Latest / Tech IPO Conversations with Fexingo: Public Software Companies and Capital Markets / How Post-IPO Companies Use AI to Predict Revenue
Transcript
- Lucas: So last week Shopify reported earnings that beat consensus by almost four percent, and the stock jumped eight percent over the next five days. But what caught my eye wasn't the beat itself — it was how they got there. Luna: How do you mean? Lucas: Shopify is part of a quiet trend among post ipo companies: they're using AI to forecast their own revenue. Not just for analysts — internally, to set guidance. And the accuracy gains are startling. Luna: I've heard whispers about this. Some companies are replacing the old spreadsheet and judgment approach with machine-learning models that crunch sales pipeline data, customer churn signals, macro indicators. Lucas: Exactly. One cloud software company I've been following — let's call it Company X, it's a recent IPO — built a custom neural net that ingests over two hundred variables every day. Their quarterly revenue forecast error went from about six percent down to under two percent. Luna: That's a massive shift. A two percent error means they can guide with real confidence. Analysts love that. Lucas: Right. And when you think about the post ipo landscape, where every earnings miss gets punished brutally — just look at some of the names in the news lately — the incentive is huge. Nvidia dropped seven-point-seven percent in the last five days on macro fears, not even an earnings miss. Imagine if a company can reduce the guidance gap. Luna: But what about the SEC? They've been wary of ai generated projections. There was that proposal last year about requiring companies to disclose if AI is used in earnings guidance. Lucas: Right, the SEC hasn't issued final rules yet, but the informal stance seems to be: you can use the models, but you have to be able to explain the logic to auditors. No black boxes. So companies are building interpretable models — feature importance scores, SHAP values, that sort of thing. Luna: So it's not about replacing the CFO's judgment. It's augmenting it. Lucas: Exactly. The CFO still sets the final range. But the model gives them a tighter starting point. And during earnings calls, when the analyst asks about the back half of the year, the CFO can say with more conviction where revenue is heading. Luna: I want to go back to Company X. How did they validate the model before going live? Because if you train on historical data, the patterns might change after you go public. Lucas: Great question. They ran a shadow period for six quarters — the model produced forecasts alongside the traditional process, but they didn't use them publicly. They compared the model's predictions to actual results and found it outperformed manual forecasts in five out of six quarters. Only then did they start incorporating it into guidance. Luna: That's rigorous. And it makes sense because post ipo companies often have limited public earnings history. You don't want the model learning from a data set that's too small. Lucas: Right. And they also added a feature that tracks macro indicators in real time — like the ten-year yield, consumer confidence data. So if the macro environment shifts, the model adjusts. Luna: So what's the risk here? If every post ipo company adopts this, guidance becomes a kind of consensus model. Everyone's forecast converges. Lucas: That's a real concern. If all companies use similar AI tools trained on similar data, you could get herding in guidance. And when the macro environment changes rapidly — like we saw with inflation spikes in 2022 — the models might all miss in the same direction. Luna: So it's not a panacea. But for now, the early adopters seem to be getting an edge. Shopify's eight percent bounce suggests the market rewarded their confidence. Lucas: Yeah. And it's not just Shopify. I've seen at least a dozen post ipo software companies experimenting with similar models. Some are building in-house, others are buying off-the-shelf tools from vendors like Anaplan or Adaptive Insights. Luna: Adaptive is Workday now, technically. But I take your point. Lucas: Right. The key takeaway is that ai driven forecasting is moving from a back-office experiment to a core part of investor relations. And for listeners who follow IPOs, this is a signal: when a company announces a new AI forecasting system, pay attention to the next few quarters' guidance accuracy. Luna: Let's talk about the cost. Building a custom neural net isn't cheap. For a post ipo company with maybe fifty million in revenue, is it worth the investment? Lucas: The company I mentioned spent about two million dollars over eighteen months to build and validate the model. That's not trivial. But if it saves them even one bad quarter of stock price damage — say a ten percent drop instead of a fifteen percent drop — it pays for itself. Luna: Especially for companies with volatile stocks. I look at some of the recent IPOs in the ARK Innovation ETF — ARKG is up almost fourteen percent in the last five days, but ARKK is flat. The swings are brutal. Lucas: Exactly. And that's where the model's ability to tighten the guidance range really helps. Investors hate uncertainty. If you can show that your revenue will be within a narrow band, they assign a higher multiple. Luna: So do you think this becomes table stakes? In two years, every post ipo company will have some form of AI forecasting? Lucas: I think it's heading that way. But not all will do it well. The ones that succeed will be the ones that don't just buy a tool — they integrate it with their sales data, their customer success data, and they keep the human in the loop. Luna: Speaking of tools, one thing that keeps this kind of deep-dive reporting going is listener support. If you find these conversations useful for what you're building or running, a couple of dollars a month genuinely makes a difference. You can head to buy me a coffee dot com slash fexingo. Lucas: Yeah, it keeps us ad-free and focused on the details that matter. And we really appreciate it. Luna: Anyway, back to forecasting. One more thing I want to dig into: how do these models handle seasonality? Cloud companies often have December-heavy quarters. Lucas: Good point. The best models include seasonal features — like month-of-year dummies, holiday calendars. They also weight recent data more heavily. So if a company had a blowout Q4 last year, the model doesn't assume it will repeat exactly. It adjusts for growth rates. Luna: That makes sense. And it's probably better than the old method of just taking last year's number and adding a growth percentage. Lucas: Much better. So for anyone tracking post ipo companies, I'd suggest looking at their earnings call transcripts for mentions of AI or machine learning in the context of forecasting. It's a leading indicator of how seriously they take guidance precision. Luna: And if you see a company that starts consistently beating by small margins — maybe that's the model at work. Lucas: Exactly. It's not magic. It's just better math. And in a market where every basis point counts, better math can give you a real edge.