Building AI search across everything your company knows
This workbook is your step by step guide to deciding what your AI should answer from, testing it until you trust the answers and putting it live in stages. By the end you'll have a content map, a real test set with a pass bar and a rollout plan, with every answer cited back to its source.
Casa Maestri
Kamiki
Part 1: Building your company brain
Turn "I should really get AI doing this" into a plan you can hold any builder to. Three steps, each with worksheets you fill in for your own business.
Map the hours spent hunting for answers, and what wrong or slow answers cost you.
The source, access and testing rules any AI search must pass, then a test set built from your real questions.
Go from a read-only pilot to customer-facing answers in four stages, each one working before the next.
Worksheets, not theory
We built a legal education company its own AI search after its vendor shut down: 88% pass on a 261-question test, zero hard failures.
pass on a 261-question test set, with zero hard failures
of speakers identified automatically across 566 programs
processing cost per video
Teams tired of answering the same questions
- Companies with knowledge spread across documents, videos, tickets, wikis and inboxes.
- Support and sales teams answering the same questions from scattered sources.
- Education, training and content businesses with large libraries of video or courses.
- Teams whose AI vendor shut down, raised prices or never got past the demo.