10 KiB
Customer
Public-clean version. Field recordings, raw transcripts, named actors, and operational vote deltas live only under
/private/on local checkouts (gitignored). Cloud agents: use this document; if a task would produce sensitive deliverables, report them in chat and do not commit them.
Kernel (day-to-day)
- Who: the General Coordinator of a federal-deputy campaign in Bahia (~45 days, scarce resources, information scattered across WhatsApp / spreadsheets / memory). Job: one shared, current operational picture so limited effort goes to the right territories and people.
- Mental model (observed use): Action → Place → Who — never entity-first. The coordinator does not scroll by instinct; an invisible confirmation (Save / Approve) means the action was not completed; a fragmented form means abandonment.
- Relative reading of the board: % of the candidate’s own vote (concentration of own capture) is the hard priority criterion — absolute statewide % is an anti-metric. Projected floor is on the order of prior-cycle performance with a stretch target above it.
- Threat #1: friendly fire inside the same party (a co-religionist ahead). Real team channel: WhatsApp 1:1 without dated audit trail.
- Dominant constraint: “legs” / field structure, not money.
- Competence anxiety (“I can’t handle the tool”) is the worst emotional problem — the product must give a sense of control, not only data.
Job Statement
When I am coordinating a statewide federal-deputy campaign across dozens of Bahia municipalities in roughly 45 days, with scarce resources and information scattered across advisors and local leaderships, I want a shared, current operational picture so I can put limited effort into precise mobilization instead of reconciling WhatsApp, spreadsheets, and memory.
(PT, field language:) Quando preciso coordenar uma campanha de deputado federal pelo estado da Bahia — votos e estrutura em dezenas de municípios, ~45 dias, pouco recurso, informação espalhada entre assessores e lideranças — quero enxergar o mesmo quadro operacional e agir no território/pessoas certas sem caçar atualizações, para transformar esforço limitado em mobilização precisa.
Job Dimensions
| Dimension | Description | Where the app underdelivers today |
|---|---|---|
| Functional | See municipalities (predefined territories), base, declared-vs-estimated votes, and agenda in one place; record updates and decide the next action | Entity-first screens invert the Action → Place → Who model; discovery requires scroll; confirmations invisible; fragmented forms. Hit: filtered “no advisor” lists × prior-cycle votes. Jargon still blocks. |
| Emotional | Feel in control and trust that the picture is current | Worst today. Competence anxiety (“I can’t see / handle it”), not only stale data. Without action-first flows, Little Hire does not start. |
| Social | Be seen as serious, aligned coordination | Real work still “wins” on WhatsApp; tool users can look bureaucratic while informal channels set the pace |
Competing Alternatives
| Alternative | Why hired today | Weakness |
|---|---|---|
| WhatsApp groups / 1:1 | Fast, everyone is already there | Fragments; no analysis; useful history disappears |
| Spreadsheets (projection maps) | Flexible; one advisor’s “source of truth” | Divergent versions; weak on phone |
| Notebook / memory / phone calls | Zero tool friction | Doesn’t scale; doesn’t share |
| Other campaign platforms (past attempts) | Promise of a single system | Team stops updating → stale data → abandonment |
| Non-consumption | Avoids another login | Burns the 45-day window |
Big Hire vs Little Hire
| Moment | What it is here | Status / intervention |
|---|---|---|
| Big Hire | First login, trust the system enough to start | Mitigated offline: in-person onboarding; seed initial data when needed |
| Little Hire | Come back and update so the shared picture stays alive | Primary leak. Past tools died here. Reduce emotional cost of “updating again today” |
Forces of Progress
| Force | Notes |
|---|---|
| Push | Fragmentation; statewide scope; 45-day crunch; cannot follow dynamics by hand |
| Pull | One operational picture; smarter targeting; instrument the team can update; territory dossier before agenda |
| Habit | WhatsApp + spreadsheet as default — “Zap is the field”; projection sheet is the morning ritual |
| Anxiety | “If others don’t update, I’m wasting time”; competence anxiety; LGPD / real-data hold; “legs” constraint |
Opportunity Solution Tree (sanitized)
Outcome: weekly Little Hire
├── O1 I don’t trust others updated → anxiety / abandonment
├── O2 Day’s delta (call/Zap) does not enter the decision queue
├── O3 Jargon + opaque AI block action
├── O4 Updating on phone is slower than voice on Zap
├── O5 “Truth” still lives in the projection spreadsheet
├── O6 Arrive in territory without a dossier
├── O7 Vote gains/losses lack dated / auditable record
├── O8 Team in isolated islands — information does not circulate
├── O9 Action-first mental model vs entity-first UI
├── O10 Discovery must be above-the-fold — no instinctive scroll
└── O11 Invisible confirmation / commit (Save / Approve / Done)
Product direction: tables = bulk bank management; redesign daily UX as continuous linear flows. Observed-session originals: private/plans/ (local only).
Leap-of-Faith Assumptions
| Assumption | Type | Cheap test | Success if |
|---|---|---|---|
| Seeded picture + onboarding → returns within 7 days | Desirability / Little Hire | Count logins + ≥1 real update after action-first redesign | ≥1 spontaneous update in 7 days |
| Action queues answer “where do I start?” | Usability | Filtered “no advisor” × prior-cycle votes | Finds queue/action without help |
| Entity-first “edit where you see” is discoverable | Usability | Observed: editable cells not discovered in free exploration | Opens vote adjust without demo — FAILED; pivot O9 |
| Inline glossary / plain language cuts confusion | Usability | Field jargon needed live translation | Zero “what is…?” in 2nd session after copy |
Discovery method (no raw evidence here)
Field discovery used Mom Test story blocks + observed product use. Full verbatim transcripts, audio, and named operational deltas are not in this public tree. Local agents: see private/customer/CUSTOMER.md, private/transcribe/, private/general-coordinator-interview/. Cloud agents: stay on this clean doc; escalate sensitive outputs in chat.
Validated patterns (safe to build on)
- Priority is share of the candidate’s own capture, not local electorate %.
- Leadership value is often political representativeness, not raw vote count.
- Friendly fire (same-party competitor) is the named #1 threat class.
- Channel of truth for day deltas is voice/Zap without dated tool trail — until the product wins Little Hire.
- Dominant constraint named in field: structure/“legs”, with money as instrument — not always the binding constraint.
- Map is briefing/panel, not the daily operating surface; filtered lists for “who is unmonitored” are the observed value hit.
- Tables remain for bulk; day-delta work needs action-first chassis.
Segments & Best-Fit Customer
- Primary:
coordinator("Coordenador Geral"),candidate("Candidato"), andadvisor("Assessor") operating/campanhaunder time pressure — advisor scoped to administered municipalities. - Secondary:
leader("Liderança") in the field (PWA) — phone-first supporter contact tool only. - Out of scope for this improve cycle: public site visitors / donors.
Links
- Architecture / campaign model:
ARCHITECTURE.md - Agent ops (incl.
/private/rules):AGENT-OPS.md - Local sensitive originals:
private/(gitignored; local checkouts only)