Resources · Playbook
How to scope an AI transformation.
Answer
Answer: a real AI transformation starts with a four-to-six week diagnostic that maps the company's processes and data, identifies the workflows where AI creates measurable value, scores each opportunity by value, feasibility and risk, and produces a prioritised roadmap with costed pilots — not a slide deck of buzzwords.
Why scoping is the work
Most failed AI initiatives don't fail in engineering. They fail in scoping — the wrong workflow was chosen, the data wasn't ready, success wasn't defined, or leadership never agreed on what "transformation" actually meant. Treating scoping as a serious project, not a workshop, is the single highest-leverage decision a company can make about AI.
The five outputs a scoping exercise must produce
By the end of a properly run AI scoping engagement, the leadership team should have:
- A process and data map of the business as it actually operates today.
- A long-list of AI opportunities, sourced from interviews and observed bottlenecks.
- A short-list scored on Value, Feasibility and Risk.
- Reference architectures and build-vs-buy recommendations per opportunity.
- A 12–18 month roadmap with costed pilots, owners and decision gates.
Anything less is consulting theatre. Anything more and you've quietly started the project before the business signed off on it.
How to rank opportunities
Use a single matrix. Score every candidate workflow against three axes: Value (revenue, margin, time saved), Feasibility (data quality, technical complexity, change management) and Risk (regulatory, reputational, model failure). Build first where the three axes line up — high value, high feasibility, low risk.
| Tier | Value | Feasibility | Risk | Action |
|---|---|---|---|---|
| Quick win | Med–High | High | Low | Pilot in 30–60 days |
| Strategic build | High | Med | Med | Production project, 3–6 months |
| Bet | Very High | Low | Med–High | Time-boxed feasibility study |
| Defer | Low | Any | Any | Document and revisit in 12 months |
Where most teams get scoping wrong
Three failure modes recur. The first is starting from the technology — "let's use an LLM somewhere" — instead of from the highest-leverage workflows. The second is treating data readiness as a footnote; in practice the data work often dominates the build and must be costed up-front. The third is skipping change management; an AI system that the team won't use is worse than no system at all.
A serious diagnostic forces all three to the surface in week one, not in production.
The six-week diagnostic, week by week
Weeks 1–2 — Discovery. Executive interviews, functional deep-dives and a structured walk of the company's highest-leverage processes. The output is a process and data map of how the business actually operates, not how the org chart says it does. This is the phase that surfaces the bottlenecks leadership already suspects but has never named precisely.
Weeks 2–3 — Long-list. Every candidate workflow is documented with the data it depends on, the volume it handles, the cost it absorbs and the failure modes it produces today. The long-list is intentionally generous: the discipline comes in the scoring, not in the brainstorming.
Weeks 3–4 — Scoring. Every candidate is scored against the Value / Feasibility / Risk matrix with evidence — not with show-of-hands voting. Data readiness is assessed in detail and costed; change-management exposure is named honestly; vendor and build options are compared on total cost of ownership over 24 months, not on sticker price.
Weeks 4–5 — Architecture and build/buy. For each short-listed opportunity, a reference architecture is sketched, the right blend of automation and AI agents is chosen, and a build-vs-buy recommendation is made with rationale. This is where most engagements quietly drop ambitious-sounding ideas that do not survive contact with the data or the integration surface.
Weeks 5–6 — Roadmap and decision gates. The output is a 12–18 month sequenced roadmap with costed pilots, named owners, success metrics and clear stop/continue gates after each phase. Leadership leaves the engagement with decisions made, not options to debate.
What a serious diagnostic costs you to ignore
The honest case for scoping is economic, not philosophical. A mid-sized company that skips the diagnostic typically spends six-figure sums automating the wrong workflow, discovers the data isn't ready in month four, and rebuilds in month twelve with a different vendor. The diagnostic is a small fraction of that cost and removes the failure modes that produce it. Treating it as optional is the most expensive decision in the project.
How this connects to Raise AI Ltd
Built AI offers this exact engagement as a fixed-fee, time-boxed commercial product: the AI Transformation Audit. It is the standard entry point for clients who want a serious roadmap before committing to a build. Once the roadmap is set, delivery moves to the AI Project Studio.
FAQ
How long does a real AI scoping exercise take?+
Four to six weeks for a mid-sized company. Less and you produce a wishlist; more and you've started the project. The deliverable is a prioritised, costed roadmap, not a slide deck.
Who needs to be in the room?+
An executive sponsor, the heads of the affected functions, IT/data leadership and a process owner from each candidate workflow. Without these people the audit produces opinions, not commitments.
What's the most common scoping mistake?+
Starting from the technology — 'where can we use an LLM?' — instead of from the workflow. Real scoping starts from the highest-leverage processes in the business and asks where AI removes the bottleneck.
What if leadership disagrees on priorities?+
Good. The audit is partly a forcing function. The Value/Feasibility/Risk matrix gives leadership a shared, evidence-based way to settle disagreements before the build starts.
Can we skip the audit and start building?+
You can. Most companies that do end up rebuilding within 12 months because they automated the wrong workflow, picked the wrong architecture or under-scoped the data work. The audit is cheaper than that rebuild.