AI Gave Me the Forecast I Wanted. Do I Trust It?

I want the forecast to be right. If it is, it gives me confidence to keep investing and growing, which is what I already want to do. 

The truth is, we have never been very good at forecasting at MultiFunding. We don’t have a CFO yet, and while we have tried to forecast here and there, most of it has been educated guessing. We would look at our pipeline, think about what was likely to close and when, and try to figure out what the next few months might look like. It has never been sophisticated enough for me to feel comfortable making major decisions based on it.

AI is changing that. I regularly feed it an updated pipeline report, our latest P&L, and our updated loan closings. Then I keep asking questions, challenging the assumptions, and trying to get it to look at the information in different ways. What we are getting back is far more detailed than anything we have had before.

It is also picking up things we never really accounted for in the past. For example, our loan advisors don’t all treat their pipelines the same way. Some put opportunities into the pipeline very early, which means their early-stage deals need to be discounted. Others pull a shaky deal out faster when they realize it isn’t going anywhere. The AI is starting to recognize those differences and adjust for them.

And the forecast looks really good, which is where I start to get nervous. If I trusted it, I would feel safer investing in the business faster. I could hire sooner, spend more on technology and infrastructure, and make those decisions with more confidence because the forecast shows what our profit and cash flow could look like over the next six months.

The problem is that the model is only eight weeks old. That may feel like a long time in the AI world, but it isn’t a long time in the life of a business. It is especially short at MultiFunding, where deals can take months to close. I don’t think it’s enough time for AI to understand what happens when several large deals get delayed at the same time, the economy changes, or something else happens that isn’t represented in the information we have given it.

We also can’t compare it to an old forecast because, frankly, we’ve never had one this detailed. I am not looking at our traditional forecast and deciding whether the AI forecast is better. I am going from having little to no forecasting process to suddenly having something that looks quite sophisticated. That said, I fear the sophisticated look makes it even more tempting to put too much faith in the forecast.

But that’s just part of why I find it so credible. The assumptions seem reasonable, the numbers are detailed, and the results make sense to me. But none of that means the forecast has to be right. It could also mean we have gotten very good at building a forecast that tells me what I want to hear.

This is where AI can get dangerous for an entrepreneur. It doesn’t necessarily have to give us bad information to lead us astray. It just has to give us a convincing case for something we already want to believe.  We have always been able to play with a forecast. We can change conversion rates, move closing dates, and tell ourselves that next quarter will be different. AI lets us do that much faster and with far more detail, which can make the answer feel more plausible than it really is.

It can also only work with what we give it. If I spend more time focusing on the strength of our pipeline and not enough time on how often deals get delayed, that will affect what comes back. I can keep changing the assumptions and tell myself I am making the forecast better, when what I may really be doing is encouraging it to support the growth I want.

None of this means we should ignore the forecast. It is one of the best tools we have ever built at MultiFunding. But before I use it to decide how quickly we should invest, I need to see how well its predictions match what actually happens. For the next several months, tracking its performance may be the most valuable part of this exercise. Where was it right? Where was it wrong? Did deals close when it thought they would? Did the revenue and profit show up when it expected them to? And when it misses, was there something we should have known that wasn’t reflected in the model?

We should also use AI to argue with itself. We can ask what could make the forecast wrong, which assumptions matter most, and where it has the least information. More importantly, I need to decide ahead of time what would make me slow down. I haven’t figured out exactly where that line is yet. But I know that if I wait until after the forecast misses, it will be easier to tell myself that the miss doesn’t really count and that it will be right next month.

AI doesn’t replace judgment. I think in some ways it makes judgment even more important. For years, we were mostly guessing about the future. Now I have something that can give me a detailed six-month forecast and make a pretty convincing case for why it is right. 

If we keep playing with it until it shows us the future we want to see, we will just be permission shopping faster. But if we keep improving the forecast based on what actually happens, we may be building something incredibly valuable.