Skip to content

Case study · AI sales assistant

Sidekick AI quote builder

An AI assistant that turns a conversation into an itemised project quote, with live prices from the catalogue, a branded PDF and an emailed quotation.

Client
Own project
Year
2026
Role
Product design, AI integration and development
Type
AI sales assistant
  • MinutesFrom first message to an itemised estimateInstead of waiting for a manual quote
  • 5Currencies quotedPHP locally; USD, GBP, EUR and AUD internationally
  • 1 stepFrom quote to branded PDF and emailSent to me and to the client together

The goal

Visitors asking "how much will my project cost?" usually wait days for an answer. I wanted them to get an honest, itemised estimate in minutes, while I still review every quote personally before work starts.

What I built

  • A chat assistant on OpenAI that asks focused questions and updates a structured quote each turn
  • A live estimate panel that recalculates as the conversation goes, with quick-reply suggestions
  • Three ways in: chat, a step-by-step configurator, or picking services from the catalogue
  • Branded PDF quotations, emailed to me and to the client in one step
  • Local rates in PHP for the Philippines and international rates in USD, GBP, EUR and AUD

Key decisions

The AI never sets the price

The model is instructed never to state prices. It only changes the scope of the quote through a validated tool call; the numbers are always calculated by the site from the catalogue, so the AI can't invent a discount or a wrong figure.

Structured output, validated

Each turn, the model must call a "respond" tool that returns a reply, a change to the quote and up to four suggestions. Every change is validated before it touches the quote, and anything invalid is ignored.

It keeps working when AI doesn't

Without an API key, or once a visitor or the whole site reaches its daily AI budget, Sidekick switches to a built-in demo engine instead of failing, so the quote builder is always usable.

Costs and abuse under control

Per-visitor limits, a site-wide daily AI budget, Turnstile bot checks and a cap on conversation length keep usage and spend predictable.