Gathering lessons learned is straightforward. Synthesizing fifty survey responses into a useful readout is where the time goes. This is the agent that does that work.
Lessons learned sessions have two distinct problems.
The first is getting people to participate. A well-designed survey sent to stakeholders at project close can solve that—it’s low friction, asynchronous, and gives people time to think before they respond. MS Forms makes this easy to set up and easy to collect.
The second problem is what you do with the responses once you have them. Fifty survey entries, each answering five to eight questions, all in slightly different language, at different levels of specificity, from different perspectives on the project. Reading through all of it, identifying themes, pulling out the most representative quotes, and organizing it into something a team can actually discuss in a 60-minute readout meeting—that’s a significant block of time. And it has to happen right at the moment when you’ve just closed a project and are already pivoting to the next one.
That’s the problem this agent solves.
The Approach
The workflow starts before the agent is involved. At project close, I send stakeholders an MS Forms survey with questions organized around the standard project closeout categories: what went well, what didn’t, what we’d do differently, what we should carry forward to future projects.
Once responses are in, I download the Excel file from Forms. I also pull together the key project documents: the project charter, runbook, issues log, risk register, and any other artifacts that help tell the complete story of the project. I’m moving toward putting all of these in a single SharePoint folder and giving the agent the folder link rather than sharing each file individually—that’s cleaner and makes the workflow repeatable without having to track down and attach every file each time. The next evolution I’m exploring is setting up a Copilot Page at the start of each project that links to those living documents as they’re updated throughout. At closeout, the agent gets pointed at the page instead of a folder—one place that always reflects the current state of the project.
The agent reads all of it—the survey responses and the project documentation together—and produces a readout document. For each survey question, it identifies the top five themes from the responses and pulls representative quotes to support each one. The project documents give it context: when the agent can see that a risk in the risk register materialized, or that an issue sat open for three weeks, it can connect the survey responses to the actual project record rather than just summarizing what people said in isolation.
What the Output Looks Like
The readout document is organized by question. Under each question, the agent surfaces the top five items that emerged from the responses—the themes that came up most often or carried the most weight—and includes direct quotes from the survey to illustrate each one.
That combination of theme and quote is what makes a readout actually useful. The theme tells you what people collectively experienced. The quote gives it texture—it’s one team member’s specific language about a specific moment in the project, and it tends to land differently in a readout meeting than a summarized finding does. People recognize the experience. The conversation gets more concrete.
A Note on Tooling
I’m still refining the Copilot version of this agent, but the core approach works. The agent can synthesize the survey and project documents and return the content you need for a readout. Formatting a polished Word document from that output is something Copilot handles less seamlessly than tools like Claude or Cursor—with those, you can get a properly formatted document directly. With Copilot, today’s workflow is: take the synthesized output and paste it into Word to handle the formatting yourself. I expect that to improve as I refine the approach.
For PMs who have access to Claude or Cursor, the end-to-end process can produce a formatted readout document ready to share. For everyone else, having the synthesis done—themes identified, quotes pulled, structure in place—is still a significant time save. The content is the hard part. Pasting and formatting is not.
This matters because not every PM on a team will have access to the same tools. Building the workflow on Copilot means anyone in a Microsoft 365 environment can use it, even if the output requires a little more manual effort to finish.
Why This Pair of Agents Works Together
This agent and the lessons learned retrieval agent are two ends of the same problem.
The retrieval agent asks: what did we learn in the past that applies to this new project? It helps at kickoff, when you need to pull institutional knowledge forward.
This agent asks: what did we learn from this project that should go into that institutional knowledge? It helps at closeout, when you need to capture what happened before the team scatters and the details start to fade.
Together, they close the loop. The readout agent creates the documentation. The retrieval agent makes that documentation findable and useful for the next PM who starts a similar project.
The Broader Point
Lessons learned processes often fail at the synthesis step—not because organizations don’t care about capturing knowledge, but because the effort of going from raw responses to a structured readout falls entirely on the PM, who is already stretched at project close.
An agent that does the synthesis doesn’t replace the judgment involved in running a good closeout conversation. You still need to read the output, validate the themes, and decide what the team actually needs to discuss. But it removes the manual aggregation work—the reading, sorting, and organizing that takes hours and produces something that could have been produced in minutes with the right tool.
Better readouts get written down. Better writeups become the source material for the retrieval agent that helps the next project avoid the same mistakes.
Links & References
Related Monday Business Posts:
- AI-Powered Lessons Learned Agent — the companion agent: using past lessons to inform a new project
- AI + PMO: Manual Work Worth Eliminating — other high-value automation opportunities in a PMO
- Better Status Reports with AI — applying AI to project documentation more broadly
The content is the hard part. The agent handles that. The formatting you can do yourself.
Coming soon: I’m working on a lessons learned orchestrator — a single agent that figures out what you need and routes you to the right tool automatically. Whether you’re starting a project or closing one, you describe the situation and it handles the rest.