Problem Definition Document: Curator — operating system for coaching
Executive Summary
Professional coaches lose disproportionate time to the operational work around sessions — transcription, structured analysis, maintaining a per-client picture between sessions, scheduling across time zones, and drafting client follow-ups — work that today is manual or scattered across tools and lives mostly in the coach's head. For coaches carrying 6–15 active clients, this between-session load, not client count, is the binding constraint on capacity and on consistent quality. Curator's wager is that the durable value is a reproducible analysis methodology (anchored in the ICF frame, human-in-the-loop), not transcription, which is commoditized. Recommended next step: validate the operational-pain and reproducibility hypotheses on a 3–5 coach pilot before building automation.
Problem Statement (evidence-graded)
- Who: practicing 1:1 / team coaches, ICF-oriented, 6–15 active clients. (T5 — founder conversation + ICF Global Coaching Study cited on page)
- What they're trying to do: run high-quality sessions and keep continuity between them without drowning in admin.
- Impact: between-session operational work (notes, transcripts, prep, follow-up, multi-tool note-keeping) consumes capacity and erodes consistency. (T5/T6 — asserted, not yet measured)
- Frequency: every session, every client, continuously.
- Confidence: M-Low. Direction is credible; magnitude is unmeasured.
Root Cause Analysis (5 Whys)
- Why is coaching capacity/quality constrained? → Between-session admin is heavy.
- Why heavy? → It's manual and spread across calendar, transcribers, notes, messengers. (T6)
- Why spread? → No tool models the coaching workflow; generic tools don't know session phases or ICF structure. (T6)
- Why does that matter? → Generic AI output isn't reproducible or safe enough to hand a client. (T5)
- Root: no coaching-native system that produces a stable, methodology-grounded analysis a coach can trust and reuse. (T6, hypothesis)
JTBD Framing (primary)
When I finish a coaching session, I want a structured, trustworthy read of what happened and what's next, so I can prepare and follow up without hours of manual work — and prove progress to the client/sponsor.
Opportunity Sizing (assumption-flagged, not fabricated)
- Driver = coach-hours reclaimed per week × value of coach time × addressable coaches.
- All three inputs are unmeasured at concept stage. Declared unscorable in dollars until the pilot measures (a) baseline operational minutes/session and (b) minutes saved. See
05-discover-market-sizing.mdfor a structured, assumption-labeled TAM/SAM/SOM frame (not a validated number).
Stakeholder Impact Matrix
| Stakeholder | Problem felt | Current workaround |
|---|---|---|
| Coach (primary user/buyer) | Admin load, inconsistent analysis, lost continuity | Manual notes, separate transcriber, ad-hoc prompts |
| Client | Progress feels subjective; value of paid sessions not retained | Memory, personal notes |
| Program lead / corporate sponsor | No transparent ROI on coaching | Surveys, anecdote |
Constraint Map
- Technical: reproducibility of LLM analysis; accurate speaker separation; depersonalisation before external LLM.
- Methodological/ethical: must stay inside coaching scope (not therapy); ICF alignment; human validates before client sees anything.
- Regulatory: consent + PII handling (US-first: CCPA/CPRA per the studio page).
- Org/resource: concept stage, no pilot data, founder-led.
Sub-Problem Prioritization (ICE, qualitative)
- Reproducible session analysis — High impact / High confidence it's the moat / Med ease → first.
- Between-session continuity (client dossier) — High / Med / Med.
- Operational glue (scheduling, follow-up draft, payments) — Med / High / High (cheap wins).
- Live in-session suggestions — Med / Low / Low (sensitive, defer).
Recommendations (O→I→R→C→W)
- Objective: prove the pain is real and paid-for, and that analysis is reproducible.
- Indicators: baseline ops minutes/session; minutes saved; analysis stability across re-runs.
- Risks: pain is "nice-to-have"; reproducibility unattainable.
- Constraints: no fabricated metrics; pilot-gated.
- What's next: Discovery research (02) → reproducibility experiment design (10).
Self-Critique (≥3 genuine weaknesses)
- Entire problem rests on a single founder conversation + a public ICF study; no independent coach interviews yet (high single-source risk).
- "Operational pain" magnitude is unmeasured — could be a low-frequency, low-cost annoyance nobody pays to remove.
- The reproducibility claim is the moat and the biggest technical risk; if LLM analysis can't be stabilized, the differentiator collapses to "another transcriber."
- No counter-evidence sought (satisfied coaches who feel no pain) — a red flag.