Skip to main content

​
What this guide is for

Experio answers questions from a client’s own data. It does that by building a knowledge graph: a connected model of the client’s clients, projects, people, contracts and everything else that matters to them. It builds the graph from their documents and their business systems. The software reads files and tables. It can’t guess the business rules behind them: which things matter, how they connect, which lists the firm already uses, when two names mean the same thing. An implementation captures those rules in a series of workshops and then enters them in Experio. That work is what makes the answers accurate. This guide is the playbook for that work. It explains how Experio works, defines the building blocks, and lays out an 8–10 week plan with a workshop for each decision. Every workshop page includes an agenda, the questions to ask, and a worked example.

​
Who it is for

Super users: you don’t need to be technical. Each workshop page says what you’ll be asked to decide and what to prepare. After handover, you’ll use the same pages to maintain the configuration.

​
The approach in one picture

The method starts from the questions the business wants answered and the agents it wants to run. It works backwards to the data model, then builds forwards to a tested graph. Three ideas run through the whole guide:
  1. Questions drive the model. Every entity type, relationship and attribute should trace back to a question someone actually asks. If no question needs it, leave it out of the pilot.
  2. Tables first, then documents. Business systems (HR, CRM, finance) supply the reliable master list of clients, people and projects. Documents then add detail to those records instead of creating near-duplicates. See How Data Gets to the Graph.
  3. Measure, then tune. Accuracy is checked against a list of golden questions with known answers, agreed in the first workshop. Tuning continues until the must-have questions pass.

​
How the guide is organised

Start Here

How Experio works, how data reaches the graph, and definitions of every building block.

Plan

The 8–10 week roadmap, phases, deliverables, roles and permissions.

Workshops

Ten workshops, from questions and agents to access control, each with agenda, script and example.

Validate & Launch

Pilot ingestion, accuracy validation, go-live and handover.

Worked Example

Northbridge Consulting, a fictional professional-services firm, followed from start to finish.

Templates

Copy-ready worksheets, agendas and checklists for every step.

​
What “done” looks like

An implementation is complete when:
  • The must-have golden questions for each use case pass (a common target is 80% or more), and every answer cites its sources.
  • The agreed agents (agent flows) produce usable output for pilot users.
  • The full agreed corpus is ingested, and Model & Define > Compatibility shows no invalid configuration.
  • The client super user can run the routine work alone: the conflict review queue, taxonomy refreshes, new data sources and small model changes.
  • The decision log, the ontology rationale and the golden-question scores are handed over.
  • Admin Guide: reference for every admin page. This guide links to it for the click-by-click detail.
  • User Guide: what end users see. Use it for training at go-live.
  • Implementation Playbook: a one-page summary of each decision and the admin page it belongs on.