Compliance, not just classification

UK customs record-keeping, done by the tool — not the inbox.

HMRC requires you to keep classification records for at least four years. Most importers manage this through shared drives and email archives, which fall apart the moment anyone asks "why did we pick that code?" TariffFlow replaces that with a signed, searchable, standard-format record for every case — built into the classification flow so there's nothing extra to maintain.

Start a case Download an example Case Pack

What HMRC expects in a customs classification record

Five specific pieces, drawn directly from HMRC's classification guidance and the Customs (Import Duty) (EU Exit) Regulations 2018. A record that's missing any one of them is harder to defend under audit.

01
The commodity code declared
The 10-digit code exactly as stated on the CDS / CHIEF entry.
02
The basis for classification
Which heading text supported it, which chapter/section notes were consulted, which GRI rule was applied, and any binding rulings that supported the decision.
03
The product description used to decide
Text description, datasheet, specification sheet, or image that the classifier worked from — not just the purchase-order line.
04
The date the decision was made
Tariff code schedules change; HMRC looks at what was correct on the date of import. A dated record anchors this.
05
Who made the decision
Either a named individual, a documented internal process, or an external advisor. "The system" isn't enough — there needs to be a traceable actor.

What fails in the typical setup

The shared-drive + inbox approach breaks at the worst possible moment.

Four ways a manual system collapses under a real HMRC query, or a broker query, or a post-merger compliance review:

HMRC emails to query a classification on a shipment from 18 months ago. Nobody on the team remembers why that code was chosen.
The original datasheet was an email attachment that's been cycled out of the inbox retention window.
The customs broker and the in-house team picked different codes on two shipments of the same product, and nobody has a record of the reasoning either way.
HMRC requires the GRI rule that was applied. The team had a consultant advise once, but the reasoning was never written down.

The TariffFlow approach

Six things the record-keeping layer handles automatically.

Not bolt-on storage. Built into the classification flow. Every case you run produces a record, and the library is indexable from the first one.

One standard record per classification

Every classification produces the same 12-page Case Pack with the same seven sections in the same order. Compliance reviewers learn the structure once and scan any pack the same way.

Signed and verifiable

SHA-256 hash computed from the case envelope, printed on every page footer alongside a verify URL. Anyone — your team, your broker, HMRC — can confirm the PDF wasn't altered after export.

Searchable library

Find any historical classification by product name, code, date, or free-text search across descriptions. No more inbox archaeology.

Broker handoff in one link

Create a read-only share link per case. The broker sees the same pack without a TariffFlow account. Revoke when the shipment clears.

Retention handled

Paid-tier cases retained for as long as your subscription is active. Exceeds HMRC's 4-year minimum. Full JSON export or account-level delete any time.

Reuse for repeat shipments

Save products to your library. Reclassify in one click when specs change. The fact sheet is reusable so you're not copy-pasting datasheets for every PO.

Questions

Record-keeping FAQ for UK importers.

How accurate is TariffFlow, and how do you measure it?: Measured, not claimed — and on two fixtures, because they measure different things. The 'everyday goods' fixture is 500 plain-English merchant-voice product descriptions chapter-weighted to the top UK import categories (machinery, electronics, vehicles, plastics, apparel, food/drink, household goods). On this fixture, TariffFlow gets the correct 2-digit chapter on 95% of cases and the correct 4-digit heading on 91% — that's where duty rates and trade-agreement eligibility live, and it's the typical broker workload. The harder 'HMRC binding-ruling holdout' fixture takes real Advance Tariff Rulings and tests with the source ruling excluded from retrieval (leave-one-out): on cases marked 'strong evidence' (76% of all cases) the recommendation matches the held-out binding ruling 82% of the time on the exact 10-digit code, while the remaining 24% are routed to 'review recommended' and reach 21% accuracy — a 60.9-percentage-point triage separation that proves the model knows when to ask for human sign-off. Every recommendation cites the UK ATaR, EU BTI, US CBP CROSS, or statutory chapter/section note that supports it. The classifier draws on 138,000+ unified precedents across UK, EU, and US classification practice, fuses dense retrieval with BM25 + ATaR + EU BTI + HSEN-derived quotes via Reciprocal Rank Fusion, and runs a parallel rerank-and-arbitrate ensemble with deterministic legal verification (ATaR d<0.10 hard override, chapter-note exclusion checks, HMRC live-measures validation). Target: 95% heading accuracy and 90% exact-code accuracy on the strong-evidence subset. Measurements are re-run on every material classifier release.
How long does HMRC require me to keep customs records?: Four years from the end of the calendar year the goods were imported / exported, per the Customs (Import Duty) (EU Exit) Regulations 2018. Some categories (excise, specific authorisations like IP or End-Use) require six years. TariffFlow stores Case Packs indefinitely on paid tiers so the retention obligation is handled by default.
What does HMRC expect to see in a classification record?: HMRC's Notice 702 and the Classification of Goods guidance expect: (1) the commodity code declared, (2) the basis for the classification — heading text, chapter/section notes, GRI applied, any binding ruling consulted, (3) the product description and any datasheet used to reach the decision, (4) the date the decision was made, (5) who made it. A TariffFlow Case Pack captures all five in one PDF.
Is a saved ChatGPT / AI output enough?: No, and it'll be actively unhelpful if HMRC asks for the reasoning. A ChatGPT answer isn't dated, isn't signed, doesn't cite binding rulings, and the model can contradict itself between runs. A customs classification record needs to be reproducible and standalone. TariffFlow Case Packs include a SHA-256 integrity hash and a verify URL so any recipient can confirm the document wasn't edited.
Can I share a record with my broker or freight forwarder without giving them a login?: Yes. Every case has a read-only share link (/s/<token>). The broker sees the same pack, the same cited rulings, the same walk — no account required. Revoke the link any time from the case's Share panel.
What happens if HMRC changes a commodity code?: The tariff is re-ingested daily, and each Case Pack stamps the live HMRC measures fetched at classify time. If a code is retired or the duty rate changes, the historical pack still reflects what was correct on the date of import — which is what HMRC looks at in an audit. Re-run the classification for future shipments and save a fresh pack.
Does the Case Pack satisfy 'Binding Tariff Information' requirements?: No — a Case Pack is a CLASSIFICATION RECORD, not a binding ruling. If you need a ruling HMRC is legally committed to, apply for an Advance Tariff Ruling (ATaR) via gov.uk. A TariffFlow Case Pack cites any existing ATaRs that support your code, which is often all you need — but for novel, high-value, or legally contested items, apply for your own ATaR.

Not legal advice. HMRC's guidance is authoritative — these answers summarise TariffFlow's reading of it for UK importers. Verify against your own compliance framework before declaring.

Build a record on your first classification.

Free to try, no card required. Every case produces a signed Case Pack you keep whether you upgrade or not.

Classify a product See the Case Pack structure

UK customs record-keeping, done by the tool — not the inbox.

What HMRC expects in a customs classification record

The commodity code declared

The 10-digit code exactly as stated on the CDS / CHIEF entry.

The basis for classification

Which heading text supported it, which chapter/section notes were consulted, which GRI rule was applied, and any binding rulings that supported the decision.

The product description used to decide

Text description, datasheet, specification sheet, or image that the classifier worked from — not just the purchase-order line.

The date the decision was made

Tariff code schedules change; HMRC looks at what was correct on the date of import. A dated record anchors this.

Who made the decision

Either a named individual, a documented internal process, or an external advisor. "The system" isn't enough — there needs to be a traceable actor.

The shared-drive + inbox approach breaks at the worst possible moment.

Four ways a manual system collapses under a real HMRC query, or a broker query, or a post-merger compliance review:

HMRC emails to query a classification on a shipment from 18 months ago. Nobody on the team remembers why that code was chosen.

The original datasheet was an email attachment that's been cycled out of the inbox retention window.

The customs broker and the in-house team picked different codes on two shipments of the same product, and nobody has a record of the reasoning either way.

HMRC requires the GRI rule that was applied. The team had a consultant advise once, but the reasoning was never written down.

Six things the record-keeping layer handles automatically.

Not bolt-on storage. Built into the classification flow. Every case you run produces a record, and the library is indexable from the first one.

One standard record per classification

Every classification produces the same 12-page Case Pack with the same seven sections in the same order. Compliance reviewers learn the structure once and scan any pack the same way.

Signed and verifiable

SHA-256 hash computed from the case envelope, printed on every page footer alongside a verify URL. Anyone — your team, your broker, HMRC — can confirm the PDF wasn't altered after export.

Searchable library

Find any historical classification by product name, code, date, or free-text search across descriptions. No more inbox archaeology.

Broker handoff in one link

Create a read-only share link per case. The broker sees the same pack without a TariffFlow account. Revoke when the shipment clears.

Retention handled

Paid-tier cases retained for as long as your subscription is active. Exceeds HMRC's 4-year minimum. Full JSON export or account-level delete any time.

Reuse for repeat shipments

Save products to your library. Reclassify in one click when specs change. The fact sheet is reusable so you're not copy-pasting datasheets for every PO.

Record-keeping FAQ for UK importers.

How accurate is TariffFlow, and how do you measure it?

Measured, not claimed — and on two fixtures, because they measure different things. The 'everyday goods' fixture is 500 plain-English merchant-voice product descriptions chapter-weighted to the top UK import categories (machinery, electronics, vehicles, plastics, apparel, food/drink, household goods). On this fixture, TariffFlow gets the correct 2-digit chapter on 95% of cases and the correct 4-digit heading on 91% — that's where duty rates and trade-agreement eligibility live, and it's the typical broker workload. The harder 'HMRC binding-ruling holdout' fixture takes real Advance Tariff Rulings and tests with the source ruling excluded from retrieval (leave-one-out): on cases marked 'strong evidence' (76% of all cases) the recommendation matches the held-out binding ruling 82% of the time on the exact 10-digit code, while the remaining 24% are routed to 'review recommended' and reach 21% accuracy — a 60.9-percentage-point triage separation that proves the model knows when to ask for human sign-off. Every recommendation cites the UK ATaR, EU BTI, US CBP CROSS, or statutory chapter/section note that supports it. The classifier draws on 138,000+ unified precedents across UK, EU, and US classification practice, fuses dense retrieval with BM25 + ATaR + EU BTI + HSEN-derived quotes via Reciprocal Rank Fusion, and runs a parallel rerank-and-arbitrate ensemble with deterministic legal verification (ATaR d<0.10 hard override, chapter-note exclusion checks, HMRC live-measures validation). Target: 95% heading accuracy and 90% exact-code accuracy on the strong-evidence subset. Measurements are re-run on every material classifier release.

How long does HMRC require me to keep customs records?

Four years from the end of the calendar year the goods were imported / exported, per the Customs (Import Duty) (EU Exit) Regulations 2018. Some categories (excise, specific authorisations like IP or End-Use) require six years. TariffFlow stores Case Packs indefinitely on paid tiers so the retention obligation is handled by default.

What does HMRC expect to see in a classification record?

HMRC's Notice 702 and the Classification of Goods guidance expect: (1) the commodity code declared, (2) the basis for the classification — heading text, chapter/section notes, GRI applied, any binding ruling consulted, (3) the product description and any datasheet used to reach the decision, (4) the date the decision was made, (5) who made it. A TariffFlow Case Pack captures all five in one PDF.

Is a saved ChatGPT / AI output enough?

No, and it'll be actively unhelpful if HMRC asks for the reasoning. A ChatGPT answer isn't dated, isn't signed, doesn't cite binding rulings, and the model can contradict itself between runs. A customs classification record needs to be reproducible and standalone. TariffFlow Case Packs include a SHA-256 integrity hash and a verify URL so any recipient can confirm the document wasn't edited.

Can I share a record with my broker or freight forwarder without giving them a login?

Yes. Every case has a read-only share link (/s/<token>). The broker sees the same pack, the same cited rulings, the same walk — no account required. Revoke the link any time from the case's Share panel.

What happens if HMRC changes a commodity code?

The tariff is re-ingested daily, and each Case Pack stamps the live HMRC measures fetched at classify time. If a code is retired or the duty rate changes, the historical pack still reflects what was correct on the date of import — which is what HMRC looks at in an audit. Re-run the classification for future shipments and save a fresh pack.

Does the Case Pack satisfy 'Binding Tariff Information' requirements?

No — a Case Pack is a CLASSIFICATION RECORD, not a binding ruling. If you need a ruling HMRC is legally committed to, apply for an Advance Tariff Ruling (ATaR) via gov.uk. A TariffFlow Case Pack cites any existing ATaRs that support your code, which is often all you need — but for novel, high-value, or legally contested items, apply for your own ATaR.

Not legal advice. HMRC's guidance is authoritative — these answers summarise TariffFlow's reading of it for UK importers. Verify against your own compliance framework before declaring.