For private-equity operating teams
Know what every company spends on AI—and what it can show for it.
Connect AI cost, governed activity, business results, and safety evidence across participating companies—without combining their prompts, keys, customer records, or local controls.
CompanyObserved costControlResult evidenceNext review
Atlas$18.6k94%82%Review ready
Harbor$14.2k88%61%Source follow-up
Juniper$21.4k76%74%Control review
Meridian$18.2k91%46%Result gap
One plain-language review: cost, work supported, measured result, and operating safety
Separate company approval for operations, provider-statement comparison, and business outcomes
A short evidence-based action queue while local management keeps control
The four-question review
Replace token totals with a management conversation.
Each answer carries its source and reporting boundary. A blank answer becomes the next action—not a fabricated score.
What did it cost?
Observed provider cost, gateway fees, approved spend, evidence coverage, and separately consented statement comparison.
What work did it support?
Named workloads, governed requests, model mix, owner role, and the business purpose the company selected.
What changed?
A before-AI baseline and latest period result, with the source strength visible and no automatic causation claim.
Was it responsible?
Budget discipline, failures, fallback, safety interventions, essential controls, and the evidence still missing.
A bounded 12-week operating cycle
Prove the review with two or three participating companies.
The 30-day example above is a reporting window inside this 12-week pilot. Before any connection, the parties document company owners, approved measures, source boundaries, responsibilities, success criteria, and commercial terms.
Agree on the questions and boundaries
Name the participating companies, executive recipients, four operating questions, consent scopes, owners, identity requirements, retention expectations, and stop conditions.
Connect evidence without pooling company data
Each company controls its own workspace, provider credentials, budgets, and approvals. The portfolio receives only separately approved operating, finance, and outcome measures.
Run the management review
Deliver the comparable operating view, company drill-down, action queue, and audience-specific executive report. Record decisions and evidence gaps for the next cycle.
Success is a repeatable decision process—not a manufactured ROI claim.
Define the pilot The cycle should identify who owns each material initiative, what was approved and observed, which results are supported, where controls or evidence are weak, and what management will change next.
In plain language
See what each company spent, what the spend supported, and what changed.
AI Gateway HQ connects four questions in one review: What did AI cost? What business work did it support? What measurable result changed? Was the usage safe and reliable? Gateway records cover the operating activity, while each company supplies the business baseline and result.
The portfolio outcomeOperating partners can find responsible operators, missing evidence, uncontrolled spend, and initiatives ready for a continue-change-stop conversation—while each company retains its own security, identity, provider, payment, and data boundaries.
For the operating team
Make the management conversation comparable, not the companies.
Every company is reviewed with the same evidence vocabulary, while its goals and baselines remain company-specific. The output is an action queue, never an employee leaderboard or universal AI score.
What administrators configure
Company permission
A company owner must approve aggregate reporting and can revoke it immediately. No portfolio view appears before that approval.
Separate outcome permission
Business measures require a second, explicit company-owner approval. Aggregate operating access alone reveals no baseline, current result, owner role, or evidence strength.
Comparable operating facts
Review observed cost and activity, budget discipline, reliability, model mix, and essential safeguards using the same definitions.
Company-specific results
Pair a named workload with its business purpose, accountable role, baseline, approved spend, latest result, and evidence source.
Actionable review
Surface missing limits, weak evidence, unexplained cost, failures, fallbacks, and missing safeguards as follow-up—not an automatic verdict or configuration change.
What happens to each request
- 01Observe
The gateway measures cost, requests, model choice, budget posture, failures, fallbacks, and safety interventions for named workloads.
- 02Define
A company names the result that matters, its owner, the before-AI baseline, target, evidence source, and approved monthly spend.
- 03Confirm
The owner or a connected system records a bounded period result; corrections are voided and replaced instead of erasing history.
- 04Review
The portfolio sees only separately approved rollups and a company-level queue showing what is ready, incomplete, or needs attention.
See the working control
Move from the portfolio signal to the evidence behind one company.
A full-page company view connects consented spend, named work, outcome evidence, operating controls, and explainable exceptions without exposing prompts, credentials, or individual activity.
- Rendered by the real customer console and control API
- Exercised with safe OpenAI- and Anthropic-compatible simulators
- Sanitization gate rejects credential-shaped values and private owner email
A full-page company view connects consented spend, named work, outcome evidence, operating controls, and explainable exceptions without exposing prompts, credentials, or individual activity.
Control surface
Useful on day one. Explainable on day one hundred.
Each control has an operating path, an owner, and evidence that can be reviewed without collecting prompt bodies by default.
A shared view without a shared data pool
Provider keys, prompts, responses, customer identities, payment details, routes, and policies stay inside each company account. The portfolio view receives only the approved aggregate measures needed for operating review.
Connect spend to a result without inventing ROI
The gateway supplies cost, activity, reliability, and control facts. A business owner supplies the outcome baseline and latest result. Source strength, reporting period, approved spend, and missing evidence remain visible so correlation never masquerades as causal proof.
Start with the decisions that need attention
Focus operating reviews on initiatives with spend but no result, weak evidence, exceeded approvals, missing budgets, unsafe usage, repeated failures, excessive fallback, or outcomes that do not support continued expansion.
Evidence, not assertions
Review every company on consistent terms.
The portfolio command center brings together operating activity, company-approved statement summaries, and business results. Reporting gaps remain visible, and each company retains control of what it shares.
Review security boundaries and current status- Company approval and revocation recorded for every reporting relationship
- Comparable activity, configured cost, budget, reliability, model mix, and essential safeguards
- Exact-versus-estimated cost coverage shown with the reporting window and event limit
- Separately consented business baselines, period results, accountable roles, and evidence-strength labels
- Outcome records with an audited correction history
- Prompts, responses, credentials, identities, payment details, routes, and policies excluded
- Executive follow-up queue linked to the company that owns the evidence and change
Questions teams ask
Know the boundary before you deploy.
Clear answers for buyers, administrators, and security reviewers.
Can a fund see every portfolio company's prompts?
No. The portfolio report contains approved operational aggregates. Provider keys, prompt and response bodies, customer identities, payment details, and local route or policy configuration are excluded.
Can a parent policy override a company's security rule?
Not in the current portfolio release. The portfolio view identifies gaps and supports operating review; company administrators retain control of local routes, budgets, and security policy.
How should external benchmarks appear in a portfolio review?
Only as separately labeled, reviewed context. Customer reporting must keep third-party research distinct from gateway-measured facts and company-supplied value evidence, with the source period and methodology clear wherever a benchmark is used.
Can it tell us whether AI caused a business result?
No. It correlates gateway-observed activity and spend with a company-supplied baseline and period result for the same named workload. The source and evidence strength remain visible, and the report explicitly does not infer causation.
What can a portfolio team evaluate now?
A portfolio team can use a guided multi-company example; review spend, activity, budgets, reliability, model mix, and safeguards after aggregate approval; and review separately approved business-measure rollups showing baselines, current results, evidence strength, and readiness.
Start safely
Request the briefingSee the portfolio view with realistic company data before connecting a company.
Use the no-login guided example to inspect the measures and company evidence, then request a focused briefing for your fund or operating team.