This comparison belongs to the complaint-to-product-change guide. It reads one Illustrative reply twice: first as an export row, then with everything around it restored.
Before: the February 17 reply as an export row
Illustrative. A fictional 220-person meal-kit subscription business exports its Intercom conversations while evaluating a new support tool. Subscribers have been writing in, confused about the deadline for skipping a week's delivery. One row, described at metadata level, looks like this:
| Field | What the export shows |
|---|---|
| Date | February 17 |
| Topic tag | Subscription changes |
| Customer message | Asks why a skipped box still shipped |
| Agent reply | Explains that skips must be made by Wednesday noon and links the help article on skipping |
| Article cited | The skipping article, as published that day |
| Status | Closed, rated helpful |
Read cold, this is a model answer. The agent was polite, quick, and pointed to the official article. Anyone scoring replies from the export would mark it correct, and a training set built from rows like it would teach a support system that quoting the help center is always safe.
There is no column for whether the article was current. Intercom could not know: the article and the rule that should govern it are separate records, and only one of them sits in this file.
After: the same reply with its context restored
Now set the row back inside its history. Three other records, kept in three places, change how it reads.
From Linear. One issue titled around the skip deadline carried links to 140 subscriber conversations from January 8 to February 12. On January 30 the product team decided to move the cutoff from Wednesday noon to Friday noon, and the change went live on February 9. Linear documents that its Intercom integration attaches conversations to issues as links and posts an internal note to linked conversations when an issue is completed, so both the grouping and the go-live date are on record.
From Notion. The internal subscription rules page was edited on February 11 to the Friday cutoff. Notion keeps page history for 7 days on Free, 30 on Plus, 90 on Business and without limit on Enterprise, so whether this edit can still be seen depends on the plan and on how long ago it was.
From Intercom article history. The skipping article was rewritten on February 25 to state the Friday cutoff. Intercom keeps a version history for public articles that shows which teammate changed what, and previous versions do not expire.
Together: the February 17 reply went out six days after the internal rule changed and eight days before the public article caught up. The agent quoted the help center faithfully, and the help center was stale. The explanation the subscriber received rested on a rule that no longer applied.
The gap table: what is lost and where it lives
Each row is a question a reviewer would put to the February 17 reply.
| Question about the reply | Intercom export alone | Where the missing context lives |
|---|---|---|
| Was this a known problem? | Not visible unless the agent mentioned it | Linear issue linked to 140 earlier conversations |
| Had a fix been decided? | Not visible | Linear issue comment dated January 30 |
| Was the fix live? | Not visible | Linear completion on February 9 and the internal note it posted |
| Which rule applied on February 17? | Not visible | Notion rules page version dated February 11, if history still reaches it |
| Was the cited article current? | Appears current | Intercom article version history showing the February 25 rewrite |
| Was the answer right? | Appears right | Only the joined timeline shows it was outdated |
The last row is why connected records beat a bigger export. Internal decisions tied to their consequences are rarely found in public data, a gap described in what data is missing from AI training.
Telling a process gap from an agent error
The two readings lead to opposite conclusions. The export says the agent did well. The connected history says the agent did well by the only rules available and the organization was slow: the cutoff moved internally on February 11 and publicly on February 25. That is a process gap with a measurable length of 14 days.
Records that separate those two situations are what an AI lab would want when training or evaluating a support assistant. Trained on the export alone, a model learns that quoting the article is enough. Evaluated against the joined history, it can be tested on whether it notices that a rule has moved.
For a referral partner, the practical question is how many of the company's rule changes can be read this way. One case is an anecdote. Several years of linked conversations, dated decisions and versioned articles is a record set. The record chain names the key that joins each step, using a different Illustrative case.
Describing the gap without opening a conversation
All of the above can be confirmed from settings and owners. Ask the support lead three things: is the Linear integration switched on, and since when; which Notion plan holds the internal rules pages; and are help articles edited in place with their history kept. No one needs to read a subscriber's message. The metadata-only Intercom inventory template gives a format for the answers, and complaint handling records covers what a complaint file holds by itself.
Before introducing, confirm the baseline: 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license the records and an authorized contact: the CEO, usually the owner in an owner-run company. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Sign the Channel Partner Agreement before submitting.
Unknown until SourceX reviews it: how much history can be joined for older periods, what subscriber terms allow, how messages would be de-identified, and whether a buyer wants this set. Partners never receive or forward records.
Next step
Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.
If your client's support archive can be read with its product decisions and rule versions beside it, register as a partner, check who qualifies and read the terms before making the introduction.