Turning a Broad AI Goal Into Focused Workflows

The firm came in with a broad goal: use AI to make the business more efficient. There were potential use cases across different parts of the business, but no clear answer on where AI would create enough value to justify the change.

The question was not simply where AI could be used. It was where the firm should start, what was worth changing, and what should remain human.

I interviewed the owner and team and walked through how work was being done day to day, where time was being lost, what was repetitive or frustrating, and where mistakes or missing information could create bigger problems.

What stood out

Recurring administrative work was taking substantial time away from design and client work. Project information was especially important because client instructions, requirements, deadlines and decisions sometimes needed to be traced back months or years later. If an important instruction was not recorded clearly, that could become a serious issue in a later dispute over who directed what.

What I chose not to automate first

I did not prioritize design production. When the project began, available AI tools were not reliable enough to handle drawing or modelling work to the standard the practice required. More importantly, design and professional judgment were the expertise clients were hiring the firm for. The point was to free up more time for that expertise, not automate it away.

Much of the information the business relied on was scattered across emails, PDFs, invoices and receipt images. A lot of the logic around what mattered, what needed follow-up and what required attention also lived in people’s heads.

Before, the owner was manually reviewing project information and transferring key details between emails, project records and calendars. I worked with the team to decide what actually needed to be captured, how it should be organized, and when something should come back to a person for review.

I created shared working records in Google Workspace for project information, invoices and bookkeeping, then built the automation around those records.

The biggest constraint was not a lack of AI tools. It was fragmented information and working knowledge that had never been made explicit. Once those were turned into shared records and clearer review points, AI could take on repetitive admin work without taking professional judgment out of the process.