Intercom Fin — August 2026
Independent verification of outcome claims billed for the period, evaluated under two lenses against the customer's systems of record.
- Statement
- stm_01m2nszv8cy168grags7fbgspc
- Contract
- intercom_fin_resolution v1
- Engine
- 1.0.0
- Generated
- Sep 16, 2026, 19:09 UTC
- Run
- run_01m2nszqhx82eka01wz08tsr35
§1Summary of billed and verified outcomes
| Measure | Count | Amount (USD) | Share of billed |
|---|---|---|---|
| Billed by vendor | 4,892 | $4,843.08 | 100.0% |
| Verified — vendor lens | 4,267 | $4,224.33 | 87.2% |
| Verified — strict lens | 3,354 | $3,320.46 | 68.6% |
| Not verified — vendor lens | 514 | — | 10.5% |
| Not verified — strict lens | 1,243 | — | 25.4% |
| Unverifiable — vendor lens | 111 | — | 2.3% |
| Unverifiable — strict lens | 295 | — | 6.0% |
| Definition delta (billed − verified, strict) | 913 | $903.87 | 18.7% |
Unverifiable counts are reported separately and are not treated as findings against the vendor. Amounts for not-verified and unverifiable claims are included in the definition delta.
§2Breakdown of not-verified and unverifiable claims by reason
Each claim carries one primary reason code per lens.
- Transcript: customer indicated unresolved transcript_label:customer_indicated_unresolved360
A transcript-assist label (supporting evidence, reviewed by sampling) indicated the customer did not consider the issue resolved. Labels never produce a Verified verdict.
- Reopened after claim reopened_within303
The record was reopened within the window after the claim.
- Escalated to a human escalated_to_human_within193
The conversation was escalated or assigned to a human within the window.
- Refund or credit issued refund_or_credit_within141
A refund or credit was issued to the customer within the window.
- Excluded by contract excluded_by_contract91
The claim matches a contract exclusion and is not billable under this contract.
- Duplicate claim for subject duplicate_claim_for_subject80
Another claim for the same subject exists in the same billing period; only the earliest counts.
- Negative satisfaction rating csat_negative75
The customer left a negative satisfaction rating after the claim.
- No structured signal no_structured_signal184
The subject exists in the fact sources but none of the record types the definition depends on were observed, so the claim cannot be verified from systems of record.
- Subject not found in fact sources identity_unmatched111
The claim's subject could not be found in any connected fact source, so nothing can be verified either way.
Verified under the strict lens: 3,354 ($3,320.46).
§3Definition delta and alternative definitions
The definition delta is the difference between what was billed and what is verified under the strict lens: 913 claims, $903.87. The table below shows how the verified count moves under alternative definitions.
| Definition | Verified | Amount |
|---|---|---|
Vendor definition minus 'reopened after claim' Vendor-lens verified claims, excluding any where the record shows: The record was reopened within the window after the claim. | 4,267 | $4,224.33 |
Vendor definition minus 'escalated to a human' Vendor-lens verified claims, excluding any where the record shows: The conversation was escalated or assigned to a human within the window. | 4,074 | $4,033.26 |
Vendor definition minus 'human agent replied' Vendor-lens verified claims, excluding any where the record shows: A human agent replied within the window after the claim. | 4,074 | $4,033.26 |
Vendor definition minus 'refund or credit issued' Vendor-lens verified claims, excluding any where the record shows: A refund or credit was issued to the customer within the window. | 4,126 | $4,084.74 |
Vendor definition minus 'negative satisfaction rating' Vendor-lens verified claims, excluding any where the record shows: The customer left a negative satisfaction rating after the claim. | 4,192 | $4,150.08 |
Vendor definition minus 'duplicate claim for subject' Vendor-lens verified claims, excluding any where the record shows: Another claim for the same subject exists in the same billing period; only the earliest counts. | 4,267 | $4,224.33 |
§4Transcript assist disclosure
Transcript assist was used for this statement. A language model labeled a 5% sample of verified claims (544 labels). 0 labels were reviewed by a human; agreement was —.
Labels inform calibration statistics only. No verdict in this statement was changed by a model.
§5Scope
This statement covers billing period 2026-08 for vendor Intercom Fin, outcome type resolution, under Outcome Contract intercom_fin_resolution version 1, evaluated by engine version 1.0.0. 4892 claims were evaluated under two lenses.
§6Vendor definition (as published)
Source: https://fin.ai/help/en/articles/10772642-fin-ai-agent-resolutions (retrieved 2026-09-11). Notes: Also: https://www.intercom.com/pricing (retrieved 2026-09-11) and https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-outcomes (retrieved 2026-09-11). Published rules encoded: confirmed resolution; assumed resolution after 24h of customer disengagement; deduction when the customer returns seeking further assistance (even across billing periods); once per conversation; chat and email only; escalations triggered by default behaviour or workspace rules are not billed. INFERRED (bounded for computation, not stated by the vendor): the 'returns seeking further assistance' deduction is evaluated over 90 days after the claim; escalation is evaluated within the 24h assumed-resolution window.
- Verified if any of: customer confirmed resolved within 30d of the claim; no customer reply within 24h of the claim
- Not verified if any of: reopened within 90d of the claim; escalated to a human within 24h of the claim; a duplicate claim exists for the same subject in the period
- Unverifiable if any of: required fact source not connected: conversations
§7Strict definition (customer-side)
- Verified if all of: customer confirmed resolved within 30d of the claim
- Not verified if any of: reopened within 30d of the claim; escalated to a human within 7d of the claim; a human agent replied within 7d of the claim; a refund or credit was issued within 30d of the claim; customer left a negative satisfaction rating; a duplicate claim exists for the same subject in the period
- Unverifiable if any of: required fact source not connected: conversations; subject could not be matched in any fact source
- Override (supporting evidence only): transcript label is customer_indicated_unresolved, abandoned → NOT_VERIFIED
§8Evaluation order
For every claim and lens, in this fixed order: (1) contract exclusions → Not verified; (2) subject not found in any connected fact source → Unverifiable; (3) unverifiable conditions → Unverifiable; (4) not-verified conditions → Not verified; (5) verified conditions → Verified; (6) subject found but none of the record types the definition depends on were observed → Unverifiable; (7) otherwise → Not verified. Unverifiable is reported separately and is never re-labeled to improve a number for anyone.
§9Identity resolution
Claims are matched to systems of record by, in priority order: exact record id; vendor-provided external ids; HMAC-hashed email or phone; a time-window heuristic (lowest confidence, flagged in evidence). Customer identifiers are hashed with a per-tenant key before storage; message bodies are not stored.
§10Evidence integrity
Each verdict carries evidence pointers (source, record id, field, value, timestamp, SHA-256). Verdicts are hash-chained in a fixed order; the statement hash is the SHA-256 of the canonical statement JSON and appears on every page. Anyone holding the hash can confirm it at the public verification endpoint.
§11Transcript assist disclosure
Transcript assist was used for 544 claims that were Unverifiable under the strict lens solely because no structured fact existed. Labels are supporting evidence only: they can move a claim from Unverifiable to Not verified and can never produce Verified. Sample rate for human calibration: 5%; samples reviewed: 0; agreement rate: not yet measured. Overrides applied: 360. Model ids: heuristic-v1.
§12Exclusions
- Excluded: channel is one of voice, phone
§13Limitations
Verdicts describe what the connected systems of record show; they do not characterize intent. A Not verified verdict means the definition's conditions were not met in the records available; an Unverifiable verdict means the records needed to decide were not available.