A week of fund reconciliation across 200 funds now takes two days.

Participants in recent AI trainings across campus have shared a similar message: people are ready to move from AI tool discovery to application. They are less interested in what a tool can theoretically do and want to experience where it actually saves time. Carol Ray and her coworkers are beginning to realize the difference fund by fund, manual page by manual page.

In a recent business training, other departments shared copies of their desk manual. Based on her personal experience with AI tools, Carol began to think that AI might be helpful in the process of updating her department's manual. Carol gave ChatGPT the specific MSU policy pages and the department's current structure and let it help assemble a draft. Before AI, she says, a rebuild like this would have taken two to three months to complete while working it in between other duties.

The process still was not instant, and Carol is careful to make this point. The work of gathering the right sources and vetting the information took time, as did planning the process. "You need a full understanding of the project you're building," she says. "The design stage is important. You need to decide first what you'd like to produce." The timesaving was realized in the document drafting and formatting, and that was a recognizable difference to the team.

Carol's department is also beginning to implement AI practices in their account management work. They manage more than 200 funds, and the monthly work of reconciling them against Banner has required a lot of time. In the new process, Claude reads the department's Banner PDF ledgers, follows the monthly tab layout her team provides for each fund, and imports the results into an Excel sheet. Her team cuts and pastes that output into their own department spreadsheet which allows them to check Claude's work.

The check is nonnegotiable. Carol's team learned early on that Claude will sometimes try to be helpful in the wrong way. It has tried to combine charges that look similar, for instance. To counter Claude's overinterpretation, they have given it a standing instruction: do not interpret and do not deviate. When they ask Claude to validate a spreadsheet, it now asks clarifying questions back, using the department-specific rules Carol's team has taught it. A human makes the final decision on what is correct. Work that used to take a full week of data entry across all 200-plus funds now runs closer to two days, and Carol expects that to improve as the tool finishes learning her department's fund structure.

Additionally, the team is beginning to implement AI assistance with proposals. They can feed a funding call to AI and ask it to pull out what is actually required, directly from the call language. This saves time that used to be spent hunting through the document by hand.

Carol is clear about boundaries in AI usage. "AI is an amazing tool, but we have to remember it is a tool," she says. "It doesn't replace the human aspect of your job. It's up to the department to see that everything is verified."

What's changed most for Carol is how her days feel. "It has taken away the pressure that I once felt," she says. "I didn't always have enough time in the day to get everything done at the level of detail needed, but AI is giving me the opportunity to get to the tasks that I didn't get to like I wanted in the past." That's the part of implementing AI processes that is easy to miss: less time spent on data entry means more time for the work that serves the department, the college, and the people who depend on it.

Carol is looking forward to sharing what she and her department have learned with other business staff across campus in the same way that she learned from others.