What a channel costs, what content returns, and how a meeting is actually produced. Spend figures on every result.
Why a lifecycle programme is safer when ingestion and activation share an engine, an illustrative programme walked through, and the freshness exit condition most programmes lack.
A tool listed in an agent's context is a mention. A tool the agent calls is a citation. Measure them apart and the case for a narrow, role-scoped Salesforce MCP server surface becomes a number.
The deploy is the easy part. A constructed routing change, traced from pull request to the meetings that arrive weeks later, and the holdout that decides whether it worked.
Trial lengths differ by vendor, from 14 to 30 days. Run every Salesforce DevOps free trial on the same 14-day clock, count admin hours as spend, and write down what the result cannot tell you.
Why monthly active rows pricing is hard to forecast, how published credit tiers change the budget conversation, and a three-month replay any team can run before choosing.
Half of the eight vendors read on 1 Oct 2026 publish a number a reader can check. The hypothesis: an answer engine can cite that half on price and only mention the rest.
Follow one constructed Salesforce team from trial to month 12 as it grows from 3 to 12 users, and watch per-seat and flat price curves separate month by month.
A paid-test design for AI Salesforce change management: two arms, 20 change requests, an illustrative EUR 1,980 in staff time, confounds it cannot remove, and what a win would not license.
A page can be fetched by AI agents four hundred times and never cited. Agent traffic measures access, and access is only the first of three signals.
A corpus built in a burst gets cited for a while and then decays. What separates durable content from a campaign is who is still maintaining it.
A homepage claim repeated near-verbatim across five capture dates isn't evidence of an AI search citation strategy working. It's evidence the copy didn't change — a different, and cheaper, finding.
A dashboard reports a citation rate to the decimal. The query set behind it is a few hundred prompts at best, and that sets how much of a movement is real.