githubEdit

file-lines2. Roadmap Generation

Outcome: Professional AI Roadmap PDF demonstrating deep business understanding with specific ROI projections Trigger: Discovery Call complete (all interviews finished, transcripts saved) Duration: 1-2 hours (generation + review) Output: {client}/AI-ROADMAP.md and {client}/AI-ROADMAP.pdf


Quick Reference

Step
Action
Time
Method

1

Verify inputs complete

5 min

Manual check

2

Run roadmap pipeline

20-40 min

./scripts/roadmap.sh {client}

3

Review & refine output

30-45 min

Manual review

4

Prepare for delivery

5 min

PDF conversion

Total: ~1-2 hours


Prerequisites

Before starting, verify:


Step 1: Verify Inputs (5 min)

Check that your transcripts include sufficient data:

Data Point
Minimum Content

Business Context

What they do, team size, why automation now

Software Stack

At least 3-5 tools mentioned

Processes

At least 3 processes/tasks with step-by-step walkthroughs

Pain Points

Quantification (time, frequency, impact) for major processes

ROI Data

Hours per week + hourly rates for key roles

Goals

Growth/future state discussed

If data is thin: Fill gaps from recordings or ask client via email before proceeding. Having multiple call transcripts (onboarding + interviews) gives the generator much richer data to work with than a single call ever could.


Step 2: Run the Roadmap Pipeline (20-40 min)

Primary method (recommended):

By 4-letter client code:

Enterprise roadmap (custom software, investment tables):

Autonomous mode (skip review gate, auto-approve):

The pipeline runs 7 phases, each spawning a fresh agent:

Phase
Name
What Happens
Duration

0

Pre-Flight

Validates client folder, finds transcripts, checks API keys

5-10s

1

Transcript Analysis

Extracts team, pain points, quotes, hours, software stack

3-5 min

2

Platform Research

Researches each platform's API, industry context, compliance

5-10 min

3

Document Generation

Writes complete AI-ROADMAP.md using template and exemplars

5-10 min

4

Mac Review

Human gate -- approve, edit, or skip PDF

Manual

5

Diagrams

Generates workflow diagram PNGs via Gemini

1-2 min

6

PDF

Generates branded PDF via WeasyPrint

10-30s

Phase 4 pauses for your review in interactive mode. In autonomous mode, it auto-approves. Resume after failure: completed phases are auto-skipped on re-run.


Step 3: Review & Refine (30-45 min)

Open generated {client}/AI-ROADMAP.md and verify:

Quality Checklist (18 items)

#
Check
What to Look For

1

Executive summary

Has the total savings number in the first paragraph (not vague)

2

Team table

Every person from the transcripts appears with name, role, hours, and cost

3

Software ecosystem table

All platforms listed with 7+ columns (Platform, API Type, Auth, Rate Limits, n8n Integration, Key Endpoints, Docs)

4

Client quotes

Direct quotes from call transcripts woven into the narrative

5

ROI math

Hours x rate x 52 = annual cost. Automation savings realistic (60-90%, not 100%)

6

Automation opportunities

Each has description, current state, automated state, time savings, annual value

7

Org chart section

"How Your Team Operates" with narrative about manual vs strategic time

8

Opportunity matrix

"Where to Start" with recommended positions explained

9

Workflow cards

Each opportunity has "How It Works" card with 4-6 bullets + savings headline

10

12-month timeline

Month-by-month table with future-state narrative

11

Deep API research

Platforms without APIs note Kernel.sh as fallback

12

Image branding

All generated PNGs use TEAI branding (navy, blue accents, green for money)

13

Compliance section

Industry-specific considerations if relevant

14

300+ lines

Shorter usually means missing depth -- expand thin sections

15

Research Sources

Includes real URLs from web research, not placeholders

16

One-Time Optimisations

3-5 items connected to automation readiness

17

Assessed but Not Recommended

2-3 items with honest reasons (builds trust)

18

Common Questions

At least 5 Q&A pairs present

Common Fixes

Issue
Fix

Generic language ("your team")

Replace with specific names from the calls ("Maddy", "the warehouse team")

Over-optimistic ROI

Cap automation efficiency at 85-90%. Some human oversight always remains

Missing limitations

Add "What It Cannot Do" for each opportunity -- builds trust

Jargon (API, webhook, endpoint)

Replace with client-friendly language ("automatic connection", "trigger")

Fabricated quotes

Cross-check every blockquote against transcripts. Remove unverified quotes

ROI math errors

Check: hours/week x rate x 52. Conservative estimate should be 75% of calculated savings

Missing platforms

Search all transcripts for tool names. Verify each appears in the ecosystem table

Wrong timeline estimates

Default ~6 weeks per workflow. LOW ~4wk, HIGH ~8-12wk only. Max 2 concurrent


Step 4: Prepare for Delivery (5 min)

  1. Convert roadmap to PDF (the client receives this, not the markdown)

  2. Filename: {Company Name} - AI Roadmap.pdf

  3. Save to {client}/context/docs/AI-Roadmap-{DATE}.pdf

  4. Do not send directly to client -- hand off to sales (see Roadmap Presentation)


Edge Cases

The agent automatically adds a compliance section when keywords are detected:

  • Healthcare: HIPAA

  • Finance: SOC2, PCI-DSS

  • Legal: Attorney-client privilege

Verify: Correct regulations identified, automation opportunities don't violate compliance, appropriate caveats included.

Legacy Systems

Verify: Workarounds are realistic, ROI accounts for additional integration effort, automation % is adjusted appropriately.

Multi-Location/Multi-Team Clients

  • May need separate opportunity sections per location

  • ROI should aggregate across locations

  • Note implementation phasing by location


Feasibility Check

For any capability that is:

  • Novel (we haven't built this exact thing before)

  • Dependent on third-party constraints (API limits, platform ToS)

  • High-value ($5K+ or critical to client's decision)

Required: Build a POC or spike before including in roadmap.

Red Flags Requiring POCs

  • Scraping websites with bot detection

  • AI content that must pass plagiarism checks

  • Real-time integrations with rate-limited APIs

  • Automations bypassing platform intended workflows

  • Any "we think we can do this" statements

If uncertain: Mark as "Exploratory" with disclaimer:

"This capability requires technical validation. We'll confirm feasibility during the first week of development before committing budget."


Document Turnaround

Complexity
Turnaround

Simple (1-2 workflows)

Same day

Standard (3-4 workflows)

Same day

Complex (5+ workflows)

Next business day


Troubleshooting

Issue
Solution

"No transcript files found"

Rename transcript file to include "call" or "interview" in the filename

Phase 1 fails with thin data

Transcript too short or lacks specific numbers. Re-listen to calls, add manual notes with hours/rates

Phase 2 timeout

Too many platforms, web search taking too long. Re-run -- the script resumes from Phase 2 automatically

Phase 3 produces short doc

Check the state file at planning/_roadmap_state.json -- are phase data fields populated?

ROI numbers seem wrong

Check arithmetic: hours/week x rate x 52. Conservative estimate should be 75% of calculated savings

Fabricated quotes

Cross-check every blockquote against the transcript. Remove any that can't be verified

Diagram generation fails

GOOGLE_API_KEY not set or quota exceeded. Non-blocking -- PDF generates without diagrams

PDF generation fails

Install: pip install weasyprint markdown

State file corrupted

Delete planning/_roadmap_state.json and re-run from scratch


Verify

Before marking complete:

Next: Roadmap Presentation

Last updated