Operations managers were spending hours assembling reports instead of running operations
Leadership came with two requests: set clear rules for AI, and show where it could make the organization more efficient. The managers who run field operations were consumed by manual reporting. Each one kept a monthly dashboard for leadership that drew on lab sheets, monthly operating reports, the maintenance system, overtime logs, inspections and field readings. One manager pulled from 17 sources and could spend five hours copying, reformatting and checking figures before any review began. Every manager built theirs differently, so leadership read each one differently, and little time was left to plan ahead.
The organization also answers to regulators, and no rules were in place for AI. No one had an organizational license, and team members used a mix of personal AI accounts without any agreement on which information belonged in them.
The rules were settled before training, then built into Claude
Before any training, leadership agreed on what AI is for at the organization: less tedium, faster clarity from the data it already has, and more time to think. It adopted a red, yellow and green classification that keeps customer personal information, payment data and NDA-protected project details out of AI tools and clears routine business data for use. The policy also requires a person to review any output before it leaves the building and numbers to be verified before they are trusted, two rules the finance team had asked for from the start. We ran Claude, ChatGPT and Microsoft Copilot side by side on the same month-end close and work order backlog, and leadership chose Claude on the output. An internal AI Leader owns the account and an intake process for new uses.
The AI Leader then loaded the classification into Claude’s company-wide instructions, so Claude pauses and asks before working with anything the policy marks as restricted. With the rules settled, the managers could put their real operating data to work without guessing where the line was.
The managers rebuilt their reporting around the decisions they make
In working sessions with the operations team, each manager answered three questions for every report: what is the metric, what is the threshold, and what do you do when it crosses? A number that does not lead to a decision came off the front page. Each manager then built a dashboard once, in a shared folder, from their own raw reports and a set of saved instructions. Every month after that, they drop in the new files and Claude produces the dashboard.
The hours of manual data work are gone, and every dashboard is built the same way
The operations managers no longer spend hours each month copying, reformatting and reconciling figures across as many as 17 sources. One manager linked Claude directly to the folder where operators upload their monthly reports, so the data is ready as soon as it is saved. Because every manager uses the same method, the whole operations team produces dashboards in the same format, leadership reads them the same way, and a colleague can run one when its owner is out.
The same approach now gives the team a one-page daily operations brief from the control system’s morning summary. Within its first weeks, it flagged two issues that were fixed the same day.