Trelium Insights
Ideas that move enterprise AI forward
Perspectives on AI agents, automation design, and building for the enterprise from the team at Trelium.
How to Run a Useful AI Pilot in an ERP or Salesforce Sandbox
A sandbox where every customer has one address and every field is filled in will make any AI pilot look brilliant. Here's how to make sandbox data resemble real work, record what the sandbox can't show, and prepare a launch decision IT and security can sign.
Measuring Branded Merchandise Campaigns Beyond Impressions
The report said 2,000 water bottles had generated countless impressions, and the CFO asked what that was worth. Here's a practical way to set an objective, an attribution window, and a simple tracking plan for a merchandise campaign.
From an LLM Demo to a Production Workflow: What Reliability Requires
The demo read the invoice perfectly and then nobody could say whether the bill got created once or twice. Here's what a production AI workflow needs beyond a good first run, with a buyer acceptance test.
Test the Fix Without Reintroducing Yesterday's AI Workflow Bug
You fixed the rush charge on Monday. On Tuesday, a partial order that used to work stopped working. Here's how to keep a small test register that remembers what already worked, and release a change only when it passes.
API, Browser Agent, or No Fit? An ERP Automation Decision Guide
A decision checklist for choosing between an API, a supported browser route, or a manual step for one ERP job, with a worked draft sales order example and a six question approval list.
Why Small, Single Purpose AI Agents Are Easier to Run
One workflow that reads orders, checks them, changes records, and answers customers is easy to build and hard to fix. Here's how to split a process into a few clear jobs with written handoffs, and when splitting just adds noise.