Keep every answer current when policies and products change.

One version of every policy and product fact, each with an owner, and a change process that retires the old one everywhere.

Start with a support review How it works

QA-77Owner: CX lead

Six conversations quoted the retired returns policy.

Obsolete: RET-12 · 14-day window
Current: RET-13 · 30 days, from Oct 15
Where it lived: Help center, AI knowledge, agent macros
Fix: RET-12 retired in all three
Test added: A return on day 20 is allowed

The correction record: what was wrong, where it lived and the test that keeps it fixed.

Sound familiar?

The returns policy changed in October, and the AI still says 14 days.
Agents, macros and the help center each give a different answer.
Nobody can say which article an answer came from.

Every answer from a retired source is a promise you didn’t make, and the customer will hold you to it.

What changes.

TodayWith the engine

Source of truth: Articles, macros and memoryA register with an owner, a version and a date
A policy change: Edit the article, tell the teamRelease it, retire the old version, run the tests
Checking: Wait for a complaintReplay the questions that matter after every change
Evidence: “Which article was that?”Every answer shows its source and version

How it works. 6 steps, 4 stages.

AI2
Rules and automation3
Your team1
Approval

1Register
1. Rules and automation
  Register every source
  Each policy and fact gets an owner, a version and an effective date.
  - Gorgias
  - Shopify
2Change
1. AI
  Find where the old version lives
  When a policy changes, finds every article, macro and AI source that carries it, and drafts the updates.
2. Your teamApproval
  Approve the new policy
  Owners approve the policy and its wording. The AI never writes policy.
3. Rules and automation
  Retire the old version everywhere
  The retired source can no longer be cited by the AI, the help center or a macro.
3Test
1. Rules and automation
  Replay the test questions
  The answers that matter are checked again after every change.
4Watch
1. AI
  Flag answers that still use an old source
  Reads new conversations for citations of anything retired.

Each miss becomes a new test

What you get. Real work, every week.

Source registerPolicies and facts

Source	Version	Owner	Effective
RET-13 · Returns	30 days	CX lead	Oct 15
RET-12 · Returns	14 days	CX lead	Retired
CAN-3 · Cancellations	Before the pick	CX lead	Aug 1
CAT-042 · Product facts	Version 2	Merchandiser	Sep 24
EXC-4 · Exchanges	Same price, in stock	CX lead	Oct 15

The register: every source, its owner, its version and when it took effect.

Regression questions · returns

5 of 5checks pass after RET-13

“How long do I have to return a jacket?” answers 30 days
A return on day 20 is allowed
A return on day 35 goes to the team
The AI and the macros can’t cite RET-12
The help center matches the register

Replayed after every policy change

The questions that run after every change, so the old answer can’t come back.

How you’ll know it’s working.

Answers that cite a retired source
Zero is the goal, and anything above it is a defect with an owner.
Time from a policy change to every answer updated
How long the old version stays in circulation.
Repeat errors after a change
Whether the same mistake came back.

Baselines agreed before we build.

Works with what you have.

Support

Gorgias
Zendesk
Intercom

Commerce

Shopify

We configure first: Each helpdesk AI syncs and removes knowledge differently. Shopify’s AI staff member keeps its behavior separate from store knowledge, and Intercom’s Fin re-reads a website on a schedule, normally weekly. We learn how yours behaves first.
We build when: Policies change often, several tools hold answers, or you need proof that the old version is gone.

What a first build includes.

Built in your accounts, around your team. Scope, fee and baseline agreed before we start.

1Design
2Build
3Run it, or hand it over

Start with a review Scope this build

We build

A register of every policy and product fact, with owners and effective dates
A release process: approve, publish, retire
Regression questions that run after every change
A report of answers that cited a retired source

We start from

Help center, macros and AI knowledge sources
Policy documents and who owns them
Past conversations that quoted outdated answers

From your team

Owner: CX lead, with the policy owners
Involves: Customer service
Merchandising
Product
Legal

Questions teams ask.

Does the AI write our policies?

No. It finds where a policy appears and drafts the matching updates. Owners approve the policy and its wording.

Our helpdesk says it syncs automatically. Is that enough?

Sync schedules differ by tool, and removal doesn’t always carry across. Intercom, for example, documents that Fin’s website sync normally runs weekly and can pause after 90 days without use. We test that the old source is actually gone.

Who owns it once it’s running?

You do. It’s built in your accounts, documented, with training for the people who run it. Ownership usually sits with your CX lead, with the policy owners. We can keep running it with you, or step back.

How do we start?

Usually with a support review: two to four weeks, a fixed fee credited to the build, and a plan that’s yours. If the job is already clear, we can scope the build directly.