Using ChatGPT to build pricing models or budgets

A friend of mine was curious about whether ChatGPT could do budgets. By the time she asked me, I had done some minimal experimentation with ChatGPT and Excel, but I hadn’t done budgeting or forecasting. I was kind of wondering how I would even approach ChatGPT.

So, we embarked on an experiment. She sent me her version that she and her friend were working on, and I ran several trials, at least 3 trials. I won’t display all of the trials here, but it can get tedious to sift through what it did.

However, the bottom line is ChatGPT is not ready for primetime in the area of budgeting/pricing models. For starters, it tends to have problems with basic math, although OpenAI is working on that. I don’t doubt that basic math will be resolved. The second thing is that the AI has problems with some of the labor concepts (for lack of a better word). In the labor there were three concepts, one of which it would miss: 1) how long somebody worked on the project – how many months; 2) percentage of their time spent on the project – 10%, 20%, etc; 3) application of benefits.

ChatGPT will get two of the 3 concepts.

If OpenAI can fix those weaknesses, budgeting will become a thing of the past because it is FAST. Like 5 minutes fast.

First Trial

I just took her budget instructions and fed it to ChatGPT just to start the ball rolling.

My calculations came to $442K and ChatGPT gave me $810K.

When comparing my calculations with ChatGPT’s, I found a few minor calculations errors, but the big standout was in the labor. It butchered the benefits totally; it might have left it off.

I went away to think how to re-write the instructions to see if I could get it closer.

Second Trial

Both my friend and I thought that the instructions for benefits should be written closer to the general labor instructions. Her original instructions to me had separated the benefits to the bottom of the instructions; it didn’t tie in with the labor. We decided I would move it to where the labor instructions were.

Same numbers but rearrange the prompt.

Again, my calculations were $442K. In this second effort, it came up to $1.1 million. In digging into its explanations, I found that it was setting the labor at 100% on the project when all were less than 50% on the project.

Once I pointed that out, ChatGPT re-did the calculations and came up with $432K.

Nice! There were still some basic math calculations errors, but we were much closer.

Third Trial

I came back another night with the intention of seeing whether ChatGPT had learned from the second trial. I came up with a similar set of instructions but with a different set of numbers. The benefits instructions came right after the labor instructions.

In this new budget, my numbers came out to be $849K; ChatGPT gave me $376K.

Then I tried guiding it by pointing out what it missed – such as the project was for 20 months but it calculated for only 1 year or the benefits were calculated at 100% rather than the part time percentage. I made several attempts to get it to reason properly, but no go, it never did come close.

Conclusions…for now

Everybody doing budgeting can breathe a sigh of relief.

But I would suggest playing around with ChatGPT and stay in touch with its progress. You want to be ready to have a plan on when it can successfully generate the bulk of your budget.

Because the day is coming, maybe much sooner than we expect. I’ve been “talking” for the last decade about preparing for the day when a lot of the work (everything?) becomes automated. We need to know what it is we can do that the AI can’t.