Side project · AI product concept

CompliancePilot AI — AI-Assisted Evidence Review for Compliance Teams

Helping teams move from scattered documents and unclear blockers to confident compliance decisions. CompliancePilot AI identifies missing, expired, weak, or insufficient evidence, explains why a control is blocked, drafts the next request, and keeps humans in control before any action goes out.

It answers one question across the whole workflow: what is blocking compliance readiness, and what evidence is needed next?

Launch the prototype React + Tailwind · opens in a new tab
Role
Product Designer · AI Product Concept
Timeline
Side project · 2026
Scope
B2B SaaS · Compliance · AI · Evidence Review
Tools
Figma · AI prototyping · React prototype
Before
  • Long control lists and scattered documents
  • Evidence uploaded but hard to judge
  • Repeated follow-ups with evidence owners
After
  • AI-prioritized compliance blockers
  • Source-based evidence review
  • Human-approved evidence requests
Concept focus
  • AI-assisted review
  • Evidence quality checks
  • Human-in-the-loop decisions

Compliance teams don’t just need a place to store evidence. They need to know whether each document actually proves the control.

Compliance teams work with long control lists, scattered documents, unclear evidence quality, deadlines, comments, and review statuses. Even when documents are uploaded, reviewers still need to manually understand whether the evidence actually proves the required control.

The main friction isn’t evidence storage — it’s evidence interpretation. Reviewers spend hours deciding whether an uploaded policy actually demonstrates that MFA is enforced, whether a one-year-old certificate is still acceptable, whether a backup policy without test results closes the control. That repeated interpretation work creates confusion, slow reviews, repeated follow-ups with evidence owners, and weak confidence in compliance readiness.

Where the friction lives
01
Hundreds of controls, no priority
  • Flat lists hide what’s blocking
  • Risk + deadline + impact aren’t combined
  • Reviewers start somewhere arbitrary
02
Uploaded ≠ sufficient
  • Policy doesn’t prove enforcement
  • Document looks right, doesn’t close the control
  • The judgment is manual every time
03
Compliance jargon blocks owners
  • Controls written for auditors
  • Owners don’t know what to upload
  • Reviewers re-explain the basics over Slack
04
Manual follow-ups take the day
  • The same chase email, rewritten
  • Status moves only when someone types it in
  • Deadlines slip while threads sit

Five people, one workflow.

CompliancePilot supports five roles around a single compliance review. Each one needs different language, depth, and surface area. The MVP focuses on the Compliance Manager.

CM
01 · Compliance Manager
Owns readiness, unblocks the team
A prioritized list of what’s blocking compliance and what to do today.
200+ controls to inspect, no clear sequence.
AI-ranked blockers on the home page, with a one-line reason and a suggested action.
RR
02 · Risk Reviewer
Judges evidence and supports decisions
A second opinion on whether evidence really proves the control.
Interpretation work is repetitive and inconsistent across the team.
An AI assessment with confidence and sources next to every uploaded document.
AR
03 · Auditor / Reviewer
Validates controls, leaves the decision trail
The control, its evidence, and its history in one place.
Decisions and comments scatter across systems.
A Control Review Workspace that surfaces requirement, evidence, AI assessment, and review history side by side.
EO
04 · Evidence Owner
Just wants to upload the right thing, fast
A plain-language list of what to send and where to send it.
Requests are vague; compliance language is hard to parse.
A simplified request view with examples of acceptable evidence and direct upload.
DM
05 · Manager / Decision Maker
Wants readiness, blockers, a decision
A short summary they can act on or forward.
Detail without synthesis. Reports overwhelm.
An AI-generated manager summary in plain language, with the blockers and recommended next steps.

An evidence review workspace where AI does the interpretation work.

CompliancePilot is structured around six things AI helps users understand. Each one resolves to a clear next action — the AI never just describes a problem.

01 · What’s missing
Surfaces evidence gaps blocking review

Required-but-not-uploaded items rise to the top of the home page, ranked by control priority and audit deadline.

02 · What’s expired
Tracks document freshness against the control

Certificates, reports, and policies carry an expiry date that flips them to action-required before they silently age out.

03 · What’s insufficient
Judges whether the document proves the control

When a policy is uploaded but doesn’t demonstrate enforcement, CompliancePilot calls that out with a plain-language explanation.

04 · Why a control is blocked
Links the blocker to its specific root cause

“Missing screenshot” or “Conflicting comment from owner” instead of a generic status badge.

05 · What source data supports it
Every recommendation cites its inputs

Control requirement, uploaded documents, reviewer comments, evidence status, audit deadline — chips on every important answer.

06 · What action happens next
Drafts the artifact, the human ships it

Evidence requests, clarifications, remediation tasks, and manager summaries are AI-drafted and human-approved before sending.

Ten screens, one decision path.

From role onboarding to manager summary — every screen earns its place by answering one of the user’s questions about evidence or action.

01 · Role setup
Personalize the workspace

Five role cards (Compliance Manager, Risk Reviewer, Auditor, Evidence Owner, Decision Maker) + priority chips. Drives what rises to the top of the home page.

02 · For You dashboard
Today’s prioritized review queue

AI briefing + top compliance blockers, ranked by evidence gaps, control risk, deadline, and review status.

03 · Audit / review detail
Readiness at a glance

Readiness score, KPIs (controls complete, missing evidence, expired docs, insufficient evidence), and the controls table.

04 · Control review workspace
Where decisions actually happen

Requirement, plain-language explanation, acceptable evidence, current upload, AI assessment, and approve / reject / clarify actions.

05 · AI assistant panel
Context-aware, evidence-grounded chat

Structured replies: Recommendation, Why, Evidence used, Confidence, Suggested next action — with action buttons that wire back into the app.

06 · Evidence & sources panel
The audit trail behind every recommendation

Tabs for Evidence, Comments, History, and Sources. Source chips (control requirement, uploaded documents, reviewer comments, evidence status, audit deadline) on the trust tab.

07 · Request evidence modal
AI drafts, human approves

Editable AI-drafted message with recipient, related control, framework, priority, due date, and the specific evidence needed.

08 · Evidence owner view
Simplified for the person uploading

Plain-language explanation, requested items with examples, upload + comment + submit. Deliberately shorter than the reviewer surface.

09 · Evidence Hub
Documents as first-class objects

Status, linked control, owner, expiry, AI match confidence, and the next review action — sortable and filterable.

10 · Manager summary
A paragraph, not a deck

AI-generated readiness summary in plain language, with blockers and recommended next steps. Copy, save, send to manager.

Structured replies, always cited.

Every AI answer follows the same five-part structure. It’s the contract that lets a reviewer trust the assistant without checking every recommendation against the source data themselves.

Example prompts
What is blocking this review?
Is this evidence enough?
Why is this control blocked?
What evidence should I request?
Explain this for a manager.
Draft a request to the evidence owner.
A CompliancePilot · reply for “Is this evidence enough?”
Recommendation
Not enough to approve.
Why
The uploaded Access Control Policy describes the expected access control process, but it does not prove that MFA is technically enforced for privileged accounts.
Evidence used
  • Control A.5.17 requirement
  • Access Control Policy.pdf
  • Evidence status: uploaded but insufficient
  • Missing MFA enforcement proof
Confidence
Medium-high
Suggested next action
Request a screenshot, IAM configuration export, or identity provider policy showing MFA enforcement for privileged accounts.

No magic. Every AI move shows its work.

In a compliance context, an assistant that can’t be audited is worse than no assistant. CompliancePilot enforces five rules on every AI surface so the reviewer can defend any recommendation.

01
Source data

Every recommendation lists the controls, documents, and comments it drew from.

02
Confidence

Stated explicitly — high, medium-high, medium, low — never hidden.

03
Related controls

Each AI answer links back to the controls and frameworks it touches.

04
Related documents

The exact files the assistant judged, with status and freshness visible.

05
Human approval

Sending a request or approving evidence always requires a human click.

Six choices that shaped the product.

Each decision was a trade-off — cleaner cognitive load against feature breadth, AI utility against reviewer accountability.

  1. 01
    An action-first dashboard, not a generic analytics one

    The home page leads with what needs review and what to do next. KPIs are supporting context; charts only appear when they answer a specific question.

  2. 02
    Evidence is a first-class object, not an attachment

    Every document carries status, linked control, owner, expiry, and AI match confidence. The Evidence Hub stands alongside Audits and Controls in the IA — not buried under a single control’s tab.

  3. 03
    AI recommendation is separated from human decision

    The AI assesses. The reviewer approves, rejects, or asks for clarification. The two never blur, even visually — AI cards have their own background, review actions have their own region.

  4. 04
    Progressive disclosure to reduce cognitive load

    Summary first. The reason on demand. The full source trail one click away. The reviewer chooses when to go deep.

  5. 05
    Source traceability is visible in every important AI response

    No floating claims. The five-part reply contract (Recommendation / Why / Evidence used / Confidence / Suggested next action) is consistent across screens.

  6. 06
    Designed for five roles, prototyped on one

    All five personas inform the IA, but the MVP commits to the Compliance Manager flow end to end — better to ship one deep path than five shallow ones.

A concept, honestly framed.

This is a design exploration, not a shipped product — so the value sits in what I’d expect it to unlock and what I learned by building it.

Expected impact
Where this would move the needle for a compliance team
  • Faster identification of compliance blockers
  • Less confusion around what evidence is required
  • Fewer repeated follow-ups with evidence owners
  • More consistent review decisions across the team
  • More confidence in audit readiness reporting
What I explored
The design questions I worked through
  • How AI can support evidence interpretation, not just storage
  • How to keep compliance decisions human-controlled
  • How to design source-based AI recommendations
  • How to reduce complexity in dense B2B compliance workflows
  • How a single workspace can serve five roles without becoming generic
What this argues

A compliance platform that stops at “document uploaded” isn’t finished. The real question is whether the document actually proves the control — and that’s where AI can earn its place, as long as it cites its work.

CompliancePilot is a sketch of what that contract looks like in product form.

More case studies