ValidIQ — Discovery Confidence Scorer

Dimension 1 of 4

Problem Validation

How well have you established that the problem is real, widespread, and worth solving?

Describe the problem you are solving Be specific about who experiences the problem, what breaks today, and what it costs them. Vague problem statements like "customers want better service" score low.

One or two sentences. What breaks today, for whom, and at what cost?

Customer interviews conducted Count only interviews with real users or stakeholders, not internal team discussions. 4+ interviews is the minimum for a defensible discovery package at a stage gate.

None yet 1 to 3 interviews 4 to 7 interviews 8 or more

Pain signal strength Validated means you have data quantifying the problem: ticket volumes, time-on-task logs, error rates. Anecdotal means a few people confirmed it in conversation. Assumed means you believe it exists but have not yet confirmed it.

Assumed — we believe it exists Anecdotal — a few people confirmed it Validated — backed by data

Segments covered in interviews Target segment means the people who directly experience the problem your feature solves. Interviewing only adjacent teams like internal ops or leadership does not substitute for target segment coverage.

Select all that apply

Target segment — the people most affected by the problem Internal ops or adjacent teams Business sponsors or leadership Compliance or risk stakeholders No interviews conducted yet

Evidence types used Structured evidence means data sources beyond interviews: support tickets, NPS scores, transaction logs, behavioral analytics. Two or more structured types signals a serious discovery effort.

Select all that apply

User interviews Support tickets or complaint data NPS or CSAT data Transaction or behavioral data No structured evidence yet

Please complete all fields before continuing.

Dimension 2 of 4

Value Clarity

Is your value hypothesis explicit, testable, and connected to a financial outcome?

State your value hypothesis A strong hypothesis names a specific behavior change and connects it to a measurable financial outcome. "AI will improve efficiency" is not a hypothesis. "Automating X will reduce analyst time-on-task by Y%, unlocking Z" is.

What does this feature change, and what does that change unlock?

Behavior change identified What will users start doing, stop doing, or do differently because of this feature? A clearly defined behavior change means you can measure whether the feature is working after launch.

Yes — clearly defined Partially — directional but not specific Not yet defined

Hypothesis status Evidenced means you have data or research confirming the value claim. Partially evidenced means you have proxies or directional signals. Assumed means the hypothesis is logical but untested.

Assumed — no external validation Partially evidenced Evidenced — validated with data or research

Primary financial unlock type Choose the category that best describes how this feature creates financial value. If you cannot identify which category applies, that is itself a signal the value hypothesis needs more work.

Please complete all fields before continuing.

Dimension 3 of 4

Cost Awareness

Have you modeled what this AI feature will cost to run at scale, including agentic loop costs?

Feature type A simple AI feature makes one inference per user request. An agentic workflow makes multiple calls, uses tools, and loops until a condition is met. Agentic workflows carry significantly higher and less predictable token costs.

Simple AI feature — single inference per request Agentic workflow — multi-step, tool calls, loops

Token cost modeling Cost modeling means estimating tokens per request multiplied by expected volume and frequency to produce a monthly cost envelope. A rough estimate is better than nothing but a full model includes worst-case scenarios.

Done — full cost model built Rough estimate only Not done yet

Estimated monthly active users at scale Use your target steady-state volume, not current pilot numbers. This is the scale at which the cost model needs to hold for the ROI case to be defensible.

Interaction frequency per user Higher frequency multiplies token cost directly. A feature used multiple times daily at high volume needs a cost model that explicitly accounts for this, not a back-of-envelope estimate.

Refinement loop costs accounted for Agentic workflows re-call the model until an output condition is met. McKinsey data shows these loops account for 60% of total agentic token spend on average. If you have not modeled loop iterations, your cost estimate is likely less than half the real number.

Agentic workflows re-process outputs until a condition is met. These loops can represent 60% of total token cost.

Yes — loop costs are included in my estimate Not accounted for Not applicable — this is a simple AI feature

Please complete all fields before continuing.

Dimension 4 of 4

Stakeholder Readiness

In regulated financial services, unvalidated stakeholder assumptions are as dangerous as unvalidated customer assumptions.

Stakeholder engagement status In financial services, compliance and risk engagement is not optional. "In progress" means active dialogue has started. "Confirmed" means you have documented alignment. "Not engaged" means you have not started the conversation.

For each stakeholder group, select the current engagement level

	Not engaged	In progress	Confirmed
Compliance
Risk team
Business sponsor

Regulatory constraints flagged Common constraints in financial services include PSD2, GDPR, SR 11-7 model risk guidance, and FCA rules on automated decisioning. Flagging a known constraint and naming it specifically scores higher than flagging it without detail.

No known regulatory constraints for this feature Yes — specific constraints identified

Name the specific regulatory constraint

E.g. PSD2 real-time reporting obligations, GDPR data minimisation, SR 11-7 model risk guidance

Please complete all fields before submitting.