Strategy and System Design Case Study

Protecting $500K to $2M enterprise deals
by translating AI governance into buyer-ready proof

Qlarc turns the governance AI vendors already have into proof enterprise buyers can evaluate.

Role
I led product design, system architecture and the financial model.
Qlarc was a 5-person capstone.
Goal
Get a vendor through enterprise procurement on the governance they already have.
02The Problem

The governance exists.
It just cannot be translated.

AI-enabled vendors have real governance work. Real people wrote it. Real teams signed off on it. But it lives scattered across the organization and none of it is in the format enterprise procurement legally requires or can evaluate.

What the vendor already has
  • EngineeringUNTAGGED
  • LegalUNTAGGED
  • ComplianceUNTAGGED
  • SecurityUNTAGGED

Several teams hold the proof. None of it is tagged or in a form a buyer can read.

What buyers require
Per regulation · per clause · non-negotiable
SR 11-7
Fed Reserve
Named model governance evidence, traceable to source article
clause-level required
FINRA
2026
Explainability documentation per AI decision type
clause-level required

One structured document. Every claim at the clause level. Incomplete submissions are discarded, not reviewed.

The governance exists on the left. The buyer requires the right. Nothing converts between them.

72%of enterprises raising AI-governance requirements (PwC, 2025)

The cost?

The vendor submits what they have. The buyer’s AI Ops team marks it incomplete.
They do not follow up.

Deal record  FIN-ENT-2026-0342
Closed lost
Revenue lost per documentation stop
$2M
Deal value
$1.8M
Days active
87
Blocked at
AI Gov Review
Day 57
Follow-up
None
VP: day 87
Deal progression
Discovery
cleared
d1-14
Proposal
cleared
d15-28
Tech Review
cleared
d29-42
Security
cleared
d43-56
AI Gov Review
blocked
d57
Legal
not reached
not reached
Contract
not reached
not reached
The aha

Buyers built systems to ask.
Vendors never built systems to answer.

From the notebook

i kept trying to design a better governance tool, and it kept feeling redundant, the vendors already had governance. took me a while to admit i was solving the wrong half. the gap wasn't missing rules, it was that one side spent ten years building systems to ask and the other side had nothing to answer with. the moment i stopped adding governance and started translating it, the whole thing clicked.

03Validation

Validated the problem from the people who enforce it, feel it and advise on it.

Validated across three named experts in two research phases, early-stage discovery and a March 2026 follow-up. Each conversation shaped a specific product decision below.

01Interview

Pratik Patel

AI Ops Leader · former Mastercard
Procurement stops are hard blocks, not delays. Unclear answers end evaluation, buyers do not follow up.
↳ How I responded
Qlarc surfaces gaps before submission, not after.
02Interview

Dr. Cari Miller

Co-Founder, AI Procurement Lab · former IEEE Vice Chair
Proactive, candid vendors win. Evasive ones do not, regardless of product quality.
↳ How I responded
Qlarc shows gaps honestly with named human approval.
03Interview

Debashis Bhattacharyya

Practice Lead, Tech Consulting · Opus Technologies
Documentation gets rebuilt from scratch every cycle. Sales teams cannot answer technical AI questions.
↳ How I responded
Qlarc makes evidence reusable, versioned and source-tagged.
04The Solution

Qlarc turns scattered governance into structured proof.

A vendor-side platform for AI-enabled FinTech companies preparing for enterprise procurement review. It takes the evidence a vendor already has and structures it into a package a buyer can actually evaluate.

Step 1
Connect & Upload

Assess readiness, connect AI metadata, and upload the governance evidence that already exists.

Step 2
Map the evidence

Map evidence to procurement questions, then fill only the gaps.

Step 3
Review & approve

A named human signs off on every claim before anything ships.

Human gate
Step 4
Export the pack

Export one verifiable evidence pack, ready for the buyer.

Step 1 of 4

Connect & Upload

The vendor connects AI systems and uploads existing governance documents.

Within minutes. Not hours. Not days.

What goes in
AIOpenAI · AWS Bedrock · Azure

Read-only · like reading a phone bill

Model
GPT-4o
v2.1 · active
Region
US-East
EU-West backup
Governance documents
  • AI Governance Policy v3
    3 evidence items extracted
  • Bias Testing Report Q1 2026
    4 evidence items extracted
  • Data Handling Procedure
    Extracting evidence · processing
What comes outautomatically translated
65%of your Evidence Pack already built.

Not created from scratch. Found.

Found in existing systems and documents35% gap fill remaining
Model gov
90%
API-verified from source
Bias testing
55%
Needs gap fill
Data gov
80%
Document extracted
AI stack
65%
API-verified from source
Exports as2 deliverables
Procurement Response Pack
One-Page Governance Summary

The Vendor has to review and approve the answers.

Illustrative example, not a measured result

From the notebook

the tempting version of this product just writes the answers for you. i killed that on purpose. the second a tool invents a compliance claim it's not a feature, it's a liability. so every claim routes through a named human before anything ships, the AI maps and drafts, but a person still signs their name to it. “translates, doesn't invent” isn't a tagline here, it's the whole product decision.

05Product Decisions / System Architecture

The hard part was knowing what to refuse.

Three decisions. Two hard boundaries. One pipeline. The structure is a direct response to the strategic bet, and each boundary exists because a specific failure mode in regulated FS procurement made it necessary.

01 · Inputs
Evidence quality is structural, not documented
Instead oftreating all inputs as equivalent and letting the vendor self-report everything
BecauseAI Ops reviewers evaluate source attribution directly. Pratik Patel confirmed: generic documentation fails review regardless of product quality. The tier system encodes what they evaluate into the architecture, not a policy document.
Inputs · Vendor's governance
API-verified signals1.0
Governance documents0.75
Gap-fill answerslabeled
Every input enters tagged by source strength. Nothing is treated as equivalent.
02 · Processing
The boundary between extraction and citation is architectural, not instructional
Instead ofletting the AI generate regulation citations alongside evidence extraction
Becausea hallucinated citation in FS procurement is a legal liability, not an accuracy issue. The Regulation Library is isolated structurally. No prompt can override it.
Processing
AI Zone
Extraction
Reads documents, tags each item by source, maps to evidence-framework items.
Reg Library
Citations
10 questions, manually verified, quarterly versioned.
The model extracts evidence. It never writes a citation.
03 · Gate
Accountability stays with the vendor. Entirely.
Instead ofa confirmation step that logs vendor intent, with approval treated as a UX courtesy
Becauseevery answer carries contractual and regulatory consequences. The gate is a liability architecture decision. Named approval, timestamped, cryptographically hashed. The vendor is accountable. Not the platform.
Human Gate
Named approval required
Review · Attest · Timestamped hash
Export locked until all 10 sections approved
Result

What the three boundaries produce: documentation that survives review because every claim is sourced, every citation is verified, and a named human stands behind it.

Outputs
Procurement Response Pack
Reg-mapped · source-attributed · versioned
One-Page AI Governance Summary
Plain language · proactive share
06Trade-Offs

I chose trust over automation.

Design principleSpeed only matters if trust survives it.
FinTech-first scope
gave up broader market coverage
gained sharper urgency
Vendor-side focus
gave up buyer-side GRC market
gained clearer pain
Evidence translation
gave up full governance platform
gained stronger wedge
Read-only integrations
gave up deeper access
gained lower security risk
Verified regulation mapping
gave up faster AI citations
gained lower hallucination risk
Delayed integrations
gave up bigger MVP demo
gained cleaner product focus

The fastest answer was not the best answer.
The most trustworthy answer was.

07Business Model

Why the pricing works.

Set the deal and the tier. One recovered deal covers the year. Then stress-test the math with your own assumptions.

Deal at risk$800K
Subscription tier
To cover the year
1 deal

One recovered $800K block pays for the $25K Scale tier for the year, many times over.

Subscription
One recovered deal
Stress-test, your odds it saves one block this year100%
floor 3%
At those odds, the subscription returns 32x.still clears
Modeled, not earned

This is a sensitivity check, not an outcome. What a pilot actually tests is whether vendors convert at the rates the model assumes.

And before a single recovered deal is counted, the manual coordination Qlarc replaces runs $25K to $40K a year in senior staff time. The operational case stands on its own.

08Building with AI

AI does the work.
I do the thinking.

AI helped

Research, synthesis, iteration · Speed and scale

Where it couldn’t

Strategic framing · Regulatory accuracy · Defining what the product is

CHANGE 01Frame shift.
AI said

Qlarc is a compliance platform that helps vendors demonstrate regulatory adherence.

I said

Vendors already have governance. This is an evidence translation engine, not a compliance tool.

New GTM, new positioning, new competitive set. One reframe changed everything downstream.

CHANGE 02$2M rests here.
AI said

Vendors face procurement delays of 15 to 20 days when governance documentation is incomplete.

I said

This is a hard stop, not a delay. Buyers mark submissions incomplete and close the evaluation.

The entire revenue-loss argument depends on this single distinction. A delay is recoverable. A hard stop is not.

09Open Risks

The next version would need to prove trust inside live procurement cycles.

The risk I would test first

Will vendors pay for this?

Why it matters

The problem is validated. Demand is not. Three experts confirmed the pain is real, but none was asked whether they would buy.

Next step

Pilot pricing with vendors in active procurement to test conversion at $15K to $25K.

The part I'm still unsure aboutAn honest note

I set out to help honest vendors prove they’re trustworthy. The risk I never closed is that the same tool helps any vendor look the part.

Qlarc packages the governance a vendor already has into proof a buyer will accept. The hardest decision wasn’t what to build, it was how to build it so it could be reused, and that’s the catch: the clean, repeatable format that helps an honest vendor read as ready helps a careless one just as well. Source-attribution and the human gate stop it from inventing evidence, not from making thin governance read as strong. Whether Qlarc raises the floor on trust or just raises the polish is the line I’m least sure I stayed on the right side of.

Margin note · to self

the honest version: i built this to get good vendors past a broken process. the same tool gets a careless one past it just as fast.

source-attribution and the gate are my answer. i still can’t tell if a clean pack proves good governance, or just good packaging.

Next case study

Mura.

StrategyProduct DesignSustainabilityConsumer

No electricity. No synthetics. Charged on sunlight.