An AI discovery sprint is the single most valuable thing an enterprise can do before committing to any AI build. In 4-6 weeks, it transforms "we should do something with AI" into a prioritised roadmap with specific capabilities, data requirements, and business cases.
What You Need to Know
- A discovery sprint is not a proof of concept. It doesn't build anything. It maps the territory, identifying where AI creates the most value in your specific business and what's needed to capture it.
- The output is a prioritised AI roadmap with business cases, data readiness assessments, and a recommended sequence. It's the strategic document that justifies and guides your AI investment.
- Most enterprises need 4-6 weeks and $20-50K. The sprint pays for itself by preventing the most common (and expensive) mistake: building the wrong thing first.
- Discovery should involve your people, not just consultants. The domain experts who do the work know where the friction is. The sprint extracts and structures that knowledge.
- 50% of the discovery cost can typically be credited toward the build phase if you proceed within 90 days. Discovery isn't a sunk cost. It's an investment in doing the build right.
3×
average cost reduction when AI initiatives follow a structured discovery vs starting directly with a build
Source: RIVER Group, enterprise engagement data, 2023-2024
The Four-Phase Framework
Phase 1: Landscape Mapping (Week 1)
Goal: Understand the current state: processes, data, systems, pain points.
Activities:
- Stakeholder interviews (6-10 sessions, 45-60 min each). Talk to the people who do the work: operations managers, team leads, process owners. Not just IT. The business.
- Process mapping for 3-5 high-potential areas. Document the current workflow, decision points, handoffs, and bottlenecks.
- Data audit. Where does the data live? What format? How accessible? What governance exists?
- Technology audit. Current systems, integration points, existing automation.
Key output: A map of your enterprise's knowledge friction points: where knowledge gets stuck, decisions get delayed, and smart people do dumb work.
Who's involved: 2-3 discovery team members + your stakeholders (operations, IT, leadership).
Phase 2: Opportunity Identification (Week 2)
Goal: Identify and describe specific AI capabilities that address the friction points.
Activities:
- Workshop 1: Opportunity brainstorm. With cross-functional stakeholders. Start with the friction map from Phase 1. For each friction point, ask: "If AI could handle any mechanical step, access any knowledge, and work at any speed, what would we build?"
- Opportunity shortlist. Typically generates 10-20 candidates. Shortlist to 6-10 based on initial feasibility assessment.
- Capability descriptions. For each shortlisted opportunity: what the AI does, what inputs it needs, what outputs it produces, who uses it, what systems it integrates with.
Key output: A catalogue of 6-10 specific AI capabilities with clear descriptions.
Phase 3: Scoring and Prioritisation (Week 3)
Goal: Rank capabilities by a consistent framework and identify the optimal starting point.
Activities:
- Five-factor scoring for each capability: business impact, data readiness, process clarity, foundation potential, organisational readiness.
- Workshop 2: Scoring session. With domain experts and leadership. Score each capability collaboratively. The discussion reveals assumptions and builds alignment.
- Foundation analysis. Which capabilities share infrastructure? What's the optimal sequence for compound value? Where does building a shared foundation create the biggest acceleration?
- Dependency mapping. Which capabilities depend on others? What needs to be true before each can start?
Key output: A ranked list of capabilities with composite scores and a recommended sequence.
Phase 4: Roadmap and Business Case (Weeks 4-6)
Goal: Produce the strategic document that guides investment decisions.
Activities:
- Roadmap development. Phased plan showing which capabilities to build, in what order, with what dependencies.
- Business cases. For the top 3-5 capabilities: expected investment, timeline, projected value, risk assessment.
- Foundation architecture. High-level design of the shared infrastructure components and how they serve multiple capabilities.
- Governance framework. Initial governance requirements and a plan for developing them.
- Workshop 3: Roadmap presentation. Present findings and recommendations to leadership. Align on priorities and next steps.
Key output: The AI Discovery Report, a strategic document that includes the prioritised roadmap, business cases, data readiness assessment, foundation architecture, and governance plan.
What Makes a Good Discovery
Domain experts in the room. The people who do the work know where the value is. Consultants who map processes from documentation miss the real friction: the workarounds, the tribal knowledge, the things that "we just know."
Honest data assessment. Don't assume data is ready. Verify it. A discovery that claims "data readiness: high" without actually checking will produce a roadmap that fails on first contact with reality.
Foundation thinking from day one. Every capability should be evaluated not just on its standalone value, but on what it builds for the capabilities that follow. The best first capability isn't always the highest-value one. It's the one with the best combination of value and foundation potential.
Realistic business cases. Don't inflate projections to justify investment. Realistic numbers build trust and set achievable targets. Overpromising is the fastest way to create AI cynicism.
The discovery sprint isn't about finding a use case for AI - it's about finding your organisation's most valuable knowledge problems and figuring out which ones AI is the right tool to solve.
Isaac Rolfe
Managing Director
What Happens After Discovery
If you proceed: The build phase starts with the top-ranked capability. The discovery roadmap guides the work: capability scope, data requirements, integration points, and success metrics are already defined. This typically saves 3-4 weeks of the build phase (which is why discovery investment credits make sense).
If you don't proceed: You still have a strategic document that guides future AI decisions. The roadmap doesn't expire. Your opportunities don't change that quickly. When you're ready, you can pick up where the discovery left off.
If you partially proceed: Some enterprises run discovery and then build internally, using the roadmap as their guide. Others build capability #1 with a partner and take over for capabilities #2+. The discovery works for any execution model.
- Can we run a discovery internally?
- Yes, if you have someone with both AI expertise and enterprise delivery experience. The risk with internal discovery is confirmation bias. Teams tend to validate their existing assumptions rather than challenge them. An external perspective brings fresh eyes and cross-industry patterns.
- What's the difference between a discovery sprint and a proof of concept?
- A POC builds one thing to prove it works. A discovery maps the entire landscape to determine what to build first and in what order. A POC without a discovery often builds the wrong thing. A discovery without a subsequent build is still valuable strategic intelligence.
- How do we know if we need a discovery or can jump straight to building?
- If you can answer these three questions confidently, you can skip discovery: (1) What's the highest-value AI capability for our business? (2) Is the data ready? (3) What shared infrastructure should we build alongside it? If any answer is unclear, run the discovery.