between Directive (EU) 2019/771 Art. 10(1) - two years of seller liability to the consumer - and Art. 18, which grants a right of redress up the chain but leaves the deadline to national law and to your supplier's own terms. The register runs both clocks, assembles the claim pack from what the counter already captured, chases the supplier automatically, and matches incoming credit notes back to the row so 'agreed but never credited' stops being invisible.

There are two clocks on every faulty item that comes back over your counter. The first one is the law and you already run it. The second one is your money, it is shorter, and almost nobody runs it. This blueprint is one row that runs both.

The architecture

the same flow as text

      [ The counter ]  a faulty item, a serial number, a photo, and what it actually does

              ▼  typed once, in front of the customer, in under two minutes

      [ n8n — the claim desk ]

              ├─ searches Paperless-ngx for the serial → finds the delivery note it came in on

              └─ opens one row and starts two clocks

              ▼

      [ Grist — the claim register ]

              ├─ customer clock → the legal guarantee. non-negotiable. settled first, always.

              └─ supplier clock → their terms. days, not months. this is the one that leaks.

              ▼  claim pack assembled from what the counter already captured

      [ n8n — the chase ]  day 7 nudge · day 14 escalate · day 21 it is on your desk, named

              ▼

      [ Ollama ]  reads the supplier credit note → pulls claim reference and amount

              ├─ matches the row → closed, recovered

              └─ no match → agreed but never credited, the pile nobody has ever looked at

              ▼  monthly

      [ Grist — the one page that changes a decision ]

              recovered vs absorbed, per supplier and per product

      The customer clock is the law. The supplier clock is your money. Run both or you are only running the cost.

Four pieces, one machine. Paperless-ngx already holds your supplier invoices and delivery notes as searchable text, so the serial number is the key that links a return back to the purchase. Grist holds the register and the two clocks. n8n does the timing and the chasing, which is the entire job nobody has time for. Ollama reads credit notes. If you already built the receipt intake or the month-end pack from this library, Paperless-ngx and n8n are running and this is a weekend on top of them, not from scratch.

The problem

A customer walks in with a pump, a laptop charger, a pair of boots, fourteen months after you sold it. You repair it or you replace it, because you have to. In the EU the seller is liable for any lack of conformity that becomes apparent within two years of delivery, and you cannot point at the manufacturer and send the customer away. That is Directive (EU) 2019/771, Article 10(1), in force since 01/01/2022. You did the right thing and it cost you an item.

The same Directive gives you the other half. Article 18 says that where you are liable to the consumer because of something a person earlier in the chain did, you are entitled to pursue remedies against that person. The catch is in the second sentence: who you can pursue, and how, and inside what deadline, is determined by national law. In Belgium that runs through the hidden-defect articles of the old Civil Code (1641–1648), where a commercial seller is presumed to have known about the defect, but the buyer must act "within a short period" that a judge decides after the fact. In France the same article gives two years from discovery, which the Cour de cassation settled in 2023 as a limitation period with a twenty-year longstop from the sale. And on top of the law sits your supplier's own terms, which typically demand notification in days, with the serial number, the purchase reference and a photo.

So the loss is not the faulty item. The loss is the claim you were entitled to make and did not, because the evidence was on a phone, the delivery note was in a binder, and by the time anyone thought about it the window had closed. The industry name for this is warranty leakage, and the four ways it happens are always the same: the claim is settled for the customer and never forwarded, it is forwarded with incomplete documentation and rejected, it is forwarded and never chased, or it is agreed and the credit note never actually arrives. One vendor in this market states the split plainly: with a formal process a business recovers 50–70% of eligible costs, and without one many recover under 20%, or nothing. Treat those as vendor-published figures, because they are. The direction is not in doubt even if the decimals are.

The software sold to fix this starts at about $49/month at the entry tier and runs to $549/month for mid-market, with the per-seat helpdesks that get bent into the job charging around $55 per user per month. Vendor guides put the realistic mid-market band at $6,000–$25,000 a year. For a shop settling five or ten claims a month, every one of those prices is larger than the leak.

Tool choices — and why

Paperless-ngx

(GPL-3.0, 44.3k stars, commit 16/08/2026) is the evidence layer, and it is the piece that makes the claim survive contact with a supplier. A rejected claim is almost always a documentation problem: no purchase reference, no delivery date, no proof the item is even yours. Paperless OCRs every incoming supplier document and makes it full-text searchable, so a serial number typed at the counter returns the delivery note it arrived on. Without this you are looking through a binder while a customer waits, which is precisely the moment the claim gets abandoned.

Grist

(Apache-2.0, 11.5k stars, commit 15/08/2026) is the register. Two clocks on one row is a formula problem, and Grist's formula columns are Python, so "days until the supplier window closes" and "claimed but not credited after 60 days" are one readable line each rather than a spreadsheet formula nobody dares touch. Access rules per row matter here too: the person at the counter should be able to open a claim without seeing what every supplier has cost you this year.

n8n

(Sustainable Use License, free for internal business use, 200.9k stars, commit 16/08/2026) runs the clocks and the chase. This is the honest centre of the blueprint. Everything else is storage; the reason claims die is that chasing a supplier for €40 is never the most urgent thing on any given Tuesday. Automate the nudge and the escalation and the unit economics of a small claim change completely, because your cost of chasing goes to roughly zero.

Ollama

(MIT, 178.7k stars, commit 16/08/2026) does one bounded job: read the supplier's credit note and pull out the claim reference and the credited amount so it can be matched to the right row. That is it. It does not decide whether a claim is valid, it does not decide whether a fault is covered, and it never writes to the customer. Those are commercial and legal calls with a real downside for being confidently wrong. The extraction job is different: the answer is on the page, and if the model cannot find it the row stays open and a person looks. An unmatched credit note is a visible problem. A wrongly matched one is an invisible one, which is worse.

Monthly cost

Software is €0 for all four. On a machine you already own, running cost is electricity. Off-premises, a small VPS with 8 GB of RAM is roughly €10–20/month, and be honest that Ollama is the reason for the RAM. There is no per-seat fee and no per-claim fee, which matters because per-claim pricing is exactly what makes a €40 recovery not worth filing.

Setup outline

1. Read your three biggest suppliers' terms before you build anything. You are looking for one number in each: how many days you have to notify a defect, and what evidence they require. Write those numbers down with the date you read them. This is the whole design input, and it is different per supplier, which is why one global reminder does not work.
2. Ask your accountant or lawyer once, in writing, what your national recourse deadline actually is. Article 18 hands the question to national law and it is not the same in any two member states. Belgium's "short period" is decided by a judge after the fact; France gives two years from discovery. Put their answer in the register as a note with the date it was given.
3. Get supplier documents into Paperless-ngx first and confirm serial numbers are searchable. Test with ten real delivery notes. If your suppliers send serials only on a packing slip you throw away, fix that before writing a line of automation.
4. Build the register with the fields the claim needs and nothing more: item, serial, date sold, customer, supplier, supplier document reference, date the fault was reported, what the fault is, photo, what you did for the customer, cost to you, claim sent on, claim reference, credit received, amount. Fifteen columns settles a claim; the sixteenth is where these projects go to die.
5. Capture at the counter, in front of the customer, or it does not happen. Serial, photo, one plain sentence of what it does. If it takes more than two minutes it will be skipped on the busy days, which are the days with the most returns.
6. Settle the customer on the customer clock and never make them wait on your supplier. Their right is against you and it is not conditional on your recovery. Confusing those two is the single fastest way to turn a warranty claim into a consumer complaint.
7. Send the claim the same week, in the supplier's format. Assemble the pack automatically from what the counter already captured and let the last step be a human pressing send. Chasing costs nothing once automated; a rejected claim costs the whole amount.
8. Wire the chase with real steps: nudge at 7 days, escalate at 14, and at 21 it lands on a named person, not "the office". Include the claim reference in every message so their side can find it.
9. Close the loop on credit notes, because this is where the money actually gets lost. Match every incoming credit against an open claim. Anything agreed and not credited after 60 days goes on a list you read out loud.
10. Read the monthly page: recovered versus absorbed, per supplier and per product. That is the only output that changes a decision, and the decision is usually about which supplier you keep buying from.

Pitfalls — the real ones

The supplier window closes long before the customer window does

The customer has two years against you. Your supplier's terms may give you fifteen days from discovery of the defect. A register that only tracks the customer's guarantee looks complete and leaks continuously, and it will keep looking complete for years. Both clocks or do not bother.

Do not make the customer wait for your supplier

The consumer's right is against the seller, full stop. If you delay a repair while a claim is in flight you have converted a warranty case into a dispute, and in most member states you also have the burden of proof running against you for at least the first year under Article 11(1). Settle, then claim. In that order, every time.

The claim gets rejected on documentation, not on merit

Missing serial, no purchase reference, no photo, described in a way that does not match a failure mode they recognise. This is why capture happens at the counter with the item in front of you rather than three days later from memory.

Agreed is not credited

The most common late-stage leak is a supplier who accepts the claim and then never issues the credit note, or issues it against a different account. Nobody notices, because emotionally the claim was won. If you build only one alarm in this whole system, build this one.

Shelf time eats your own warranty

A part that sits in your stock for eight months and is then sold with a two-year guarantee to the customer may already be most of the way through the supplier's window. That gap is a real and permanent exposure on slow-moving stock, and the register makes it visible per product. It does not make it go away.

Be careful what you keep about the customer

A name, a contact, a purchase reference and a description of the fault is what a claim needs. Health detail, ID scans and anything else swept up "just in case" is data you now hold for no reason. Send the supplier the item detail and the fault, not your customer's file.

Business-to-business sales are a different regime entirely

Everything above about the two-year guarantee is consumer law. If you sell to other businesses, your terms and theirs govern, the presumptions do not apply, and the notification deadlines are usually much shorter. Keep the two apart in the register from day one, or you will apply the wrong clock to the wrong row.

National law decides the recourse, and this blueprint does not know your country

Article 18 exists in every member state and means something different in each. What the register does is keep the evidence and start the clocks. What the clocks are set to is a question you ask a professional once and hard-code, with the date you asked it.

Verified repos

Paperless-ngx — GPL-3.0, 44.3k stars, commit 16/08/2026
Grist — Apache-2.0, 11.5k stars, commit 15/08/2026
n8n — Sustainable Use License (free self-hosted internal use), 200.9k stars, commit 16/08/2026
Ollama — MIT, 178.7k stars, commit 16/08/2026

Sources for the claims above: Directive (EU) 2019/771, Articles 10(1) liability period, 11(1)–(2) burden of proof, 12 notification, and 18 right of redress — text via the European Commission and ECC-Net. Belgian hidden-defect recourse (old Civil Code Arts. 1641–1648, commercial seller presumed to know, claim "within a short period") per published Belgian product-liability practice notes; the French two-year limitation under Art. 1648 and its twenty-year longstop per the Cour de cassation chambre mixte, 2023. Recovery rates (50–70% with a formal process, under 20% without) and the warranty-leakage taxonomy are vendor-published figures from Claimlane, quoted as such. Software pricing from published 2026 vendor comparisons. National recourse deadlines are not covered here and are not the same in any two member states.

Get the next blueprint as it publishes — free, practical, verified tooling only.

Subscribe · Want it audited or extended? DM @RXed_EU