# SaidWell

> Voice-rep practice for the founder-sales calls you keep losing.

SaidWell is a voice-based sales-rehearsal app for technical founders who run their own sales. You voice-act a 3-minute scenario (investor pushback, demo discovery, pricing, asking for the close), get scored on five fixed rubric dimensions with the exact phrases you said cited as evidence, and watch the trend across 14 days.

## Core Concept

Founder-led sales fails under time pressure, not at the desk. Reading text-based AI feedback on a typed-up pitch is not the same skill as carrying a 5-minute call without flinching. SaidWell closes that gap by making you actually speak under a 3-minute clock and scoring what you actually said against a consistent rubric, so the trend across reps is meaningful.

The wedge over a do-it-yourself Claude project is three things structured together: voice instead of text (it reproduces the freeze that typing cannot), a fixed rubric every rep (so the trend is comparable), and a 14-day trend (so you measure behavior change, not a single-rep reaction).

## How It Works

1. Pick a scenario. Seven founder scenarios cover the calls that fund the company: investor pushback, demo discovery, pricing, asking for the close, and more.
2. Record. You speak under a hard 3-minute clock. Single take, no editing. The clock is part of the point.
3. Get scored. The transcript is scored on five fixed dimensions, each with a verbatim phrase you said as the citation, the moment in the audio it happened, and a one-line rationale.
4. Watch the trend. Five sparklines track each dimension across 14 days, with a focus suggestion for the week.

## The Rubric (the five fixed dimensions, scored every rep)

These five are locked and identical on every rep, which is what makes the trend mean something. Source of truth: lib/ai/rubric/dimensions.ts.

- Directness: how often you hedged versus made a clear, owned claim.
- Discovery: whether you asked before telling, instead of pitching blind.
- Concision: whether your answer length matched the question, or over-explained past it.
- Opening: first-sentence quality, anchored to the outcome the buyer cares about, not the mechanism.
- Frame: whether you held your own frame or accepted the questioner's.

Each dimension returns a 0 to 10 score, a verbatim citation, the citation's audio timestamp, and a short rationale. No streaks, no badges, no "great job" overlays. Just the number and the evidence.

## Pricing

- Free tasting: 1 rep, lifetime, per account. There is no recurring free tier.
- Pro: $19/month or $190/year (about two months free). Includes up to 5 reps per day.

## Tech Stack

- Next.js 16 (App Router, Partial Prerendering)
- React 19 (Server Components, Server Actions)
- TypeScript (strict mode)
- Supabase (PostgreSQL, Auth, Storage)
- Drizzle ORM
- Tailwind CSS v4 + shadcn/ui
- OpenAI Whisper (transcription) + Anthropic Claude (rubric scoring)
- Stripe (subscriptions)
- PostHog (analytics)
- Vercel (hosting)

## Links

- Marketing: https://saidwell.app
- App: https://web.saidwell.app
- Pricing: https://saidwell.app/pricing
- About: https://saidwell.app/about
- Sign up: https://web.saidwell.app/signup

## Company

Operated by Fantastic Online Stores PTY LTD, Australia.