Defining Jobs To Be Done

Why this matters

Clear Jobs To Be Done (JTBD) keeps AI products focused on real customer progress, not features. For AI Product Managers, JTBD anchors discovery, scoping, metrics, and experimentation.

• Prioritize: Decide which customer struggles justify AI vs. simpler automation.
• Scope: Write solution-agnostic acceptance criteria and data requirements.
• Measure: Define outcome metrics before model selection.
• De-risk: Align stakeholders on the customer job, not the algorithm.

Concept explained simply

JTBD describes the progress a user wants to make in a situation, independent of your product. A simple job story template:

When [situation], I want to [motivation/struggle], so I can [desired outcome].

• Functional job: The core task (e.g., assess risk).
• Emotional job: How they want to feel (e.g., confident, not rushed).
• Social job: How they want to be perceived (e.g., competent to peers).

Desired Outcome Statements (DOS) make jobs measurable. Use a direction + metric + object + context:

• Minimize time to draft a customer reply for high-priority tickets.
• Increase recall of relevant policies in compliance reviews.
• Reduce variance in forecasts for long-tail SKUs.

Mental model

Think of a pipeline:

1. Situation triggers → 2. Struggling moments → 3. Job story → 4. Desired outcomes (measurable) → 5. Acceptance criteria → 6. Data and constraints → 7. Candidate solutions (AI or not).

Quick self-check: Is your job story strong?

• Solution-agnostic (no references to models, features, or UI).
• Specifies a situation, not just a persona.
• Contains a measurable outcome or a path to one.
• Connects to a business outcome you can track.

Practical framework and template

1. Define the situation
   • When/where does the struggle happen? What triggers it?
   • Who is present? What tools or data are available?

   Mini task: Write one sentence starting with “When …”.

2. Capture the struggle
   • What slows them down or creates risk?
   • What trade-offs are they making today?

   Mini task: Write “I want to …” without referencing features.

3. State the desired outcome
   • What does “better” look like? How would they know?

   Mini task: Finish “so I can …” with a measurable result.

4. Write Desired Outcome Statements (DOS)
   • Use verbs like minimize, reduce, increase, improve, avoid.
   • Attach a metric, baseline, and target range when possible.

   Mini task: Draft 3 DOS. Example: “Reduce average review time from 45m to <20m within 3 months.”

5. Acceptance criteria (solution-agnostic)
   • Define observable behaviors or thresholds, not model names.
   • Include quality bars and guardrails (safety, compliance).
6. Data and constraints
   • Data needed, availability, quality checks, privacy rules.
   • Latency, cost, and interpretability constraints.
7. Only then: Candidate solutions
   • Consider AI, rules, UI flows, or process changes.

Copy-paste JTBD template

Job story: When [situation], I want to [struggle/motivation], so I can [desired outcome].

Desired Outcome Statements (3–5):
- [Direction] [metric] of [object] in [context]
- [Direction] [risk/error] for [segment]
- [Direction] [time/cost] while maintaining [quality/safety]

Acceptance criteria:
- Success threshold(s): …
- Guardrails: …
- Observability: …

Data/constraints: Sources, freshness, privacy, latency, cost.

Candidate solutions: AI / rules / process / UI.

Worked examples

1) Customer Support: Response drafting

Job story: When I receive a high-priority ticket with a long conversation history, I want to quickly understand context and propose a correct reply, so I can resolve the case fast without missing policy details.

Desired outcomes:

• Minimize time to first draft from 10m to <2m.
• Reduce policy violations in replies to <1%.
• Increase customer satisfaction (CSAT) on these tickets by +0.3 points.

Acceptance criteria (solution-agnostic):

• Drafts reference correct order/account info 95%+ of the time in audit samples.
• Drafts include links to relevant policy sections or quoted policy text.
• Latency to draft < 5s for 95th percentile.

Data/constraints: Access to ticket history, policy corpus, PII protection, redaction for training data.

Candidate solutions: AI summarization + drafting; or rule-based templates + dynamic merge fields. JTBD helps compare both.

2) Fintech Risk: Transaction review

Job story: When a transaction is flagged as suspicious, I want to assess risk quickly with explainable evidence, so I can make a defensible decision and minimize false positives.

Desired outcomes:

• Reduce manual review time from 8m to <3m.
• Reduce false positives by 20% without increasing false negatives beyond 2%.
• Increase proportion of reviews with documented rationale to >99%.

Acceptance criteria:

• Every recommendation includes top 3 contributing factors with human-readable reasons.
• Audit log captures inputs, versioning, and decision maker notes.
• Model suggestions must be overrideable with rationale.

Data/constraints: Transaction features, graph data, explainability requirement, latency < 2s, regulatory retention.

3) Retail: Demand planning

Job story: When planning inventory for seasonal items, I want reliable demand projections with uncertainty ranges, so I can place orders that avoid stockouts without overstock.

Desired outcomes:

• Reduce MAPE for seasonal SKUs from 28% to <18%.
• Reduce stockouts by 30% for top 100 seasonal SKUs.
• Provide 80% prediction intervals per SKU-week.

Acceptance criteria:

• Forecasts include P10/P50/P90 and SKU-level feature importance.
• System flags data gaps and outliers before forecast generation.
• Weekly re-forecast completes within 30 minutes for 10k SKUs.

Data/constraints: Sales history, promotions, weather, holidays; cost ceiling; interpretability for planners.

Exercises (hands-on)

Do these before the quick test. Keep outputs short and solution-agnostic.

1. Exercise 1 (mirrors ex1)

   Scenario: A B2B sales rep prepares for a call with a new lead after a long email thread and attachments.

   • Write 1 job story.
   • Write 3 Desired Outcome Statements with tentative metrics.
   • Draft 3 solution-agnostic acceptance criteria.
2. Exercise 2 (mirrors ex2)

   Scenario: A healthcare claims reviewer must decide if a claim requires additional documentation.

   • Write 1 job story.
   • List required data and constraints (privacy, latency, explainability).
   • Write 2 guardrail criteria.

Quality checklist for your answers

• Job story avoids feature/model mentions.
• Outcomes use clear direction verbs and metrics.
• Acceptance criteria are observable and testable.
• Data/constraints cover privacy, safety, and latency.

Common mistakes and how to self-check

• Jumping to solutions: Mentions of “LLM”, “classifier”, or “chatbot” in the job story are red flags.
• Persona-only framing: “For analysts” is not a situation. Add triggers and context.
• Vague outcomes: Replace “better/faster/smarter” with target ranges or proxy metrics.
• Metric tunnel vision: Balance speed with quality and safety guardrails.
• Ignoring data reality: Validate data access, freshness, and quality before committing.

Self-audit

• Can a stakeholder read your JTBD and imagine multiple solutions?
• Do outcomes connect to a business KPI you can measure?
• Are acceptance criteria independently checkable by QA?

Who this is for

• AI Product Managers and aspiring PMs shaping ML/AI features.
• Data Scientists and Designers collaborating on discovery.
• Founders and PMMs defining outcome-first product narratives.

Prerequisites

• Basic understanding of product discovery and stakeholder interviewing.
• Awareness of metrics and experimentation (A/B or offline evaluation).
• High-level knowledge of AI capabilities and limitations.

Learning path

1. Learn JTBD basics and draft job stories.
2. Translate to Desired Outcome Statements and acceptance criteria.
3. Validate with users/stakeholders; refine metrics and constraints.
4. Only then explore candidate solutions and feasibility.
5. Set up measurement and guardrail monitoring.

Practical projects

• Create a JTBD dossier for one workflow at your company: job story, outcomes, acceptance criteria, and data map.
• Run a 3-interview discovery sprint to validate the job and sharpen outcomes.
• Design a solution-agnostic experiment plan comparing AI vs. non-AI baselines against the same JTBD outcomes.

Next steps

• Complete the quick test to validate understanding.
• Apply the template to one real use case this week.
• Share your JTBD and outcomes with a partner for feedback using the checklist above.

Note: The quick test is available to everyone. If you log in, your progress and test results will be saved.

Mini challenge

Pick any recurring decision in your product (e.g., prioritizing a backlog, triaging reports). Write:

• 1 job story,
• 3 Desired Outcome Statements with rough targets,
• 3 acceptance criteria and 2 guardrails.

Tip if you get stuck

Ask: What makes this task slow, risky, or frustrating today? What would a confident, repeatable result look like tomorrow?