We can talk ad infinitum about AI as a tool rather than an end-all-be-all but it bears repeating for as long as we see over-reliance lead to costly errors. This can manifest itself in numerous ways but there are a few glaring examples we’ve recently seen.

Some fall into the camp of trusting data that is just flat out wrong. Maybe the models are extrapolating based on patterns to fill in gaps rather than citing factual information. Or maybe the data it is pulling is wrong, but the tool cannot verify the accuracy and trusts it as reliable. Or, perhaps the information was true at some point but it’s no longer the case. At any rate, somewhere the AI tools are pulling information that is out of date or hallucinated, so the end results might seem compelling but are completely inaccurate.

Take for example a founder who tried measuring TAM using an AI generated market sizing query. The model was built on information that was entirely wrong, but there were no red flags to signal this, or at least none that were readily apparent to the founder. Thankfully we were able to dig deeper and see that there was no meaningful foundation to the model and we’d have to gather data another way.

There’s another crucial example we’ve seen, and interestingly in this case, the data itself is accurate. The issue with what happens to founders in this second example is that the data is just data, and they are missing the important human element that provides crucial nuance required for interpreting the data correctly when planning future strategy.

For instance, we had a founder who came to us with an AI generated competitive analysis that was technically accurate, and frankly did contain useful information. However, there were some qualitative points that the analysis simply missed, namely, how poor word-of-mouth of a competitor’s product was hurting its market share as opposed to design flaws. Without this crucial detail the information was not wrong, but it would have been a dangerous misstep to proceed based solely on the AI analysis.

Another founder used AI to synthesize customer conversations that his team had had with prospects. Well, at least in this case they were out talking to people, which we wholeheartedly and enthusiastically advocate for! However, as useful as these summaries were, they completely missed all the nuance of tone and other cues that can’t be easily captured. In this case we had those team members review the summary and they quickly noted the discrepancy and manually modified the summary to better reflect the crucial takeaways from the customer conversations.

The connection in all three of these cases is that the AI model produced very compelling results. As noted, there were no red flags that caused alarm, and that is precisely what makes it so concerning. A founder who doesn’t know the market will read the data and potentially rely on it as credible. After all, the AI presents the conclusion confidently whether that underlying data is accurate or not, so founders may feel inclined to trust it. This is especially true in cases where the founder is feeling pressure to grow and may not have the bandwidth to dig deeper. (And let’s be real here. What founder isn’t feeling that pressure?!)

Additionally, there is another element at work. AI is so ubiquitous now that relying on a readily-available tool feels like efficiency. Of course, this speed is really only helpful if the founder is moving down the right path. The bottom line as usual is that AI can produce research but it still needs to be validated by actual people, be that the team having conversations or those customers themselves.

Before dictating a strategy based on an AI generated analysis, consider the following:

Where did the data actually come from? Examine the underlying sources that the analysis cites and determine if the information is verifiable. If you can’t pinpoint where the information is coming from, or the validity of that data, tread lightly.

Did a real person from the team, founder or otherwise, have eyes on this before moving to the action phase? Remember data may be literally accurate but still miss crucial details that a human wouldn’t, so be sure to trust yourself and your team enough to parse the data points.

Have you actually tested the analysis? That is, did you bring the conclusions to actual customers before assuming its accuracy? Treat it like any hypothesis and validate before moving forward.

Are you weighing if an AI generated analysis has merit? If you need support in using data to plot your next steps, let’s connect.