How I Structure Every AI Consulting Engagement (The 5-Phase Framework)
Source: Dev.to
Why Frameworks Matter for AI Consulting Specifically
AI consulting has a unique credibility problem. Clients have been burned by overpromised automation projects that delivered nothing, so they are skeptical. A clear, documented framework signals professionalism and manages expectations before the work begins.
It also lets you quote fixed‑price projects with confidence. When you know exactly what each phase costs in time and effort, you stop guessing on proposals.
Phase 1: Discovery (Weeks 1‑2)
Goal
Understand the client’s current state before proposing anything.
Key activities
- Stakeholder interviews (using AI‑generated question banks tailored to their industry)
- Current‑state process mapping (I use Claude to turn interview transcripts into visual workflow descriptions)
- Technology audit (tools, data flows, bottlenecks)
- Pain‑point prioritization matrix
AI acceleration
I feed all interview transcripts and notes into a structured discovery prompt that outputs a Current State Summary doc in under an hour. What used to take a week of synthesis now takes an afternoon.
Typical duration
8‑12 hours (2 weeks elapsed)
Deliverable
Current State Assessment + Opportunity Inventory
Phase 2: Diagnosis (Weeks 2‑3)
Goal
Quantify the problems and size the opportunity.
Key activities
- Data analysis (volume metrics, error rates, cycle times)
- Gap analysis between current state and industry benchmarks
- Opportunity sizing (conservative, moderate, aggressive scenarios)
- Root‑cause analysis on the top 3‑5 friction points
AI acceleration
A Diagnosis Prompt takes raw data exports and outputs a structured Gap Analysis with opportunity sizing in three scenarios. A task that used to require 6+ hours of Excel modeling now takes ~45 minutes.
Typical duration
6‑10 hours
Deliverable
Diagnosis Report with quantified opportunity size
Phase 3: Design (Weeks 3‑4)
Goal
Build the solution architecture and get client buy‑in before building anything.
Key activities
- Solution option development (2‑3 approaches at different investment levels)
- AI tool/workflow selection and rationale
- ROI model (built on the Diagnosis data)
- Implementation roadmap with milestones
- Risk register
AI acceleration
The ROI model template and solution architecture framework are AI‑assisted. I brief the system on the diagnosis findings and get a draft architecture and financial model in about 90 minutes, then refine it.
Typical duration
10‑15 hours
Deliverable
Solution Architecture + ROI Model + Implementation Roadmap
Phase 4: Delivery (Weeks 4‑12+)
Goal
Build and deploy the agreed solution.
Key activities
- Prompt engineering and workflow development
- Integration work (connecting AI outputs to client systems)
- Pilot testing with real client data
- Iteration based on feedback
- Change management (getting users to adopt the new system)
AI acceleration
Every deliverable in this phase—training materials, workflow documentation, test scripts, change‑management emails—is drafted through AI and refined. Speed‑to‑draft has roughly doubled.
Typical duration
Varies by scope (40‑200+ hours)
Deliverables
Working AI systems + documentation
Phase 5: Documentation & Handoff (Final 1‑2 weeks)
Goal
Ensure the client can run what you built without you and position for a retainer.
Key activities
- Operations manual creation (AI‑generated from implementation notes)
- Training sessions (recorded, with AI‑generated transcripts and summaries)
- 30/60/90‑day optimization checklist
- Retainer proposal (optional ongoing optimization support)
AI acceleration
The handoff documentation package—previously 2‑3 days of writing—now takes about 4 hours with AI assistance. I input project notes and outputs; the system drafts the full ops manual structure.
Typical duration
8‑15 hours
Deliverable
Operations Manual + Training Materials + Retainer Proposal
The Business Case for Repeatable Frameworks
With this framework I can now:
- Quote fixed‑price projects accurately (because I know what each phase costs)
- Set client expectations up front (they know exactly what they are getting)
- Delegate phases to subcontractors without rebuilding from scratch each time
- Identify upsell opportunities naturally (Phase 5 retainer conversion rate ≈ 60 %)
The framework also compounds over time. Every engagement adds examples, prompt refinements, and calibration data that make the next engagement faster.
If you want the full framework with the actual AI prompts I use in each phase, see the documentation at .
What does your consulting engagement structure look like? Do you run fixed‑price or time‑and‑materials? I’m curious how others handle scope management.