Verified AI Cost & Unit-Economics Report — synth-01

Period reconciled: 2026-06-01 to 2026-07-01 · Prepared for: synth-01

This report is an evidence pack of reconciled and verified AI-cost data — deterministic software output computed on company-provided inputs, prepared for and addressed to synth-01. It reflects reconciled historical data for the period 2026-06-01 to 2026-07-01: analysis of a closed, backward-looking window, not a forecast. [ Sample — in a delivered report, the client-specific legal framework appears here. Wording under counsel review. ] Full restricted-use and limitation-of-liability terms appear at the end of this report.

How to read this report

Two vocabularies recur below; read them once here.

Verification tiers — how firmly each dollar is established:
- T1 Reconciled — recomputed for each request and tied to the provider invoice; the firmest tier.
- T2 Invoice-verified — confirmed against invoice totals but not per request (for example, fixed subscription pools).
- T3 Management-declared — asserted by the company and not independently confirmed here.
- Uncovered — spend seen but not placeable in a tier; disclosed, never presented as declared.

Reconciliation quantities — how the invoice ties out:
- Coverage — the share of the invoice reconciled per request (the T1 share).
- Residual — the reconciliation error on tracked spend; near zero means tracked dollars tie to the invoice to the cent.
- Untracked — invoice spend the telemetry did not capture. Untracked is not a residual: a near-zero residual does not mean full coverage.

Inputs register and company assertion

[T1 RECONCILED] sources · [T3 MANAGEMENT-DECLARED] assertion. [EV-08]

Reconciliation draws on 3016 reconciled records (hashed for integrity) and effective-dated price tables. The company assertion (click-wrap) is captured in-product at delivery; in this synthetic demo it is a placeholder.

AI cost reconciles to the provider invoice to the cent; margin level warrants attention and vendor concentration warrants attention.

Reconciled AI cost of $9,716.21 ties to the provider invoice: $9,181.07 of tracked spend reconciles per request — the 94.49% coverage figure — while $283.14 (2.91%) is untracked spend the telemetry did not cover; the residual on tracked spend is $0.00 (cent-exact — a distinct quantity from untracked, not the same gap). Recoverable waste is estimated at $4,322.57/mo; the achievable-efficiency ratio (AER) — how close current spend already runs to the cost-optimal baseline — is 85.6%. Figures below carry a verification-tier badge. Each dimension measures a distinct thing — see the column — so a red margin and a green attribution score are not in conflict. [EV-01, EV-07]

Assessment (benchmark-relative index, never an absolute figure):

[T1 RECONCILED]

The method is deterministic reconciliation, not estimation

Per-request token counts are recomputed against effective-dated provider price tables and tied to the provider invoice total (F2). Zero LLM in the cost path. Verification tiers: T1 RECONCILED (per-request, tied to invoice), T2 INVOICE-VERIFIED (invoice totals), T3 MANAGEMENT-DECLARED (company-asserted), UNCOVERED (seen but not placeable — never folded into T3). Reconciliation tolerance is 2.0%. [EV-08]

Every reconciled dollar ties to the invoice, to the cent

The signature exhibit. [EV-01]

[T1 RECONCILED]

Three distinct quantities reconcile to the invoice total — do not conflate them:

1. Tracked spend, reconciled per request — $9,181.07, the 94.49% coverage figure, tied to the invoice at a residual of $0.00 (cent-exact).
2. Untracked spend — $283.14 (2.91%), invoice spend the telemetry did not cover (UNCOVERED tier).
3. Invoice-verified fixed pools — $252.00, subscription or committed lines not billed per request (T2 INVOICE-VERIFIED).

With $0.00 unexplained residual, these sum to the invoice total of $9,716.21. Residual and untracked are different things: the residual is the reconciliation error on tracked spend — it measures how exactly tracked dollars tie to the invoice, not how much of the invoice is covered. Untracked spend is invoice dollars the telemetry never saw. A small residual therefore does not imply full coverage; the uncovered remainder is disclosed above, not lost. Credits are shown as a separate line, never allocated per feature.

Naive-vs-reconciled variance drivers:

[T1 RECONCILED]

Cost attributes cleanly by feature and provider

Gross cost basis; credits never allocated per feature. [EV-03, EV-04]

By provider × model:

[T1 RECONCILED]

By feature (top):

[T1 RECONCILED]

Tag coverage 95.86%.

Unit economics carry a heavy-user margin tail

[T1 RECONCILED] cost · [T3 MANAGEMENT-DECLARED] revenue. [EV-04, EV-07]

Negative-margin cohorts: 1, totalling $423.59 of margin gap (shown separately, outside waste — a cost legitimately incurred). Unit-economics-quality index 80.9 (green). Revenue is management-declared (T3).

Reconciled cost trends across the period history; revenue remains a single period

Historical reconciled cost. No forward projection. [EV-01]

[T1 RECONCILED]

Revenue is available for a single period only; a margin trend requires two or more revenue periods (limitation disclosed).

Under price and load stress, costs move as the price table dictates

T1 rows are deterministic replays of the reconciled request set; T3 rows are labelled hypotheticals. The Basis column states each scenario's cost basis — most replay the reconciled usage, while credit-expiry moves the invoice from net-of-credit to gross (gross and net are never mixed unlabelled). Routing envelopes are mathematical bounds — the cost if the entire existing workload were forced onto a single model, assuming no quality constraint — not achievable configurations and not a recommendation; unlike a provider price change, which moves every request's actual price, an envelope changes no price and only re-routes hypothetical traffic. [EV-06]

[T1 RECONCILED + T3 HYPOTHETICAL]

No vendor credits in this period — credit-expiry exposure is nil.

Vendor concentration is a critical-path exposure

[T1 RECONCILED] [EV-03, EV-07]

The largest single line is anthropic claude-sonnet-4-6 at 77.04% of tracked spend. Vendor-concentration index 17.3 (red — warrants attention); index = 100 − largest-provider cost share %.

Pricing and contract risk (declared)

[T3 MANAGEMENT-DECLARED]

Flat-plan exposure and committed-spend / credit-expiry cliffs are company-declared items; the credit-expiry stress replay above quantifies the list-price downside. Provider contract terms are not connected in this report.

Seat-based AI tooling is not connected in this report

[UNCOVERED] — NOT COVERED IN THIS REPORT — data stream not connected. Seat/GPU/cloud connectors (secondary-spend ingestion) are out of scope for this phase. This is disclosed as context; it is outside the AI-spend coverage percentages above.

GPU / compute rental is not connected in this report

[UNCOVERED] — NOT COVERED IN THIS REPORT — data stream not connected. Invoice upload for GPU providers is out of scope for this phase. This is disclosed as context; it is outside the AI-spend coverage percentages above.

Cloud spend context is not connected in this report

[UNCOVERED] — NOT COVERED IN THIS REPORT — data stream not connected. The FOCUS 1.x cloud importer is out of scope for this phase. This is disclosed as context; it is outside the AI-spend coverage percentages above.

Coverage is disclosed by verification tier and accounts for all AI spend

Liability shield and credibility signal. [EV-02]

[ALL TIERS]

Bars sum to 100.0% of total AI spend. Unpriced and untracked spend is UNCOVERED, never presented as declared.

Identified efficiency opportunities, as estimated ranges

Reconciled-cost-derived observations — estimated ranges, not commitments. The Opportunity column is the plain remediation name from the library; the Ref column carries the detector and remediation codes for traceability. [EV-05]

Total identified opportunity: $2,515.32/mo across 9 items.

[T1 RECONCILED]

Evidence appendix

Every figure above traces to one of these evidence sources (EV-xx). A structured XLSX export accompanies this report.

[EVIDENCE INDEX]

Benchmark anchors (public sources, web-verified — see METHODOLOGY_SOURCES):

[PUBLIC BENCHMARK]

Restricted use and limitation of liability