Calibrating Copilot Ep. 2: Can Copilot Make the Bid or No-Bid Call on a Complex RFP?
In the first episode of Calibrating Copilot, Dino Caputo and I tested a relatively simple task: summarizing a single project document. For Episode 2, we raised the difficulty considerably.
Dino found a real-world request for proposal (RFP) for a Cisco-to-Microsoft Teams Phone migration. It consisted of nine separate documents: Word files, Excel spreadsheets, location data, telephone information, and question-and-answer workbooks. The information was spread all over the place, the kind of package that would take a person at least a couple of days to fully digest.
The test
Our prompt asked Copilot to act as an experienced bid capture analyst, analyze every document (we placed them all in a SharePoint folder), summarize the opportunity in no more than 4,000 words, and, most importantly, end with a clear recommendation: bid, bid with conditions, or no bid. This is a commercial decision where hundreds of thousands of dollars could be on the line.
As always, we ran the same prompt through Copilot Chat, Microsoft 365 Copilot, the Researcher agent, and Cowork, scored the results independently, and then averaged our ratings for accuracy, completeness, speed, and cost.
Cowork was in a league of its own
The results were not close. Cowork was the only Copilot that identified the significant risks you would absolutely need clarified before responding. Among them:
A requirement to place source code in escrow, which makes no sense for a cloud-based Teams Phone deployment (Microsoft is not going to hand over its code)
Unbounded liability
Insurance requirements that, as Cowork noted, no national carrier (and the bidder also had to be an Operator Connect provider) would ever agree to
Confusion over whether the published Q&A answers were even contractually binding, since the contract said only the main document mattered
Cowork told us the bid should only proceed if specific items were confirmed before the August 31 deadline, and it supplied the exact clarification questions to ask. It even flagged the highest commercial risks right in its chat response, before we opened the Word document.
It was the slowest option at 11 minutes and 11 seconds, and it consumed 588 AI credits, just under $6. We took half a star off for accuracy because it confidently reported that its assessment was about 3,000 words, well under the 4,000-word limit, when it was actually 16 pages and roughly 5,500 words. In both of our runs it went over. In this case, it did not really matter.
What the other Copilots missed
Interestingly, every Copilot recommended bidding with conditions, but only Cowork clearly explained what those conditions were. Researcher and Microsoft 365 Copilot partially identified the risks; in Dino's run, M365 Copilot edged out Researcher. Copilot Chat did a good job summarizing the project, but it gave you almost no help with the actual commercial decision.
Quirks worth knowing about
Copilot Chat cannot access SharePoint links, and its error message (the folder requires authentication or is not accessible) is confusing. I spent ten minutes convinced it was a sign-in issue.
Copilot Chat then recommended zipping the files and uploading them. The problem: neither Copilot Chat nor Microsoft 365 Copilot can process ZIP files. Cowork can.
The workaround is to upload the files in batches of three. Tell Copilot up front that you will upload all nine files in batches and that it should wait for all of them before processing. It followed that instruction perfectly.
Dino's M365 Copilot run got stuck on 'working on creating the Word document.' After about two minutes, quitting and reopening the app revealed that the document was there. If it is taking an unusually long time, it probably is not doing anything.
The verdict
For complex RFP analysis, Cowork is absolutely worth the AI credits. Spending six dollars to inform a bid worth hundreds of thousands of dollars is an easy decision, and a person doing this work would take days and might still miss the subtleties spread across nine documents. That is a very different conclusion from Episode 1, where Cowork's extra cost was harder to justify.
Dino offered a smart idea for making the analysis even safer: run it more than once, and consider a separate prompt focused specifically on legal risks. That helps smooth out the run-to-run differences we saw between our results.
Matching the right Copilot to the right work
The lesson from this episode is that value depends entirely on the task. For a quick summary, the free or included Copilot may be all you need. For high-stakes analysis across many documents, a more capable agent is a completely different value proposition. The same logic applies to the people in your organization: your bid managers and analysts may justify Cowork, while others are well served by Copilot Chat.
That is exactly what our Copilot Calibration service is built to solve. Using your existing Microsoft 365 usage data, we deliver person-by-person recommendations: who should get a Microsoft 365 Copilot license, which licenses should be reassigned, where Cowork makes sense, and where targeted training would pay off. It is fixed-price and requires minimal effort from your team.
