Skip to content

borghei/Claude-Skills / project-management/discovery/identify-assumptions

identify-assumptions

Verified

Surfaces the assumptions behind a product bet, scores impact against confidence and picks the few to test first. Use when an idea feels right but is unproven, the team disagrees on risk, or before designing experiments.

examples/shared-dashboards-assumption-mapping.md

View on GitHub
# Example: Acme Analytics — Assumption Mapping for the Shared Dashboards Feature

> Real-world scenario showing how to apply this skill end-to-end.

## Context

Acme Analytics (Series-B B2B data platform, ~220 employees, ~3,200 paying accounts) is considering "Shared Dashboards v1" — a feature that lets a customer build a dashboard and share it with people outside their org via a public link with optional embedded login. Sales is pushing for it as a competitive table-stakes parity feature. The PM (Priya Rao) has 6 weeks before the team commits engineering.

Priya is using the identify-assumptions skill to surface and rank the assumptions underneath the feature. She is intentionally NOT scoping the PRD yet; she wants to know which assumptions are most likely to falsify the whole bet so she tests those first.

## Inputs

- Feature concept: Shared Dashboards v1 — public/auth-gated dashboard sharing
- Sales claim: ~40% of mid-market deals lose to a competitor at the dashboard-sharing question
- Engineering: 4-5 engineer-months estimated; touches the embed framework
- Existing product: dashboards work great internally; no sharing at all today
- Customer interview data: 14 interviews in the last quarter mentioned sharing in some form
- Team: Priya (PM), Liang (Designer), Reza (Engineer)
- Timing: 2 weeks for assumption mapping + risk validation, then commit/no-commit decision

## Applying the skill

1. **Ran devil's advocate from all three roles separately.** Each role wrote their own challenges before the group session. Group-mode devil's advocate produces 4-5 assumptions; silent independent mode produces 8-12.
2. **Categorized into the 4-category model.** Shared Dashboards is an existing-product feature, not a new product, so Priya used Torres' core 4 (Value, Usability, Viability, Feasibility). Did NOT use the 8-category extended set.
3. **Surfaced 8 distinct assumptions.** Some are obvious, some are non-obvious. The non-obvious ones (Viability A2, Usability A4) are exactly what assumption mapping is for.
4. **Ranked on Impact x Risk (likelihood of being wrong).** Pre-set scoring rubric before scoring; team scored independently then reconciled.
5. **Picked the top 3 to test first.** The remaining 5 are documented but parked.
6. **Tied each top-3 assumption to a specific lean experiment.** Assumption mapping that does not lead to experiments is theatre.

## The artifact

```
================================================================
  ASSUMPTION MAP — SHARED DASHBOARDS v1
  Owner: Priya Rao (PM, Activation)
  Drafted: 2026-05-22
  Decision date: 2026-06-12 (commit / kill / re-scope)
================================================================


PART 1 — FEATURE STATEMENT (used as the anchor for assumptions)

"Acme customers will be able to share a dashboard with people
 outside their account via either (a) a public link with no
 login, or (b) a magic-link login that creates a guest viewer.
 Shared dashboards will refresh on the same cadence as the
 source dashboard. Pricing: included in tier-2 and tier-3;
 limited to 1 shared dashboard on tier-1."


PART 2 — DEVIL'S ADVOCATE ROUND (silent, 25 min)

Each role challenged the concept independently, then shared.

PM Devil's Advocate (Priya):
  - Is there real demand or are we chasing competitor parity?
  - Will paying customers actually share more often than
    they downgrade?
  - Does sharing cannibalize tier-3 seat licenses?
  - Will this distract us from Activation, our actual
    quarterly goal?

Designer Devil's Advocate (Liang):
  - Will external viewers understand a dashboard built for
    internal context?
  - Is the sender confident the data they're sharing is
    correct AT the moment of share, not just when they built
    the dashboard?
  - Will "guest viewers" feel like real users or feel like
    a degraded experience that hurts our brand?

Engineer Devil's Advocate (Reza):
  - Do we have the row-level security primitives to make
    this safe at all?
  - What happens if a shared dashboard contains a viz that
    accidentally exposes PII from another customer (cross-
    tenant data risk)?
  - Can our embed framework scale to 10x the current load
    from external traffic?
  - Who maintains the guest-auth flow once we're past v1?


PART 3 — ASSUMPTIONS, CATEGORIZED

8 assumptions surfaced. Mapped to the 4 categories:

VALUE — Will customers want this?

  V1  "At least 25% of mid-market customers will share at
       least one dashboard within 60 days of the feature
       launching."
       Evidence today: 5 of 14 interviews mentioned wanting
       to share with stakeholders. Sales claim of 40% deal
       loss reason is unverified.

  V2  "Customers who share will value the feature enough
       that it materially affects their renewal."
       Evidence today: none. We are inferring from
       competitor presence in deals.

USABILITY — Can customers figure out how to use it?

  U1  "Sharers will be able to set the right permission
       level (public vs. magic-link) on the first try
       without help."
       Evidence today: none.

  U2  "External viewers will understand the dashboard
       context (filters, date ranges, definitions) without
       the sender having to explain."
       Evidence today: contradicting signal — internal
       Acme staff frequently misread other teams'
       dashboards.

VIABILITY — Can the business sustain this?

  V3  "Sharing will not cannibalize tier-3 seat licenses
       in a way that reduces ARR."
       Evidence today: unknown. CS leadership and finance
       have different intuitions.

  V4  "Engineering and ongoing maintenance cost (estimated
       at 0.5 engineer-quarters/year ongoing after launch)
       is justified by the upgrade and retention impact."

FEASIBILITY — Can we build this?

  F1  "Our embed framework can support 10x external traffic
       on the same shared dashboard without performance
       degradation."

  F2  "Row-level security primitives are sufficient to
       prevent any cross-tenant data leak via a shared
       dashboard."


PART 4 — SCORING (Impact x Risk)

Rubric (pre-defined before scoring):

  Impact: how much does this assumption being wrong damage
          the bet?
    5  Falsification = kill the whole feature
    4  Falsification = major re-scope
    3  Falsification = launch delay
    2  Falsification = mitigation in v2
    1  Cosmetic — does not change the path

  Risk: likelihood the assumption is wrong
    5  Strong contradicting signal or no evidence at all
    4  Weak supporting signal
    3  Mixed signal
    2  Strong supporting signal, minor gaps
    1  Validated

Each team member scored independently. Reconciled values:

                                    Impact  Risk  Score
  V1  25% of mid-market will share    4      3     12
  V2  Sharing affects renewal         5      4     20
  U1  Permission UI is intuitive      3      3      9
  U2  External viewers understand     4      4     16
  V3  No tier-3 seat cannibalization  4      3     12
  V4  Maintenance cost is justified   3      2      6
  F1  Embed framework scales 10x      4      2      8
  F2  No cross-tenant data leak       5      2     10

  (Note F2: Impact is 5 because a leak is brand-ending;
   Risk is 2 because Reza has strong evidence the existing
   primitives are sufficient. Score is 10 — important to
   verify but not a top-3 spend.)


PART 5 — RANKED LIST

  Rank  Assumption  Score  Category
  ----  ----------  -----  -----------
  1     V2          20     Value
  2     U2          16     Usability
  3     V1          12     Value
  4     V3          12     Viability
  5     F2          10     Feasibility
  6     U1           9     Usability
  7     F1           8     Feasibility
  8     V4           6     Viability


PART 6 — TOP 3 PICKED FOR VALIDATION

The team will validate the top 3 BEFORE engineering commit.
The other 5 are documented; we will only revisit them if
the top 3 pass.

Top 3:

  V2 (score 20) — Sharing will materially affect renewal
  U2 (score 16) — External viewers understand dashboard
                  context
  V1 (score 12) — 25% will share within 60 days


PART 7 — EXPERIMENT BRIEFS FOR THE TOP 3

[V2] "Sharing affects renewal"

  Experiment: 30-customer concierge MVP. PM manually
  produces shared dashboard links for 30 mid-market
  customers known to have external stakeholder needs.
  Track for 60 days:
    - Do they use it?
    - In renewal conversations, do they cite it?
    - In win/loss interviews on these accounts vs control,
      does sharing change the renewal narrative?
  Pass: in qualitative interviews, >=15 of 30 spontaneously
  reference the shared dashboard as part of their value
  perception. Document in dovetail.
  Fail: <5 of 30 mention it.

  Cost: ~12 hours of PM time over 60 days.
  Decision rule: PASS = green light. FAIL = kill the
  feature outright; the renewal signal isn't there.
  AMBIGUOUS (5-15) = run a structured renewal-conversation
  follow-up.

[U2] "External viewers understand the dashboard"

  Experiment: 8 unmoderated usability sessions on Loom or
  Maze. Show the same shared dashboard to 8 external
  viewers (recruited via UserInterviews; given the dashboard
  context only — no setup explanation).
  Ask: "what is this dashboard telling you?" Score whether
  the takeaway matches the dashboard creator's intent.
  Pass: >=6 of 8 produce the correct takeaway.
  Fail: <=3 of 8 produce the correct takeaway.
  Ambiguous: 4-5 of 8 = redesign of the external view.

  Cost: ~$800 + 4 hours.
  Decision rule: PASS = ship as scoped. FAIL = redesign
  external-viewer experience first (different UI, different
  context affordances). AMBIGUOUS = redesign smaller.

[V1] "25% will share within 60 days"

  Experiment: Fake door. Add a "Share dashboard" button to
  the dashboard view, route the click to a modal saying
  "Coming soon — early access waitlist." Track click rate
  by customer tier.
  Pass: >=25% click rate among mid-market.
  Fail: <10% click rate.

  Cost: 1 engineer-day. Already feature-flagged in adjacent
  code.
  Decision rule: PASS = strong demand signal. FAIL = no
  demand, kill feature. AMBIGUOUS = run sales-call follow-up
  to deal-loss claim.


PART 8 — WHAT WE ARE NOT TESTING (and why)

  V3 (cannibalization) — Cannot be tested before launch.
       Mitigation: launch with monitoring; CS + finance
       align on a roll-back threshold.

  F2 (no cross-tenant data leak) — Cannot be "tested" the
       way a value assumption can. Treated as an
       engineering security requirement. Reza commits to
       a threat model + penetration test BEFORE GA, but
       this is a build-time guarantee, not a validation
       step.

  U1 (permission UI intuitive) — Will be tested during
       implementation usability testing. Not blocking the
       commit decision.

  F1 (10x scale) — Engineering will build load test before
       GA. Not blocking the commit decision.

  V4 (maintenance cost) — Internal estimate, validated by
       Reza's engineering plan, not by experiment.


PART 9 — DECISION FRAMEWORK

  After 2 weeks of running the 3 experiments:

  ALL 3 PASS         Commit to engineering. Begin PRD.
  V2 fails           Kill the feature. The renewal signal
                     was the load-bearing assumption.
  U2 fails           Re-scope: external-viewer redesign
                     before any commit.
  V1 fails           Likely kill, unless sales can produce
                     specific deal evidence we missed.
  Any AMBIGUOUS      Run the documented follow-up.


PART 10 — WHAT THIS PROCESS PROTECTED US AGAINST

  - Building Shared Dashboards because sales said so,
    without testing whether customers actually share.
  - Skipping the external-viewer experience design until
    after engineering had locked the architecture.
  - Treating cross-tenant data risk as a product question
    rather than a build-time engineering requirement.
  - Confusing "feature in competitor product" with
    "feature that drives our renewals."
```

## Why this works

- **Three roles, independent surfacing.** PM, Designer, and Engineer each wrote challenges silently before sharing. This produced 8 distinct assumptions vs. the 3-4 a group-discussion session typically yields.
- **Used the right category model.** Shared Dashboards is an existing-product feature, so Priya used the core 4 categories (Value, Usability, Viability, Feasibility). The 8-category extended set is for new-product decisions — using it here would have produced noise.
- **Impact and Risk scored separately.** A common mistake is to score "importance" as a single number. Splitting impact (consequence of being wrong) from risk (probability of being wrong) is what surfaces the asymmetric cases — like F2 (cross-tenant leak), which is high impact but low risk because Reza has evidence the primitives work.
- **Top 3 testable in 2 weeks.** Each top-3 assumption has a concrete experiment, a cost, and a decision rule. Less-experienced teams produce a beautiful assumption map and then never test anything.
- **Decision rule before experiment, not after.** The PASS/FAIL/AMBIGUOUS thresholds for each experiment are written before the data arrives. This is the single biggest defense against post-hoc rationalization.
- **F2 (cross-tenant leak) handled as engineering requirement, not experiment.** Some assumptions are not validated via PM experiment — they are guaranteed by build-time controls. The map names which is which.
- **V3 (cannibalization) explicitly parked.** Cannot be tested pre-launch. The mitigation is monitoring + rollback threshold, not pre-launch validation. Naming this honestly is the discipline.

## What's next

- Each of the 3 experiments is designed in detail using [`../brainstorm-experiments/`](../brainstorm-experiments/).
- If all 3 pass, the PRD itself uses [`../../execution/create-prd/`](../../execution/create-prd/).
- A pre-mortem on Shared Dashboards using [`../pre-mortem/`](../pre-mortem/) will classify the parked feasibility assumptions (F1, F2) as Tigers / Paper Tigers / Elephants.
- For V3 (cannibalization), the post-launch monitoring metric is wired into [`../../execution/north-star-metric/`](../../execution/north-star-metric/) and [`../../execution/activation-funnel/`](../../execution/activation-funnel/).
- Cross-tenant data leak (F2) — treat as a compliance question, hand to security and audit teams; out of scope for product discovery.
- After 60 days post-launch (if shipped), use [`../interview-synthesis/`](../interview-synthesis/) on the concierge customers from V2's experiment.
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall