Overview
A two-week audit that pays for itself.
Then a retainer that proves it monthly.
Then a retainer that proves it monthly.
FlowLens runs silently on an employee's machine, capturing only behavioural metadata — never content. The AI engine turns that data into a ranked automation roadmap, specific to their role and industry. Even the client onboarding form is enhanced by AI — so recommendations are sharp from day one.
Initial Review
Days 1–7
Activity Capture
Activity Capture
FlowLens logs app usage, workflow sequences, and interaction patterns. Nothing leaves the client machine.
Client fills in onboarding form on havaron.com
AI enriches the form input — expands sparse descriptions, infers automation context, standardises tool names
Pre-configured FlowLens downloaded and installed
Logger runs silently on every login
Pattern engine runs nightly, improving daily
AI generates initial recommendations
Discovery questions surfaced for consultant
Deep Dive
Days 8–14
Enhanced Analysis
Enhanced Analysis
Client selects which apps to analyse more deeply. Enhanced tracking begins. Discovery answers refine the AI output.
Client reviews Initial Review findings
Deep Dive activated per app (opt-in)
Discovery conversation with consultant
AI re-runs with client context injected
Developer-ready automation briefs generated
Build & Retain
Ongoing
Implementation
Implementation
Havaron builds the automations. Client runs them in parallel with an AI assistant. Monthly ROI reports track every minute saved.
Client selects priority automations
Havaron builds and deploys
AI chatbot guides the transition
Monthly ROI report vs audit baseline
Retainer for enhancements and monitoring
Architecture
Data flows from client machine
to consultant — never the reverse.
to consultant — never the reverse.
Raw behavioural events never leave the client machine. Only processed summaries — patterns, recommendations, app usage — are pushed to the cloud. The consultant sees results, not raw data.
Data Flow — Client Machine → Supabase Cloud → Consultant Portal
Client Machine · Windows PC
flowlens.db
✅ events table — stays here always
✅ excel_events — stays here always
↑ app_usage summary (minutes per app)
↑ patterns (descriptions + confidence)
↑ recommendations (AI output)
↑ discovery_notes (Q&A answers)
↑ workflow_aggregates (anonymised stats)
✅ excel_events — stays here always
↑ app_usage summary (minutes per app)
↑ patterns (descriptions + confidence)
↑ recommendations (AI output)
↑ discovery_notes (Q&A answers)
↑ workflow_aggregates (anonymised stats)
sync_agent.py
nightly · HTTPS
summaries only
nightly · HTTPS
summaries only
Supabase Cloud · Singapore Region
Cloud Database
› clients (profile per engagement)
› app_usage (time per app)
› patterns (detected workflows)
› recommendations (AI output)
⇄ discovery_notes (consultant editable)
› workflow_aggregates (benchmark data)
› sync_log (audit trail)
› app_usage (time per app)
› patterns (detected workflows)
› recommendations (AI output)
⇄ discovery_notes (consultant editable)
› workflow_aggregates (benchmark data)
› sync_log (audit trail)
REST API
read + write
any browser
read + write
any browser
Consultant Portal · Any Device
consultant_portal.html
✓ All clients in one view
✓ Live pattern and recommendation data
✓ App usage with Deep Dive badges
✓ Enter discovery answers remotely
✓ Last sync timestamp per client
✓ Live pattern and recommendation data
✓ App usage with Deep Dive badges
✓ Enter discovery answers remotely
✓ Last sync timestamp per client
Privacy guarantee: Raw events never leave the client machine · No keystrokes or file content ever captured · Clipboard stored as hash only · Deep Dive is opt-in per app · Architecturally PDPA compliant
AI Engine
How patterns become recommendations.
The pattern engine detects workflow signals from raw events. After the discovery call, client answers sharpen the engine — filtering noise and focusing detection on confirmed workflows — so Deep Dive recommendations are built on facts, not hypotheses.
Onboarding
AI Enhanced Client Onboarding
Triggered at sign-up. The client provides their role, industry, daily tasks, and tools — Claude AI enriches every field, expanding sparse descriptions and inferring automation context before the audit even begins.
Expands sparse descriptions into rich context
Infers automation opportunities from role + tools
Standardises tool names across industries
Generates targeted discovery prompts
Output feeds Pattern Engine + Recommendation Engine
↓ Enriched client context
Detection
Pattern Engine
Reads the local events table nightly with 5 core detectors — copy-paste flows, context switching, daily sequences, time-of-day habits, and short frequent tasks. Deep Dive adds 6 Excel-specific patterns. Gets smarter after discovery.
Client answers narrow detection to confirmed app pairs
Excluded apps filtered out entirely
Confidence scores recalibrated against confirmed processes
+6 Excel-specific patterns in Deep Dive mode
↓ Patterns + confidence scores
Recommendations
Claude AI · Recommendation Engine
Receives pattern descriptions, client profile, Singapore tool library, and discovery answers. Discovery context also feeds back into the pattern engine to sharpen detection before the Deep Dive run.
Automation recommendation per pattern
Time saving estimate + complexity rating
Tool suggestion from Singapore library
Likely processes + confirmation questions
Priority score (1–10)
↓ Ranked automation briefs
Delivery
Client Dashboard
Delivered as a standalone HTML report — no install required. The Initial Review shows app usage, pattern cards, and recommendations. Deep Dive adds developer-ready briefs, automation scenarios, and next steps.
App usage chart + pattern cards
AI recommendations with priority scores
Discovery Q&A section
Developer-ready automation briefs
Actionable next steps
Timeline
Every day, the data gets
richer and more accurate.
richer and more accurate.
The pattern engine runs nightly via Windows Task Scheduler. Each additional day of logging adds more signal — patterns strengthen, confidence scores rise, and recommendations sharpen. By Day 14 the AI has two weeks of context to work from.
14-Day Audit Timeline — Nightly Automated Pipeline
Logging active
Pattern engine runs
AI recommendations
Synced to cloud
Deep Dive active
Dashboard delivered
Data Quality — Cumulative Improvement Over Audit Period
Events logged
~19,000+
Patterns detected
~300+
Recommendation quality
High
Pattern confidence
75–95%
Privacy
What stays on the machine.
What goes to the cloud.
What goes to the cloud.
FlowLens is built privacy-first by architecture — not policy. The client machine physically cannot send raw events to the cloud. The sync agent only has access to processed summaries.
Stays on client machine — always
Never transmitted, never accessible remotely
Raw events table — every app focus, click, keyboard burst
Window titles — document names, email subjects, browser URLs
Clipboard content — stored as hash only, content never read
Excel cell data — sheet names captured, cell content never
Keystrokes — burst frequency only, no characters ever captured
File contents — never accessed or read under any circumstance
Synced to cloud — processed only
Anonymised summaries with client consent
App usage summary — time spent per application in minutes
Pattern descriptions — human-readable workflow signals
AI recommendations — automation briefs and priority scores
Discovery Q&A — consultant questions and client answers
Deep Dive config — which apps the client opted into
Workflow aggregates — anonymised statistics for benchmarking
Engagement Model
From first install to
ongoing automation partner.
ongoing automation partner.
Havaron operates as your organisation's external CTO — automating processes, saving time and cost, and keeping you ahead of every development in AI as it unfolds.
Phase 1
Initial Review
Days 1–7 · Audit Fee
Client fills in onboarding form — AI enriches input before download
FlowLens installed, logging begins
Nightly pattern + AI pipeline
Initial Review dashboard delivered
Discovery questions surfaced
Phase 2
Deep Dive
Days 8–14 · Audit Fee
Discovery conversation with client
Deep Dive activated per app
Enhanced pattern detection
AI re-runs with discovery context
Final dashboard + report delivered
Phase 3
Build & Implement
Weeks 3–6 · Implementation Fee
Client selects priority automations
Havaron builds and tests
Client runs old and new in parallel
AI chatbot guides the transition
Full handover once validated
Phase 4
Maintain & Improve
Ongoing · Monthly Retainer
Full automation deployment
Monthly ROI report vs baseline
AI chatbot for enhancements
Havaron on call for debugging
Retainer justifies itself monthly