What Does a Paid AI Discovery Sprint Include? How to Tell If It's Worth $5,000
Quick Take / Direct Answer
A legitimate AI discovery sprint for a professional services firm delivers five things: a working prototype (functional software on your real documents, not a mockup), a data architecture assessment, a prioritised use case list with ROI estimates per use case, a technical specification for the full build, and a fixed-fee project cost and timeline estimate. If you receive only a strategy deck or PowerPoint, you received consulting, not AI engineering.
Why a Paid Discovery Exists (And Why It Benefits You)
The AI consultancy market has two failure modes:
Failure mode 1: Firm signs a large contract for a full AI system without understanding their data landscape, their actual technical requirements, or whether the vendor can actually build what they are promising. The project drags, costs overrun, and the delivered system does not work as expected.
Failure mode 2: Firm spends months on unpaid discovery conversations with multiple vendors, receives conflicting advice, and makes no progress.
A paid discovery sprint — properly structured — solves both problems. It costs $4,000–$6,000 and produces working software in 3 weeks. The firm sees exactly what the AI does on their actual documents before committing to a full build. The vendor proves they can deliver before receiving a large contract.
The discovery fee should always be credited toward the full build if you proceed. Any vendor who does not credit the discovery fee is treating it as a profit centre, not a commitment mechanism.
What a Legitimate Discovery Sprint Includes
Week 1: Data landscape and use case mapping
- Access to your document systems and data sources (read-only)
- Inventory of document types, volumes, and current formats
- 2–3 stakeholder interviews (managing partner or CIO, lead practitioner, operations manager)
- Identification of the top 3 highest-value AI use cases based on data availability and operational impact
- Preliminary ROI estimates for each use case
Week 2: Prototype build
- Ingestion of up to 200–500 of your real documents into a working prototype system
- Private, web-based query interface allowing natural language search of your document sample
- Basic accuracy testing on 20 known-answer queries
- Integration scoping — documentation of what APIs and systems the full build needs to connect to
Week 3: Finalisation and delivery
- Accuracy report from prototype testing (with specific query examples and results)
- Technical specification document: architecture, integrations, infrastructure requirements
- Fixed-fee quote for full system build (not a range — a number)
- Timeline for full build with milestone schedule
- 60-minute presentation and Q&A session with your team
What You Should NOT Receive From a Discovery Sprint
- A slide deck with no working software
- A "roadmap" with no technical specification
- A cost estimate expressed only as a range without a basis for the range
- Recommendations to use specific tools without evidence from your actual data
- Any request to begin the full build without completing the prototype phase
Discovery Sprint Evaluation Criteria
Score your vendor's discovery on these five criteria:
| Criterion | What Good Looks Like | Red Flag |
|---|---|---|
| Working prototype | Functional system on your real documents | Mockup, wireframe, or demo on generic data |
| Data assessment | Specific findings from your data, not generic observations | Generic "your data needs cleaning" without specifics |
| Use case prioritisation | Named use cases with specific ROI calculations | Generic list of AI possibilities |
| Technical specification | Architecture diagram + integration list + infrastructure requirements | Vague description of what will be built |
| Fixed-fee quote | One number with itemised breakdown | Wide range or "we'll quote after more discovery" |
FAQs
Q: Should the discovery fee be credited toward the full build? A: Yes — at any reputable AI engineering firm. Govistudio credits the full discovery fee toward the build cost if you proceed within 60 days. This aligns incentives: we only earn full value if we deliver a discovery that convinces you the full build is worth doing.
Q: How long should an AI discovery sprint take? A: 3–4 weeks for a professional services firm. Longer than 6 weeks suggests the vendor is not working efficiently. Shorter than 2 weeks suggests they are not doing a proper data assessment.
Q: What questions should I ask an AI vendor before agreeing to a discovery sprint? A: Key questions: (1) What exactly will I receive at the end of the discovery — working software or a document? (2) Is the discovery fee credited toward the full build? (3) Can you show me a prototype you built for a similar organisation? (4) Who specifically will work on this discovery — a senior engineer or a junior analyst? (5) How do you measure accuracy on the prototype?