What is the first problem?
Not governance software. Visibility and decision evidence.
09 / 09 · Enterprise AI Platform & Economics Product
Aperture AI
AI spread faster than visibility.
A company buys copilots, connects model APIs, launches internal agents and lets teams experiment. Six months later, leadership cannot answer what AI exists, who owns it, what it costs, how good it is, what data it touches, or which investments deserve to scale. Aperture turns that fragmented estate into governed decisions.
ENTERPRISE PLATFORM + ECONOMICS
How does an enterprise decide which AI investments deserve to scale?
This case starts before the UI: estate discovery, outcome contracts, governance, cost attribution and change management.
01 / THE PRODUCT PROBLEM
Start with the decision the user cannot make well today.
A company buys copilots, connects model APIs, launches internal agents and lets teams experiment. Six months later, leadership cannot answer what AI exists, who owns it, what it costs, how good it is, what data it touches, or which investments deserve to scale. Aperture turns that fragmented estate into governed decisions.
The portfolio test: a client should understand the human problem before seeing a model, agent or framework.
02 / HOW RIKA THINKS
From ambiguity to a decision system.
The five-step reasoning pattern stays consistent; the actual product logic is specific to this problem.
What must leadership know?
Asset, owner, model, vendor, cost, quality, risk, data and outcome.
What looks like value but is not?
Seats, calls, runs and tokens prove activity, not ROI.
What should the platform create?
An AI estate graph + Outcome Contracts + governed decision queue.
What should every decision produce?
Evidence that the asset improved, consolidated, failed or should be retired.
03 / TRANSFERABLE PROBLEMS
The pattern travels. The constraints change the solution.
This is how I would reuse the thinking without copying the product.
Cloud FinOps
Shared visibility/cost optimization pattern; AI adds quality, autonomy and model/data risk.
SaaS management
Shared tool inventory and consolidation pattern; AI usage can be API-embedded and agentic rather than seat-based.
Security asset management
Shared ownership/risk inventory pattern; AI adds model/provider/eval and prompt/tool surfaces.
Model risk management
Shared validation/governance pattern; Aperture extends it to business outcomes, spend and operational ownership.
04 / USERS + BUYER
Who receives value, who operates it, and who pays?
AI products often fail when one persona is treated as ‘the user.’
Head of AI / AI platform leader
Needs one estate view and a prioritized decision queue.
Security, privacy, legal, risk and procurement
Need policy evidence, data classification, vendor context and audit trails.
Use-case owner + CFO/FinOps
Need to know whether usage translates to measurable outcome and whether spend should scale.
JOBS TO BE DONE
Functional, emotional and social value.
Discover AI assets, reconcile spend/quality/risk/outcomes and create governed scale/fix/stop decisions.
Replace uncertainty and board-level ‘AI spend anxiety’ with inspectable evidence.
Let business, finance, security and AI teams argue from the same underlying asset state.
CUSTOMER JOURNEY
The question changes as evidence arrives.
Connect
Integrate model providers, SSO, finance, tracing/evals and BI.
Establish read-only inventory first.
Discover
Normalize AI assets and owners.
Create estate graph.
Assess
Join cost, quality, risk and Outcome Contract evidence.
Expose gaps and duplication.
Decide
Prioritize scale / fix / consolidate / stop.
Human approval records rationale.
Review
Recheck after decision window.
Outcome evidence changes future policy/investment.
05 / PRODUCT DECISION RECORDS
The product is the sum of choices and trade-offs.
Each decision includes an implicit reversal test: better evidence can change the choice.
Usage is not ROI
AI runs, seats and tokens show adoption but do not prove business value.
Inventory before governance theater
The platform starts by discovering what exists and who owns it before adding heavy policy workflows.
Outcome Contract as the unit of value
Every important AI use case defines baseline, expected outcome, owner, review period and quality/risk tolerance.
Recommendations, not autonomous shutdowns
Consolidation or stop decisions have financial and organizational consequences; Aperture prepares evidence and workflow, humans decide.
06 / AI SYSTEM
The architecture separates AI judgment from exact rules, evidence, tools and human authority.
Every connector has a defined job. Animated dashed lines represent active evaluation/learning loops rather than decorative motion.
Primary decision/data flowContext/evidenceHuman/consequential pathContinuous evaluation loop
HOW I WOULD BUILD IT
Point by point, from state to operation.
Connectors
Read model-provider usage, agent/tracing tools, SSO, procurement/finance and BI/outcome systems.
Normalize estate
Map vendor-specific data into Asset, Use Case, Owner, Model, Cost, Eval, Data Class, Risk and Outcome Contract.
Reconcile cost
Attribute shared spend and usage to assets/use cases with confidence/coverage.
Quality normalization
Map heterogeneous eval metrics into a comparable quality evidence layer without pretending they are identical.
Analyst agents
Estate Analyst finds duplication/orphans; Quality Analyst finds weak evidence; ROI Analyst tests outcome coverage.
Policy engine
Deterministic rules route assets by data sensitivity, autonomy, spend and evidence gaps.
Decision queue
Generate scale/fix/consolidate/stop recommendations with reasons, owner and missing evidence.
Audit + review
Human decision, follow-up date and observed business outcome feed the estate history and policy tuning.
07 / EVALUATION ARCHITECTURE
How do I know the AI deserves to ship?
Component eval, system eval and product outcome are deliberately separated.
Asset discovery precision/recall
Asset discovery precision/recall, owner mapping and connector freshness.
Reconciliation coverage
Reconciliation coverage, unallocated spend and vendor-bill variance.
Metric mapping validity
Metric mapping validity, stale eval detection and evidence completeness.
Correct policy routing
Correct policy routing, tenant isolation, permission enforcement and audit completeness.
Duplicate-tool detection and scale/fix/sto
Duplicate-tool detection and scale/fix/stop recommendation agreement with expert review.
Outcome Contract coverage
Outcome Contract coverage, decision completion, realized savings/value and false consolidation risk.
Closed loop: trace → classify failure → add/refresh eval case → change source/retrieval/model/prompt/rule/tool → regression suite → controlled release → monitor outcomes/overrides → repeat.
08 / FEATURE PRD
A concrete example of how strategy becomes engineering work.
The PRD is intentionally feature-level and includes AI behavior, deterministic controls, telemetry and non-goals.
PRD example — Outcome Contract + consolidation review
Problem
Marketing has three overlapping AI tools, high spend, one orphaned owner and no common outcome/evaluation definition.
Outcome
Create an inspectable consolidation review that joins spend, usage, quality, ownership and business-outcome evidence before leadership chooses what to keep.
User stories
- As a Head of AI, I can see duplicated capabilities and missing owners across the AI estate.
- As a Finance partner, I can see cost per accepted/business outcome instead of only tokens or seats.
Functional + AI requirements
- Normalize each tool/agent into the AI asset schema.
- Create/assign owner and business use case.
- Attach Outcome Contract: baseline, target, measurement method, review window and confidence.
- Join vendor spend + usage + eval/quality evidence.
- Generate recommendation with missing-evidence list, not a black-box verdict.
- Create governed review with assigned owners and recheck date.
Acceptance criteria
- No asset can be auto-disabled by recommendation.
- Unallocated cost remains visible rather than silently distributed.
- Conflicting quality metrics are shown separately with mapping notes.
- Decision record stores participants, rationale and evidence snapshot.
- Review completion updates estate status and realized outcome.
Telemetry
asset_discovered · owner_missing · outcome_contract_created · duplicate_cluster_found · review_created · decision_recorded · recheck_completed · savings_verified
Non-goals
- Replacing every provider's observability tool
- Claiming cross-model eval metrics are perfectly comparable
- Fully autonomous procurement or shutdown actions
Engineering handoff
Typed state/schema · API/tool contracts · error states · permissions · eval fixtures · analytics events · rollout/rollback.
09 / BUILD VS BUY + RELIABILITY
Do not custom-build commodity infrastructure—and do not pretend every API always works.
BUILD DIFFERENTIATION
Normalized AI estate + Outcome Contract + decision layer
Use standard infrastructure for storage, models, tracing and connectors where it does not create strategic advantage.
BAD-DAY MODE
Last-known telemetry + freshness warning
Timeouts, stale data, partial results, rate limits and provider failures are explicit product states—not invisible backend details.
10 / WORKING PRODUCT JOURNEY
Use the product from input to changed state.
Every click represents a user/product decision and explains why the information is needed.
APERTURE AI · ENTERPRISE DECISION COCKPIT
The dashboard begins with uncertainty.
The product does not assume that every AI tool is valuable—or even known.
Five assets discovered. Two need attention.
HEALTHY
Support Copilot
Owner known · eval current · cost attributed · outcome defined.
REVIEW
Marketing AI cluster
3 overlapping tools · one owner missing · no common outcome contract.
QUALITY GAP
Internal Search
High usage · stale eval suite.
EARLY
Sales Agent
Pilot · spend low · outcome window not complete.
Usage is not ROI.
SPEND
$18.4k / month
Across three overlapping vendors.
OWNER
1 missing
Governance gap.
QUALITY
Not comparable
Each tool reports a different metric.
OUTCOME
Undefined
No baseline or measurement window.
Why no automatic shutdown?
The system has evidence of duplication, not proof that the tools create no value. The correct product action is a governed review.
The system has evidence of duplication, not proof that the tools create no value. The correct product action is a governed review.
Define the Outcome Contract before deciding.
Was this case study useful?
A quick signal helps me understand what is useful to recruiters, founders and product teams.
11 / PRODUCT LIFECYCLE
Autonomy and market scope are earned in stages.
Inventory proof
Connect 2–3 providers and reconcile asset/owner/spend.
Decision cockpit
Add quality, risk, Outcome Contracts and decision queue.
Governed workflows
Review/approval/audit for scale/fix/consolidate/stop.
Optimization
Vendor consolidation, spend forecasting and quality/cost frontier.
Enterprise platform
Policy APIs, business-unit rollups and ecosystem integrations.
GTM + ADOPTION
A credible route into the market.
Head of AI / CIO / CTO with FinOps, security and finance stakeholders.
50–1,000 person SaaS/knowledge companies already using multiple AI vendors/tools.
Inventory + owner + spend + quality evidence. Avoid selling ‘complete AI governance’ on day one.
Outcome Contracts → risk workflow → vendor consolidation → policy/decision platform.
KILL / PIVOT CRITERION
What would make me stop?
Stop or narrow the platform if customers can make the same scale/fix/stop decisions by joining existing FinOps + tracing + governance tools with minimal manual effort, or if connector normalization is too costly to maintain.
12 / TOOL + MODEL DECISIONS
Tools are mapped to responsibility—not displayed as decoration.
Where the source material does not prove an implementation, the portfolio says proposed/candidate rather than “built with.”
Model provider APIsusage/cost connectors
SSO / SCIMidentity + owner mapping
FinOps / procurementspend evidence
LangSmith / tracing toolsquality/run evidence
BI / warehousebusiness outcomes
NIST-aligned policy modelgovernance vocabulary
13 / SECONDARY RESEARCH
Evidence establishes the problem environment. It does not magically validate the solution.
McKinsey — State of AI 2025
McKinsey reports most organizations are still early in scaling AI; governance and workflow redesign correlate with greater value capture. Product implication: the enterprise problem is operating-model evidence, not simply adding more AI.
Open source ↗IBM Cost of a Data Breach 2025
IBM reports large governance/access-control gaps among organizations experiencing AI-related incidents. Product implication: visibility, ownership and controls need to grow with the estate.
Open source ↗NIST AI RMF
NIST provides Govern / Map / Measure / Manage risk-management structure. Aperture can map evidence to that vocabulary without claiming the product itself creates compliance.
Open source ↗PwC — Responsible AI agents
PwC recommends risk tiering based on agent autonomy and potential impact plus centralized/transparent oversight. Product implication: asset inventory should include autonomy and consequence.
Open source ↗EVIDENCE PACK
Download Aperture AI Product Evidence.xlsx ↗The working assumptions, PRD, eval suite and roadmap are inspectable.
No spreadsheet preview is embedded. Open the workbook only if you want the detail.