AI PM Ready

Scoping AI products

Vague requirements produce vague AI outputs. Scoping is where PM craft translates business intent into a specification a model can be built and evaluated against.

The AI scoping canvas

Seven dimensions every AI feature scope must address. Leaving any one out creates an undefined gap that engineering will fill with their own assumption — which may not be the right product decision.

Business Goal + KPI

What business problem are we solving and how will we measure success? This must be a specific, measurable KPI — not "improve CX" but "reduce first-contact resolution time from 6.2 min to 4.0 min."

Real example

Salesforce Einstein: business goal was to reduce manual data entry for sales reps. KPI: time spent on CRM data entry per rep per week. Baseline was 2.1 hours. Target was <45 minutes.

Common trap

Defining the goal in AI terms ("build a classification model") rather than business terms. If you cannot state the business KPI, you cannot evaluate whether the AI succeeded.

Worked example — customer support ticket router

All 7 canvas dimensions applied to a realistic AI feature.

Business GoalReduce manual routing time by 30% (from avg 4.2 min to <3 min per ticket). Secondary: reduce misrouting rate from 18% to <5%.
AI TaskGiven customer support email body (max 2,000 chars), classify into: Billing / Technical / Returns / Sales / General. Output: JSON { department, confidence }.
Data Strategy50,000 labeled historical tickets. Labels reviewed by 3 senior agents — inter-annotator agreement >90% required. PII stripped. 80/10/10 train/val/test split.
Eval PlanPrimary: top-1 accuracy ≥ 88% on 5,000-item golden test set. Secondary: precision ≥ 0.85 per category. A/B test: routing error rate in treatment vs. control for 4 weeks.
Human-AI TeamingL2 at launch: AI classifies, agent confirms before routing. Agent can reclassify — corrections logged as training signal. Move to L3 when 4-week accuracy ≥ 91% sustained.
EthicsAudit classification accuracy by customer language and account tier to detect disparate impact. Bias threshold: <3% accuracy gap across groups.
TechnicalCloud-hosted fine-tuned BERT. P50 <100ms, P99 <400ms. Cost <$0.001/classification. Fallback: rule-based keyword routing when model unavailable.
Was this page useful?