AI Business Case Template: What to Include
A defensible AI business case has nine sections. Missing any one of them is why most decks get bounced back for another round of analysis.
Problem, not solution
Open by naming a specific business problem and quantifying its current cost. 'We want to try GenAI' is not a problem statement. 'Contact-center handle time has risen 12% year over year and 38% of agent time is spent on repetitive lookup tasks' is. Reviewers who don't see a real problem in the first paragraph stop reading.
Opportunity, sized
State what becomes possible if the problem is solved, and put a number on it. Sizing does not have to be precise; it has to be defensible.
Proposed AI initiative — scope and approach
Describe what will actually be built or bought, why AI is the appropriate approach (versus process change or classical software), and what is explicitly out of scope. Reviewers reject business cases that hide scope.
Expected value in four categories, discounted
Productivity, direct cost savings, margin-adjusted revenue and risk avoidance — kept separate. Every category discounted by adoption and realization. Do not book gross theoretical value; every line should show its discounted number.
Investment — one-time and recurring
Split one-time from recurring. Include the frequently missed items — model/API consumption at production volume, monitoring, human oversight, ongoing governance. A three-year total exposes the recurring drag.
ROI summary — four numbers, all four
3-year net value, 3-year ROI %, payback in months, and benefit-cost ratio. Payback is often the number an executive actually cares about; leaving it out is why business cases stall.
Assumptions, exposed
List every material assumption — adoption rate, realization rate, ramp-up period, gross margin used, hours saved per user. A business case that hides assumptions is a business case that will lose the room.
Risks and mitigations, paired
Every named risk gets a named mitigation and a named owner. Unowned risks are how projects go sideways in month four.
Recommendation — with conditions
State a clear recommendation (Fund / Pilot / Investigate / Defer) and the conditions under which it holds. Reviewers reward a recommendation that survives its own caveats.
Turning this into a document
The paid PyratechAI Business Case Toolkit ships a populated Word template using this exact structure with the Customer Service Copilot worked example, and a matching PowerPoint deck with real charts. See /toolkit.
Put this into practice.
Run the numbers for your own initiative in the free AI ROI calculator.