Three dimensions to review when implementing agents in a real process.
TL;DR
This page keeps its historical 10-20-70 URL. The percentages are not measured outcomes or a universal rule: effort is determined by each project’s scope and dependencies.
A pilot needs a baseline, acceptance criteria and accountable people. Choosing a model is one part of the work.
Review data, integrations, permissions, exceptions and adoption with the team. Diagnosis establishes a priced scope before work starts.
Working principle: define what success means, record what happens and use that evidence to decide the next step.
Define baseline, metrics, test cases and acceptance criteria before integrating agents.
Agree models, tools, permissions, memory, infrastructure and failure recovery around process needs.
Agree owners, human review, training, escalation, support and tracking of actual use.
Signals that need attention
Practices we look for
The URL is retained for existing links. Evaluation, technology and adoption are working dimensions, without claiming a universal ratio or exclusive methodology.
Diagnose one process, define a baseline and price the scope. Then design and build the agents, with timelines based on access, data and dependencies.
Evaluate the system and measure it against the agreed baseline over a period suited to the process. Expansion depends on results, acceptance criteria and a newly agreed scope.
The form prepares a conversation about scope, dependencies and metrics.
Start initial diagnostic →Related resources