Using AI to do Win-Loss Analysis
And why the Art of the Prompt Matters
At one company I worked for I wanted to understand why we had a churn problem so I did some win-loss analysis. This process took me a week to pull the data, organize it, and then analyze it. Analyzing it took the longest.
So today I tried testing this with AI to see if it would speed up the process.
I challenged both ChatGPT and Claude for answers.
At first, I found ChatGPT was inaccurate a few times and both were inaccurate in taking the data for face value, but Claude was able to get to the conclusions faster. Yet, I started to wonder, was it my prompt? Keep reading.
Here’s what I did:
My initial prompts
Prompt 1:
I uploaded a spreadsheet that looked like this (this is all fake data):
Granted, there were only 15 deals, but still…..there is enough information in here for me to do win-loss analysis on my own which would be:
We tend to lose to competitor B - find out why and dig deeper to see if we can create any objection handling
We seem to win in SaaS, although, no specific industry rises to the top. Explore more deeply into the closed lost reasons as to why we lose in some SaaS and why we win in others.
We seem to be able to go head to head with Competitor A so maybe look into messaging there to see what we can improve.
MM and Enterprise are where we’ve won the most recently. Dive deeper into closed lost reasons to see why we lost other MM/ENT deals.
Webinars seem to work. See which webinars worked to close these deals vs the one where we lost. Anything to glean there?
Look at historical data and decide if you need to possibly pull out Snowflake as an outlier.
It came back with:
So then I ask it how can we outperform competitor B and lose more frequently to competitor B?
Prompt 2:
It came back with this:
I am glad it admitted it was wrong, but I’m still not happy with this. Why would it suggest with only five deals that FinTech, Cloud Storage, and Security were good verticals?
Prompt 3:
Then it came back with this:
So it finally came back with our strongest performance was MM and ENT (which is correct). And it came to the conclusion that we do not consistently win in any specific industry.
So then I asked Claude… and it got to the conclusion a lot more quicker than ChatGPT did.
But I still didn’t like the fact it recommended to broadly target SaaS. So I asked it:
Why would you recommend targeting SaaS if there were a good bit of SaaS in lost deals?
This is a much better nuanced answer. To conduct a deeper analysis of why we’re losing in specific SaaS verticals, I would have to upload or look at Loss Reasons.
However, the one thing Claude didn’t address without me prompting it, even though it had all of the information in the spreadsheet, was whether or not we won or lost more against any particular competitor.
All of this analysis took me less than 2 hours rather than 5 days, but I still felt it needed some human oversight.
Could my prompts be better? Yes.
This IS the prompt you should use for Win/Loss Analysis
I came back the next day to ChatGPT and asked it the following:
Based on this quarter’s wins and losses, please summarize and identify trends where we win and where we lose. Please advise if there are any commonalities among where we win by company type, industry, acquisition channel, deal type, deal size, or days to close.
It provided the following and this is WAY better and more accurate:
It is about the art of the prompt. I also told it to clear its memory and to start fresh. This prompt took me 5 minutes and the recommendations are spot on.