Tender agent
An AI agent for a tender team that moves data from client spreadsheets with different layouts into one bid format. The Power Platform experiment showed that further changes were faster to make in code with agents than in low-code.
Every customer’s spreadsheet looks different
The tender team receives spreadsheets with different layouts. Their data needs to fit a shared format before the team can prepare a quote. The agent was meant to identify the structure, map fields and leave a result for a person to review.
My role
I worked on the workflow, integrations and whether the approach could handle additional formats. The spreadsheet translation rules were developed with the team.
When a shortcut adds work
We connected Power Apps, Power Automate, Office Scripts and Dataverse with AI. Some examples worked. New exceptions and changes meant more manual configuration across interfaces.
There was something ironic about it. Tools designed to save us from writing code were making the work take longer. I could already describe similar changes to an agent and implement them in code more quickly.
What I took away
This approach did not deliver what we needed. We ended development in Power Platform and chose to rethink the process. When considering low-code today, I also compare the cost of the next change with having an agent work in a repository.
Result
We stopped development in Power Platform and decided to redesign the solution.
An internal experiment. Client spreadsheets and data stay confidential, so this case has no screenshots.