# AI Pilot Scoping Worksheet

A two-hour exercise to scope an AI pilot before you commit budget. Fill this out with the team that will use the tool, not just the team that will build it. Print the result and walk into your kickoff with shared assumptions.

> Brought to you by ByteWorthy. byteworthy.io

---

## Section 1: The problem (15 minutes)

**1. What specific task is the pilot targeting?**

> [Example: "Generating discharge summaries from clinical notes after each patient visit."]

Your answer:

---

**2. How many times per week is this task done today?**

> [Example: "Roughly 60 times per week across 4 providers."]

Your answer:

---

**3. How long does each instance take today (manual time)?**

> [Example: "10 to 15 minutes per summary."]

Your answer:

---

**4. What goes wrong with the current process?**

> [Example: "Summaries are inconsistent across providers, often missing key elements, and frequently delayed."]

Your answer:

---

## Section 2: Success criteria (20 minutes)

**5. What does success look like at the end of the pilot?**

> Be specific. "Better discharge summaries" is not specific. "70% of pilot summaries are approved by the supervising physician with zero edits, time per summary reduced from 12 to 4 minutes" is specific.

Your answer:

---

**6. How will you measure success?**

> What will you track? Where does that data live? Who is responsible for the measurement?

Your answer:

---

**7. What is the no-go signal?**

> What result would tell you to stop the pilot and not deploy? Decide this before you start so you can act on it without ego.

Your answer:

---

## Section 3: Constraints (20 minutes)

**8. What data does the AI need access to?**

> List every data source: EHR fields, document repositories, calendars, payer portals.

Your answer:

---

**9. Is any of that data sensitive (PHI, privilege, trade secrets)?**

> If yes, your stack choice is already constrained. Most cloud AI is off the table without a BAA.

Your answer:

---

**10. What systems must the AI integrate with?**

> EHR, billing, scheduling, internal tools. List read endpoints separately from write endpoints.

Your answer:

---

**11. What is the budget ceiling for the pilot?**

> Be honest with yourself. A $5,000 pilot is not the same project as a $50,000 pilot.

Your answer:

---

**12. What is the time horizon for the pilot?**

> Ideal: 4 to 8 weeks. If the answer is "as long as it takes," redefine before starting.

Your answer:

---

## Section 4: Stakeholders (15 minutes)

**13. Who reviews and approves before the pilot launches?**

> Name them. Not "leadership." A specific person.

Your answer:

---

**14. Who is the daily user of the tool?**

> The person who has to make this work in their actual workflow. Their feedback is the source of truth.

Your answer:

---

**15. Who reviews the output during the pilot?**

> Every AI output gets human review during a pilot. Who is on the hook for that review?

Your answer:

---

## Section 5: Path forward (20 minutes)

**16. Are you building, buying, or hybrid?**

> See our build versus buy primer (byteworthy.io/learn/primers/build-vs-buy-ai). Decide before you start scoping vendors.

Your answer:

---

**17. What is the kickoff date?**

Your answer:

---

**18. What is the pilot review date?**

> Calendar this now. Do not let the pilot drift. A defined end date is the friction that forces a decision.

Your answer:

---

**19. What is the criteria for moving from pilot to production?**

> What has to be true for you to commit to a full deployment?

Your answer:

---

**20. Sign off**

> Names of everyone who has read and agreed to this scope before kickoff:

---

## Quick reference

A pilot fails when:
- The team that uses the tool was not in the scoping conversation
- Success criteria were vague
- The no-go signal was not defined up front
- Budget and time horizon kept expanding silently

A pilot succeeds when:
- One person is the daily user and the source of truth on whether it works
- The success criteria are measurable and the measurement is automated
- The no-go signal is honored without ego
- The pilot ends on time, with a clear decision to deploy, iterate, or stop

## Want help running this exercise?

We facilitate this scoping in 00-discovery for every Build engagement. byteworthy.io/services/build