Ask four questions instead of chasing one magic score
A large AI bill does not prove adoption is valuable, and a low bill does not prove a company is using AI well. Begin with four questions a non-technical leader can use: What did it cost? What business work did it support? What measurable result changed? Was the usage safe and reliable?
The answers can be compared as an evidence checklist. They should not be collapsed into a universal score that hides different goals, evidence quality, or operating risk.
Name one workload and one result
Start with a stable gateway workload such as a support assistant, document-review service, or coding workflow. Give the initiative a plain-language purpose and an accountable business role—not an individual employee performance target.
Choose one measure the business already understands: cases resolved, review time, defect escape rate, qualified opportunities, processing cost, or another bounded measure. Record the before-AI period, baseline value, optional target, and approved monthly AI spend.
Keep objectives, key results, and KPIs distinct
An objective states the business outcome the organization wants. A key result defines the measurable change that would demonstrate progress toward that objective. A KPI monitors ongoing performance; it may stand alone or support a key result, but it is not automatically an objective or proof that AI caused a result.
Map each initiative to a value-creation lever and choose the correct alignment. When several initiatives support the same key result, assign explicit attribution and keep their combined share at or below 100%. This prevents a portfolio report from counting the same result several times.
Keep the two evidence sources visible
AI Gateway HQ supplies gateway evidence: requests, observed provider cost, gateway fees, models, failures, fallback, budget use, risk signals, and safeguard interventions for the named workload and reporting window.
The company supplies the business result. A business owner can attest to it, an operator can identify a system export, or a future supported integration can supply it. The source-strength label remains visible beside the result. Prompts, responses, uploaded documents, customer records, and employee rankings do not belong in the value record.
Read the result as decision support, not causal proof
The report can show that AI activity and cost occurred during the same period as a measured business change. It can calculate the change from baseline, target status, indicative budget use, and—for suitable count measures with complete retained event coverage—observed AI cost for the outcome period per current unit.
It cannot prove that AI alone caused the result. Process changes, staffing, seasonality, demand, pricing, and measurement quality may also matter. Finance and the business owner still validate material savings, revenue, or accounting treatment.
Build the report around capital decisions
An investment committee or operating partner does not need a token dashboard. The review should begin with capital exposed, the business objective, the result observed, the strength of the evidence, and the decision required. Technical operating facts remain available as drill-down evidence when an exception needs investigation.
Use five report lines for each initiative: approved and observed spend; adoption in the intended workflow; the current business result against its baseline or key result; safety, reliability, and control exceptions; and management's next decision—scale, hold, change, or stop. A portfolio summary can aggregate readiness and exception counts, but it should not average unlike business outcomes into one opaque score.
Turn missing evidence into the next action
A useful system refuses to fill blanks with confidence. If spend exists without a result, ask the owner to record the period outcome. If an outcome is only attested, strengthen it with a system export or supported integration. If spend exceeds its approval, inspect the workload and enforce a hard gateway budget separately.
Corrections preserve history: an incorrect observation is voided with a reason and replaced. Measurement definitions, observations, and their audit evidence are committed atomically so a business record cannot exist without its corresponding audit event and delivery intent.
Use a portfolio rollup without pooling company data
Each company remains a separate tenant. Aggregate operations, provider-statement comparison, and outcome evidence are separately requested and approved by the company owner. Revoking the outcome scope removes its business measures from the next portfolio read.
An operating partner sees company-level counts of defined initiatives, results present, evidence readiness, attention states, and a bounded initiative summary. Provider keys, prompts, responses, customer identities, raw invoices, payment data, routes, policies, and employee-level activity remain outside the rollup.
Sources and attribution
PwC's 2026 framework evaluates value capture alongside strategy, investment, workforce, data and technology, governance and risk, and innovation maturity.
McKinsey's 2026 analysis distinguishes foundational productivity use cases from deeper revenue, operating-model, and exit-value opportunities.
BCG describes a balanced measurement approach that includes financial outcomes and forward indicators such as data readiness, API coverage, and pilot success.