AI Project Discovery: 20 Questions Before You Start Building

Why Discovery Decides Outcomes

The 2026 data is clear: most failed AI projects failed at discovery, not at engineering. The team did not know what success meant, what data they had, what compliance applied, or what stakeholders needed. Building the wrong thing well is still failure.

This piece is the working list of 20 questions every AI project should answer at discovery.

The 20 Questions

flowchart TB
 D[Discovery] --> B[Business: 1-5]
 D --> U[User: 6-9]
 D --> Tech[Technical: 10-13]
 D --> R[Risk: 14-17]
 D --> O[Org: 18-20]

Business (1-5)

1. What is the measurable business outcome? (Specific metric and target)
2. What is the dollar value of success? (Saving X, earning Y)
3. What is the cost of doing nothing? (Status quo vs intervention)
4. Who owns the P&L impact? (Single accountable person)
5. What is the timeline for measurable impact? (Pilot vs production milestones)

User (6-9)

6. Who is the user, specifically? (Persona, role, context)
7. What problem are they currently solving how? (Existing workflow)
8. What does their day look like with this AI? (Specific scenarios)
9. What is success from the user's perspective? (Their criteria, not yours)

Technical (10-13)

10. What systems must this integrate with? (List with API surfaces)
11. What data is available? (Sources, freshness, quality)
12. What is the latency budget? (User-experienced latency)
13. What is the volume? (Per day, peak vs average)

Risk (14-17)

14. What is the worst-case failure? (Concrete scenarios)
15. What compliance applies? (HIPAA, EU AI Act, sector rules)
16. Who reviews high-stakes outputs? (Human-in-the-loop strategy)
17. What is the audit requirement? (Logs, retention, access)

Org (18-20)

18. Who builds and owns this long-term? (Team)
19. What governance applies? (Approval gates, ongoing oversight)
20. What is the deprecation plan? (When this is sunset, how)

Why Each Matters

flowchart TD
 Q[Unanswered question] --> Risk[Project risk]
 NoMetric[No metric] --> Drift[Goal drift]
 NoData[No data] --> Stuck[Engineering stuck]
 NoComp[No compliance check] --> Block[Late-stage block]
 NoOwner[No owner] --> Cancel[Cancellation]

Each unanswered question is a future risk. Discovery surfaces them while they are cheap to handle.

The Discovery Workflow

For a typical project:

1. Stakeholder kickoff (collect first answers)
2. User research (validate user-side questions)
3. Technical discovery (data, integrations)
4. Compliance review (legal / risk team)
5. Spec writeup
6. Stakeholder sign-off

Total: 2-6 weeks for non-trivial projects. Skipping it doesn't save time; it shifts cost to engineering and beyond.

What Good Answers Look Like

For "What is the measurable business outcome?":

• Bad: "Improve customer experience"
• Good: "Increase first-call resolution rate from 65 percent to 80 percent within 6 months, measured by post-call disposition + CSAT survey"

The good answer is testable; the bad one is decoration.

When Answers Aren't Available

Some questions don't have crisp answers at discovery time. The discipline:

• Mark them as open
• Define what would make them answerable
• Schedule the work to answer
• Don't pretend the answer is "TBD"

What CallSphere Asks

For client deployments, we run a 90-minute discovery call covering all 20 questions. Clients sometimes resist; we have learned to insist.

The clients who push back hardest at this stage are typically the ones whose projects have the most discovery gaps. The clients who answer cleanly typically have successful deployments.

Red Flags at Discovery

flowchart TD
 Red[Red flags] --> R1[Vague success criteria]
 Red --> R2[No clear owner]
 Red --> R3[Compliance "we'll figure it out"]
 Red --> R4[Volume estimate is wishful]
 Red --> R5[Timeline driven by external pressure not feasibility]

Each is a sign discovery is incomplete. Push back; do not start building until they are addressed.

What Comes After

Once discovery is complete, the project moves to:

• Spec writing (covered elsewhere)
• Stakeholder sign-off
• Engineering kickoff
• Pilot phase

The 20 questions are revisited at the end of pilot to confirm assumptions held.

Sources

• "Discovery in product management" SVPG — https://www.svpg.com
• "AI project discovery" Anthropic engineering — https://www.anthropic.com/engineering
• "Why AI projects fail" BCG — https://www.bcg.com
• "Effective project discovery" PMI — https://www.pmi.org
• "Pre-mortem" research — https://hbr.org

Where this leaves operators

If "AI Project Discovery: 20 Questions Before You Start Building" reads like a prompt for your own roadmap, it usually is. The teams winning the next two quarters aren't the ones with the loudest demos — they're the ones who have wired AI into the parts of the business that compound: pipeline coverage, NRR, CAC payback, and time-to-onboard. That means picking a bounded use case, instrumenting it from day one, and refusing to ship anything you can't measure within a single billing cycle.

When AI infrastructure pays back — and when it doesn't

The honest test for any AI investment is whether it compounds. Models, prompts, fine-tunes, and slide decks don't compound — they decay the moment a new release ships. What compounds is structured data on your actual customers, evals tied to revenue events (not BLEU scores), and agents that get better as more conversations land in your warehouse.

That's why the operating model matters more than the tech stack. CallSphere runs on 37 specialized voice agents, 90+ tools, and 115+ Postgres tables across six verticals — but the reason customers stay isn't the count. It's that every call writes to a CRM event, every event feeds a sentiment model, and every sentiment score routes the next call through an escalation chain (Primary → Secondary → six fallback numbers). The infrastructure does the boring, expensive work of making each interaction worth more than the last.

For most B2B operators, the right sequence is unambiguous: pick one funnel leak (inbound qualification, demo no-shows, win-back, expansion), wire an agent into it for 30 days, and measure ACV influence and NRR delta before touching anything else. Logos and category-creation slides are downstream of that loop, not upstream.

FAQ

Q: Is there a meaningful risk of getting ai project discovery: 20 questions before you start building?

Explore a live demo and compare current plans to find the right fit for your business.

Q: What's the failure mode when ai project discovery: 20 questions before you start building?

Measure two things and ignore the rest at first: a primary outcome (booked appointments, qualified pipeline, recovered reservations) and a guardrail (containment vs. escalation, sentiment, AHT). Anything else is dashboard theater. The most common pitfall is shipping without an eval set — once you have 50–100 labeled calls, regressions stop being invisible and prompt iteration starts compounding instead of going in circles.

Q: How does this connect to ACV, NRR, and category positioning?

ACV moves when the agent influences deal velocity (faster qualification, fewer demo no-shows). NRR moves when the agent owns expansion-trigger calls (renewal, usage-spike, success outreach). Category positioning is downstream — buyers don't pay for "AI-native" framing, they pay for a reproducible motion. CallSphere pricing reflects that ladder: $49 Act, $99 Orchestrate, and $149 Custom Build, billed monthly, with the same 37-agent / 90+ tool stack underneath each tier.

Talk to us

If any of this maps onto your roadmap, the fastest path is a 30-minute working session: book on Calendly. You can also poke at the live agent stack at realestate.callsphere.tech before the call — it's the same infrastructure customers run in production today.