Product Case · Phase 3 of 3 · NoMoBo Case Series
Field access was solved. Completion still depended on manual template search, typing, and self-checking.
This is a Product Strategy Automation-at-Scale problem.
Portfolio learning exercise — a product specification only, not built or delivered. Builds on Phase 1 (real) and Phase 2 (illustrative). Every number below is a design target, not a measured result.
Phase 1 got assessors into the field with a working app. But finishing an assessment still meant hunting for the right template, typing everything by hand, and double-checking your own work before submitting. I specified an AI copilot layer — voice capture, smart templates, pre-submit validation, and adoption-risk alerts — designed to speed up completion without taking control away from the assessor.
The automation thesis
Every site, protocol, and assessor is a little different, so "automate it" was never one decision — it's three. I split the automation strategy into what should run without asking, what should be solved by configuration instead of inference, and what should always go to a person — and measured progress by whether abandonment held steady as coverage grew, not by how much got automated.
Two things field access didn't fix
Template selection and long-form typing still slowed assessors down, and manual capture kept transcription errors alive even after field access was solved.
Managers had no way to see where adoption was still breaking down without digging through raw data.
What we specified
- A smart template recommenderSuggests the right template by GPS and history, replacing manual search.
- Voice-to-form captureAssessors speak observations; the copilot maps them to structured fields.
- A pre-submit validation agentChecks completeness before submission — the assessor keeps final say.
- Adoption-risk scoringFlags assessors likely to abandon before they do, not after.
- A manager insights copilotPlain-language queries over field adoption data, no custom reports needed.
What scalable automation means here
Template pre-fill when GPS and assessment history match a known site/protocol pair with high confidence. No confirmation step, full audit log.
Required-field sets and validation thresholds that differ by site type or jurisdiction — the variation lives in a rules engine, not in a model guessing at intent.
New site types, conflicting history, or ambiguous GPS/history matches route to the assessor or manager, and get logged for pattern mining, not silently retried.
The same three-lane split used to decide what to automate, configure, or escalate; the target is holding abandonment rate steady as coverage grows, not raising an automation-coverage number for its own sake.
What escalation patterns taught us
- Recurring escalations became configuration, not exceptionsA cluster of escalations from one region's site type turned into a new configuration template class, instead of a growing list of special-cased overrides.
- Ambiguity got a name, not a confidence-tuning loopRepeated low-confidence GPS/history matches at multi-protocol sites became a distinct "site ambiguity" flag with its own lightweight resolution flow, rather than chasing a better model score.
- One-off escalations stayed manual, on purposeEscalations that never recurred were left for a person to handle each time — automating a case that happens once isn't scale, it's overhead.
Transformation scale
- 7 AI-assisted capabilities specified end-to-end
- 1 orchestrator coordinating 6 specialised agent tools
- 4-phase rollout roadmap defined — Foundation → Capture → Intelligence → Admin
A specification for faster, more confident field capture, with a human always in the loop.
What it's targeted to move
Business impact (targeted)
- +15–20% targeted lift in field activation
- Additional 20–30% targeted reduction in time-to-value beyond Phase 1
- Abandonment rate targeted to hold at ≤7%, vs. 74% pre-redesign baseline
- +10% targeted lift in sustained 30–90 day submission rate
- Full audit trail and human-in-the-loop guardrails specified for every AI suggestion