product-skill/ Field Guide/ Industry calibration Calibration

Seniority & Industry Calibration

Detect the reader's sector before you answer — startup advice applied to a regulated enterprise is confident, and wrong.

Pick the user's sector, adapt the defaults, and don't carry one industry's playbook into another. The gates below are load-bearing, not footnotes.

Same question

“What's the right default here?”

8 answers

B2B SaaSNRR as North Star; multi-persona buy, land-and-expand

B2CHabit & retention; viral loops, network effects

FintechCompliance-by-design; volume metric only behind Ring-1 floors

HealthcarePatient outcomes + clinical-workflow fit, not usage

MarketplaceTwo North Stars — supply & demand; liquidity first

Platform / APIDevEx & contract stability — breaking changes are forever

Hardware + SWHigher discovery bar — releases you can't recall

AI / MLEvals-as-QA owns quality; fairness is a guardrail

Pick the user's sector, then adapt — don't apply startup advice to enterprise.

Above the fold

The three calibration moves that matter most.

Detect the sector first.

Before recommending anything, name the user's industry and adapt to its defaults. The most expensive error is a fluent answer aimed at the wrong sector.

Low reversibility means Type-1 rigor.

In fintech and healthcare, decisions are one-way doors: staged approvals, pre-mortems, and Compliance in the room — gating, not advisory.

A volume metric needs quality floors.

Marketplace liquidity and fintech circuit-breakers exist so a raw-volume North Star can't be gamed against the customer. Ship the floor with the metric.

Per-sector defaults

Eight sectors. Eight sets of defaults.

Each card carries what to emphasize and — where it applies — the regulatory gate that changes how you ship. Pick the user's; don't apply startup advice to enterprise.

B2B SaaS

Multi-persona · land-and-expand

Multi-persona buy — buyer / user / admin each need a reason
Land-and-expand motion; watch implementation & onboarding cost
NRR as North Star — net revenue retention over raw signups

B2C consumer

Habit · network effects

Viral loops and habit formation drive growth
Engagement / retention and emotional design
Network effects and freemium conversion

Fintech / regulated

Compliance-by-design · low reversibility

Compliance-by-design, audit trails, trust/security as features
Low reversibility → Type-1 rigor + staged approvals
Distinct underwriting / actuarial / risk-management lens — not the same as compliance

Regulatory

For ML credit/risk: model-risk governance — drift, override rates, explainability, valid adverse-action reason codes; ship adverse-action notices on time under ECOA / Reg B.

Capital / liquidity & reserve-adequacy guardrails for balance/credit-holders. A volume NSM is defensible only if quality sits in circuit-breaker (Ring-1) floors derived via a Doshi pre-mortem with Compliance in the room.

Intl: PSD2 / SCA, FCA safeguarding + Consumer Duty, GDPR / data-privacy, local AML.

Healthcare

Patient outcomes · clinical safety

Patient outcomes, clinical-workflow fit, HIPAA / FDA
IRB / clinical-study design is a multi-month long-pole — budget it
Human-in-loop fallback, always

Regulatory

The 21st Century Cures Act CDS carve-out decides clinical-decision-support vs FDA SaMD. Sign BAAs with every subprocessor; PHI minimization + encryption in transit/at rest + access-control & audit logging.

FHIR / HL7 / SMART-on-FHIR interoperability is both a buyer requirement and a moat; distribution via EHR app-markets (Epic App Orchard) can substitute partner trust for your own evidence cycle.

Named clinical safety governance — CMO/CMIO sign-off + change-control so velocity doesn't outrun safety.

E-comm / marketplace

Liquidity first · two-sided

Value is bounded by the short side — min(supply,demand)
Two North Stars (supply + demand); GMV / take-rate
Two-sided trust & safety; search quality

Liquidity diagnostic

Diagnose the binding side via fill rate / seller utilization / time-to-match / zero-result-search %. Instrument per-segment liquidity before scoring (Wave 0), then sequence to the constraint (Theory of Constraints).

Platform / API

DevEx · contract stability

DevEx and API / contract stability — breaking changes are forever
Ecosystem health; docs-as-product
Partner economics

Hardware + SW

Long cycles · can't recall

Long cycles; manufacturing / firmware constraints
Releases you can't recall → higher discovery bar before commit

AI / ML

Evals-as-QA · responsible-AI

Evals-as-QA owns quality; test Model-Capability first on a golden set
Non-determinism, cost-per-call envelope, weekly failure-mode review
Data flywheel = product design

Responsible-AI pass

Model-metric literacy — precision / recall / F1 / ROC-AUC, basic MLOps — plus bias detection, fairness across segments, explainability (xAI), human-in-loop for high-stakes. Fairness is a guardrail metric, not a launch afterthought.

Calibration trap

In regulated sectors a raw-volume North Star (transactions, originations, GMV) reads as momentum but can be gamed against the customer. It's defensible only when quality lives in circuit-breaker floors — Ring-1 in fintech, the short-side liquidity metric in marketplaces — set before the metric ships, not after an incident.

Base rates

Three numbers that anchor realism.

Directional priors, not targets — each is point-in-time. Cite them to calibrate expectations, then verify the current figure before you lean on it.

~80%

of shipped features are rarely used → optimize adoption, not output count.

Pendo Directional · verify current

~80% / ~5%

GenAI: ~80% adopt but ~5% see EBIT/P&L impact → judgment is the scarce input.

McKinsey / BCG Directional · verify current

~5%

of large orgs hit target time-to-market at 40+ PMs → scale erodes speed by default.

McKinsey Directional · verify current