Visa APM - Product Judgment Lab
Earn the next product decision before you build.
Start with a case, then move into the prompting toolkit. The two areas open as separate pages so the case context stays clear and the prompt flow stays focused.
01 - Case studies
Choose the problem that will stretch your judgment.
These are fictional cases. Read one case in full, name the solution you are tempted to jump to, then move to the toolkit to research, frame, develop, and test a better next move.
Case 01 - Embedded B2B payments inside the workflow
Pay Where Work Happens
A visible client request and competitor signal are creating pressure for a 90-day pilot concept. Understand the workflow, authority, exception paths, and segment differences before treating embedded payments as the answer.
Starting situation
FlowForge is a fictional vertical-SaaS platform used by 48,000 U.S. construction and maintenance businesses. Customers already manage jobs, purchase orders, supplier invoices, and approvals there. Three of FlowForge's ten largest customers have asked to pay suppliers directly inside the platform. A fast-growing competitor is promoting "instant supplier payments with real-time reconciliation." The CEO wants a pilot concept in 90 days.
What you know at the start
- There is visible client pull and a competitive signal.
- The workflow already contains purchasing, invoice, and approval activity.
- The request combines product, payment, supplier, integration, and operational questions.
- A 90-day expectation creates pressure to define the problem quickly.
Research runway and public starting points
Use public material to understand how embedded commercial payments fit into real business workflows, what different parties value, and what system conditions might matter. Do not treat an existing Visa or competitor capability as the answer.
Case 02 - Agentic procurement under real-world guardrails
Agent on a Leash
A competitor demo has accelerated an autonomous purchasing request. Distinguish work that is truly repetitive from work that merely looks repetitive, then make authority, uncertainty, reversibility, and exception handling explicit.
Starting situation
BistroCloud is a fictional procurement and back-office platform for regional restaurant groups. Its CEO has seen a competitor demo in which an AI agent monitors inventory, reorders routine supplies, chooses a payment method, and pays automatically. The request: autonomous procurement for routine purchases under $5,000, with no human click, before the holiday period.
What you know at the start
- The sponsor sees repetitive purchasing and approval work as an automation opportunity.
- The proposed threshold is $5,000, but the case has not established what "routine" means.
- An agent could affect buyers, finance, suppliers, payment partners, and operations.
- The launch ambition is time-sensitive and highly visible.
Research runway and public starting points
Use the public agentic-commerce landscape to understand product questions around identity, authority, trust, and automation, without assuming this case should mirror an existing Visa product.
Case 03 - Cross-border payouts when speed, certainty, and control collide
Friday Night Funds
A headline promise of "faster" hides different outcomes: arrival time, status visibility, usable funds, FX clarity, exception handling, and control. Map the lifecycle and actors before turning all of that into a speed feature.
Starting situation
CraftLink is a fictional marketplace connecting agencies with independent studios and freelancers in 40+ countries. Sellers complain that international payouts can feel slow and opaque, particularly over weekends. A competitor advertises "global payouts in under 60 minutes." CraftLink asks for instant, 24/7 cross-border payouts with transparent FX.
What you know at the start
- Seller frustration exists, but the exact drivers have not yet been separated.
- The proposed promise combines timing, status, FX, recipient experience, and cross-border movement.
- The product must work across multiple actors and endpoints.
- A competitor promise is creating urgency.
Research runway and public starting points
Use public money-movement material to understand how payouts can move across accounts and wallets and how a transaction lifecycle extends beyond a single API request. Focus on actors, states, trade-offs, and uncertainty; do not try to become a payments engineer.
Case 04 - Developer experience from first test to live launch
Sandbox Is Not Production
A successful sandbox call can make the next step feel obvious while leaving launch readiness unproven. Distinguish information gaps, sequencing gaps, approvals, and specialist decisions before treating an AI copilot as the answer.
Starting situation
HelioPay Network is a fictional global payments platform with a public developer program. Fintech partners can often make a successful sandbox API call quickly, yet production launches take longer than expected and create repetitive support work. The Head of Developer Products proposes an AI Integration Copilot and a goal to cut time-to-production by 50%.
What you know at the start
- The visible pain is time from early integration work to live launch.
- The proposed solution is an AI copilot, but the source of delay has not been established.
- Developers are only one part of the launch system.
- The case sits at the intersection of developer experience, product, implementation, and production readiness.
Research runway and public starting points
Use public developer material to understand the shape of a developer journey across sandbox, certification, and production. Do not infer confidential Visa architecture or internal process.
02 - Prompting toolkit
The original flow, sharpened for Product Judgment.
Eight core prompts keep the original Discover, Define, Develop, Deliver structure. One pre-mortem joins Stage 3 to pressure-test the options before a team chooses what to advance. Every prompt includes the framework definition it needs; web access is optional, not assumed.
Discover
Understand the landscape before defining the problem. Look beyond the obvious solution, keep the customer plural, and use external forces only when they could change a product decision.
Prompt 1.1
Doblin's 10 Types of Innovation
Where might you create value beyond the obvious product feature?
Product Judgment lens: Do not let the first feature request define the problem. The useful output is an option map, with evidence gaps and alternative mechanisms, not a recommendation to build the most familiar thing.
- "Which type are we overlooking because it does not look like a product feature?"
- "What evidence would make our preferred innovation type less plausible?"
- "Turn the strongest opportunity into a one-week learning test."
Prompt 1.2
Persona and Deep Empathy
Understand the person, the recent episode, and the tension - not a fictional profile.
Product Judgment lens: A plausible persona is not evidence. Use this to separate what is observed about an actor from what is inferred, then design questions that could change your mind.
- "What would this actor do, not merely say, if the problem became urgent?"
- "Which assumption in this profile could invalidate the HMW?"
- "Create a second guide for the person who owns the exception."
Prompt 1.3
PESTLE Horizon Scan
What external forces could make today's answer fragile tomorrow?
Product Judgment lens: This is not a generic trend list. Every force must alter a decision, dependency, adoption condition, trust boundary, or test. Do not generalise from a city, country, or region without a source.
- "Which local assumption would be most expensive if wrong?"
- "Show only forces that could change our next 90-day decision."
- "What source is authoritative enough to settle this question?"
Define
A brilliant solution to the wrong problem is still a failure. Stress-test the How Might We statement until a real actor, evidence, tension, and boundary are visible.
Prompt 2.1
The HMW Stress Test
Human, specific, evidence-based, assumption-free, and ambitious enough.
Product Judgment lens: A coherent HMW can still hide an unsupported leap. The test requires evidence, exposes the solution smuggled into the question, and states what would make the team rewrite it.
- "Which phrase in this HMW is doing the most unsupported work?"
- "What would the person least represented in this workflow say we missed?"
- "What observation would make us rewrite this HMW?"
Develop
Go wide before going deep. Borrow mechanisms, generate genuinely different routes, then use the pre-mortem to surface what could make the strongest-looking option fail.
Prompt 3.1
Forced Associations: Borrowing Brilliance
What principle from another domain could create a different product bet?
Product Judgment lens: Borrow a mechanism, not a brand's feature. The point is to expose a different theory of the problem and test the limit of the analogy before you copy anything.
- "What is the limit of analogy we are most likely to ignore?"
- "Combine two mechanisms, then name the new risk created."
- "Which idea changes the theory of the problem, not just the interface?"
Prompt 3.2
Crazy Eights: 25 Solutions
Quantity before quality. Five categories of genuinely different moves.
Product Judgment lens: Go wide, but prevent 25 variations of the same feature. Each category changes the mechanism, workflow, service, partner, or theory of the problem.
- "Which ideas move risk or work elsewhere rather than reduce it?"
- "Pick three ideas that operate on different theories of the problem."
- "Which bolder bet is still testable without a production commitment?"
Prompt 3.3
Product Judgment Pre-Mortem
Assume the project failed. Make failure useful now.
Use the original stance: This is not a risk checklist or an approval exercise. First imagine the bet failed, then define the bounded conditions under which it could become a trusted, high-value capability. The final human judgment cannot be delegated.
- "Which failure mechanism is specific to our bet rather than generic AI or product risk?"
- "What must be true before we increase autonomy, scale, data access, or customer exposure?"
- "Which red flag would make us pause even if the sponsor wants to continue?"
Deliver
You have options. Now sort, test, and explain them without creating false certainty. Choose a provisional bet, state what it depends on, and make the next learning move clear.
Prompt 4.1
How / Now / Wow / Chow Sorting
Not every good idea is a ready idea. Sort by impact, originality, feasibility, and evidence.
Product Judgment lens: Sorting is a conversation, not a magic score. The prompt requires the team to separate a high-impact idea from a well-supported, testable one and to name the assumptions hidden in feasibility.
- "Which idea looks feasible only because a dependency is invisible?"
- "What would move the strongest HOW idea toward a bounded test?"
- "What evidence would turn the top WOW idea into a no-go?"
Prompt 4.2
Pitch Sharpener
Make the reasoning clear, the uncertainty honest, and the ask easy to evaluate.
Product Judgment lens: Do not make the idea "undeniable." Make the logic traceable. A good pitch leads with the decision, states what is evidenced and unknown, and tells the audience what would change the recommendation.
- "Roleplay the skeptical risk, operations, or technical owner."
- "Rewrite this for a different decision-maker without hiding the unknowns."
- "Which single sentence would most damage trust if it proved untrue?"