Find out what your AI spend actually produced.
Fixed-price AI efficiency audits and a published ratio — the Klade Ratio — for operational work. The pen test for your AI spend.
Fixed price, three to four weeks, deliver-and-exit. You create the read-only keys and can revoke them any time; we install nothing, see metadata only, and destroy your data at close.
From the sample engagement: $9,366 of measured waste against $60,130 of AI spend in a 90-day window. Demonstration data; Meridian Ridge Capital is fictional.
Companies can tell you their AI spend to the dollar. Almost none can tell you what it produced.
70–95%
of enterprises cannot demonstrate AI ROI.
MIT, PwC, Forrester, IBM, Gartner — five independent surveys.
#3
“Determining AI value/ROI” ranks third among AI-spend challenges.
State of FinOps 2026, n=1,192.
0
vendors in “AI value management,” the category Forrester named in June 2026.
Forrester, June 2026.
Boards have started asking for proof, and budget freezes follow when there isn’t any. Everyone is asking the same question — what did we get for it — and there is still no standard way to answer it.
How the audit works, week by week
-
Week 1
Intake and mapping
Structured interviews map your workflows to spend streams and to the systems that already count their output. You create read-only, scoped keys for your providers’ admin APIs: usage and cost endpoints only, metadata only. Interviews map; they never measure.
Deliverable: the workflow map, with anything that fails the qualification test named unvalued
-
Weeks 2 and 3
Measurement
Metered spend is joined to the counted outputs in your systems of record, over at least 60 days of overlapping telemetry. Tier 1 waste is arithmetic on observed telemetry. Benchmark comparisons disclose the size of the comparison set.
Deliverable: none yet. Findings are checked before they are shown.
-
Week 4
The memo and the dashboard
The audit memo leads with the waste ledger in dollars, each finding with its remediation path. Klade Ratio ranges follow per workflow, with the attribution method stated, then your position against the benchmark. Your sponsor dashboard renders from the same data file as the memo.
Deliverables: the audit memo and your firm’s dashboard
-
After
Monthly audits
The audit is fixed-price and ends at delivery. Monthly audits re-run the measurement and refresh the memo, the dashboard, and your benchmark position as the cross-company dataset grows.
Deliverable: the monthly memo, citing the methodology version it used
Every waste finding comes with its remediation path — a routing policy, a step budget, a workflow change. If you want help making those changes, that is a second, separately priced engagement, and it is optional. The ratio you get never depends on whether you buy it.
Every engagement runs inside Klade’s SOC 2-scoped environment. Nothing is installed in your environment; no agents on employee machines. Customer data is destroyed at engagement close, per contract.
What lands on your desk
Monday morning, the memo is in your inbox. Three numbers matter: what the firm spent, what part of it was measured waste, and what the measured workflows returned. Every other page exists so you can check those three.
Every engagement comes with your firm’s dashboard — the same views as the sample here, built from your data. When the engagement starts, you tell us who should see it: usually the CFO, the FinOps lead, and whoever owns the engagement internally. Individual usage appears as cost accounting only — who spent what, on which programs — the way any expense system reports it. Output, ratios, and scores exist only at the workflow level, so there is no performance view of any person to leak.
- Fully-loaded AI spend — metered inference plus seats, by provider and model tier, over the observation window.
- Spend by program and by seat — every subscription and API key, the weekly trend, and per-seat cost accounting.
- The Tier 1 waste ledger — each finding in dollars, with its mechanism and remediation status. Leads the view.
- Cost per counted output — per workflow, trended weekly, against your own baseline and the peer benchmark.
- Klade Ratio ranges — per workflow, with the attribution method and every assumption enumerated.
- Unvalued spend — what we measured but refused to value, listed plainly.
You will not find a single company-wide score anywhere in it. One blended number is easy to quote and impossible to audit, which is the problem we were hired to fix.
Walk through it yourself
This is the actual dashboard a sponsor gets, filled with fictional data: the waste ledger, the workflow table, the ratio ranges, the limitations. Nothing is gated.
Demonstration data — Meridian Ridge Capital is fictional. The sample audit report PDF is available on request: adam@kladeai.com.
The Klade Ratio
The unit of analysis is the workflow — never the prompt, the session, or the person. Three confidence tiers, reported separately, never blended into a single composite score.
| Tier | What it measures | How it is reported |
|---|---|---|
| Tier 1 — Hard waste | Observed dollars: wrong model tier for the task, redundant inference, runaway agents, idle seats. | Measured $. Arithmetic on observed telemetry — no attribution model. Leads every report. |
| Tier 2 — Efficiency vs. benchmark | Cost per counted output unit vs. your own baseline and/or peers in the same workflow class. No dollar value is assigned to the output. | Compared. Percentile or multiple vs. baseline, with the comparison set size disclosed. |
| Tier 3 — Value attribution | Market-priced output attributable to the workflow ÷ fully-loaded AI cost. Only where a system of record prices the output. | Always a range, never a point, with the attribution method stated. |
| Workflow | Tier | Klade Ratio | Attribution basis |
|---|---|---|---|
| Diligence memo drafting | Tier 3 | 3.2–4.6× | Pre-adoption baseline, trend-adjusted |
What we refuse to do
Every incumbent ROI number is proprietary and can’t be compared with any other. Ours is published and auditable, and it stays that way because of the things we won’t do.
-
No dollar values on uncounted work.
Strategy memos and dashboards are listed as unvalued, not guessed at.
-
No per-person scoring in the product.
Measurement is workflow-level by default. Per-employee detail exists only as a contractually gated enterprise SKU, with no compensation or termination use.
-
No reading prompt or completion content.
Metadata only: tokens, costs, timestamps, model tiers.
-
No client-side capture in customer engagements.
Nothing installed in customer environments; no agents on employee machines. Individual self-measurement is a separate, opt-in track — never employer-deployed.
-
No LLM-judged “output quality.”
And no self-reported hours saved in any ratio.
Who it’s for
We are sector-agnostic. The qualification test is a data condition, not an industry: is the output of this work already counted in a system of record?
Claims processed, deals screened, contracts reviewed, tickets closed, filings, billable hours by task code — most of the white-collar economy outside of code qualifies. If a workflow’s outputs aren’t counted anywhere, we will tell you that, and we won’t put a number on it.
Software engineering is excluded on purpose. Per-developer AI ROI is already a well-served category, and we don’t compete there.
Our methodology is versioned and will be published.
Every audit report cites the methodology version used. The document is written so that a skeptical CFO, auditor, or economist can pick it apart and find the limitations already stated. We wrote it to be attacked.
Not ready for an audit? Follow the ratio.
We publish the methodology and open the benchmark to new sectors as design-partner audits complete. Leave an email and you’ll hear when that happens. Nothing else is sent.