Methodology & limitations

How Kehai works

Nahim Sayed, Founder · Updated 29 July 2026

An operator-led, engine-agnostic indirect-tax change-control workspace for cross-border businesses. It sits upstream of any tax engine, turning monitored VAT/GST developments into human-review-ready, business-specific decisions, owned actions, deadlines and audit evidence.

Kehai doesn't replace tax judgement. It gives that judgement an operating system.

Kehai (気配) is the Japanese word for a sensed sign that something is about to happen: the shift in the air before it rains. It's almost a literal description of regulatory horizon-scanning.

Human-review readyPrimary-source prioritisedSource-citedBuilt for triage and change control

Become a design partner See the live demo. No signup.

Sourced from official sources

For each jurisdiction you configure, Kehai searches for changes and prioritises official primary sources (tax authorities, official gazettes, the courts and official e-invoicing platforms) over newsletters and commentary, verifying a source as official by matching the authority's approved domain rather than running a dedicated crawler per authority. Every item is labelled ◆ primary or ◇ secondary, and secondary items carry a verification warning wherever they appear.

Matched to your business

Each change is assessed for relevance to your specific business and marked affects, maybe or no, with a plain-English reason, a suggested owner and deadline, and a forward time-to-act. It is presented as an AI suggestion until a reviewer in your workspace checks it, never as settled fact, so the same development can legitimately matter to one business and not another.

Cited, with human review built in

Every item carries its source, and material items are reviewed in your workspace before they are relied on. The distinct trust signals (official source, claim support, reachability, human review) are shown separately and never collapsed into a single tick, and an unverified link is never treated as proof. For EU law, where enabled, items can be cross-checked against the EUR-Lex register; for other jurisdictions a source earns 'official' status by matching the authority's approved domain rather than via a register.

Monitored with measured coverage

Monitoring runs continuously on a schedule, without manual intervention. Coverage is measured against the official sources you have configured, and any gap (a source not seen recently, or never reached) is surfaced rather than hidden, so you can see what is and is not being watched.

Audit-ready

Reviews, assessments and sign-offs are kept in an append-only audit trail, and any item can be exported as a self-contained evidence pack (the claim, its source and its review history) to hand to an auditor.

Local-language sources, handled honestly

Sources are monitored in their original language (French, German, Dutch and beyond) and Kehai does not require an English-language source to exist. The summary and business impact you read are AI-generated in English from that original: a working summary, not a certified or word-for-word translation, and labelled AI-generated until a reviewer in your workspace signs it off. When a source is checked, its original-language text is captured point-in-time, so a reviewer (and an auditor, via the evidence pack) can compare the English claim against the source's own words.

Sources watched for your footprint

What is watched, by type. Coverage is configured to your footprint; these are the kinds of sources behind it, with examples from the approved register.

Tax authorities and finance ministries

The primary watch: the authority and ministry sites for each jurisdiction you configure. Examples from the approved register: DGFiP, Bundesministerium der Finanzen, AEAT, HMRC, FTA, ZATCA, IRAS, LHDN, SARS.

Official gazettes and legislation portals

Official journals and consolidated law: EUR-Lex and the Official Journal for EU footprints, national gazettes elsewhere. An item's source domain must match this register to count as official.

Courts

The CJEU and national supreme and tax courts, for rulings that change how the rules read in practice.

E-invoicing and digital-reporting platforms

National mandate platforms and their technical specifications: KSeF, SdI, PPF, Verifactu, FATOORA and peers, per configured jurisdiction.

Secondary commentary

Adviser notes, newsletters and trade bodies can flag a development, but they carry a verification warning everywhere and are never treated as proof until matched to the primary source.

Coverage, cadence and escalation

Concretely: what is watched, how often, and how a source earns trust.

Update cadence

Monitoring runs on a continuous schedule, not on demand. The freshest developments can take a refresh cycle to surface; any configured source not seen recently is flagged rather than hidden.

Coverage benchmarking

Coverage is reported against the official sources you configure, so you can see what is and is not being watched, not an unverifiable claim of total global coverage.

Configured to your footprint

Kehai monitors the jurisdictions you configure (the markets that actually touch your business) and nothing you don't. Add or remove markets as your footprint changes; it is not a fixed list.

Human review is a built-in step, not an afterthought. Items stay labelled AI-generated until a reviewer in your workspace checks them, and every sign-off records who reviewed it and when, in an append-only audit trail. Kehai is built by an operator who has run multi-country indirect tax in-house, so the review workflow mirrors how the work is actually done.

A worked example

Fictional demo data

01Detected: impots.gouv.fr publishes a new PPF e-invoicing milestone. The source domain matches the approved register; the page is captured with a content hash.
02Drafted: the agent assesses the change per entity. For the French entity it affects AR data readiness, so it drafts an action with an owner and a 45-day deadline, labelled AI-generated.
03Signed off: a reviewer checks the original-language excerpt against the claim and approves. The sign-off lands in the append-only audit trail; the item now reads Human approved.

Read the full sample evidence pack

Known limitations

Coverage depends on what web search surfaces at run time: items can occasionally be missed, mis-dated or mis-classified, and the freshest developments may take a refresh cycle to appear. Coverage is measured against the configured scope, not the whole world. Relevance assessments are AI-generated and labelled as such until a reviewer signs them off. Kehai is a triage aid, not legal or tax advice: always confirm any item against the cited authority, and verify material conclusions with a qualified adviser, before acting on it.

What Kehai does not claim

Complete or guaranteed global coverage of every relevant development.
Legally verified conclusions or autonomous tax advice.
That a reachable source URL proves a particular claim.
That AI-labelled “primary” content is independently official until the domain is matched.
That an AI relevance assessment is correct until a reviewer has checked it.
A measured precision or recall figure for relevance: assessments are AI-generated and human-reviewed, not yet benchmarked against a labelled gold set.
That the English summary of a local-language source is a certified or word-for-word translation of it.
Enterprise-grade security or availability before those controls exist and are evidenced.

Talk it through

Leave your work email and I'll set up a consultation scoped to your jurisdictions, entities and flows.