Tom Siebel: Hi Mike, it’s Tom. Doubling down, we’re doubling down on data scientists, we’re doubling down on large language model engineers, we’re doubling down — a lot of it is going into engineering, but also candidly in lead generation. I mean, there’s an opportunity now as we move to these marketplaces to be dealing transactions in hundreds to thousands to tens of thousands of units rather than scores. And that I can assure you is the plan that we have. As it relates to — I’m not familiar with Bloomberg article that you talked about. It sounds like somebody mentioned something that we did some layoffs in the quarter. Mike, we do performance-related layoffs every quarter, okay? And the — so we — I think last quarter, we had 42,000 job applicants.
We — how many people did we hire, Juho? Order of 100. And these people, yes, they went to MAT. Yes, they worked at Bank of America. Yes, they went to Chicago TSB and they command an F1-8 squadron. And so, we’re constantly upgrading our human capital, and we move underperformers out regularly. So if somebody said that in a Bloomberg article, I don’t know what they said. What I told you is the truth.
Operator: Our next question comes from the line of Kingsley Crane of Canaccord Genuity.
Kingsley Crane: I wanted to touch on the pilot program. You mentioned that you’d move to a lower entry price point for pilots. Could you give us a sense of the magnitude of that change? And then has the minimum fee post pilot also changed? I’m curious what kind of upsell you’re seeing upon conversion, if any?
Tom Siebel: I think the standard pilot that we have at generative AI and the enterprise is like $250,000. But that being said, you can get the AWS — generative AI for AWS, which basically handles documents like every other LLM, handles text, it’s not really multimodal, but that’s free for 14 days. So, that would be pretty available. Is there a question that you asked that I didn’t answer?
Kingsley Crane: Okay. Yes. Thank you. That’s helpful. And I just want to touch on OpEx as well. So, I think it makes sense that you want to invest more in both, LLM engineers and lead gen. And it looks like that’s particularly hitting harder in Q4 of this year. But as we think about fiscal ‘25, it seems like some of the nature of those investments would naturally continue as you scale in the some large opportunities. So, is it about timing in this year, or are you expecting those to continue next year?
Tom Siebel: Kingsley, I expect them to continue next year. But if you look at the guidance that we gave you in terms — about six quarters ago, what we see is the consumption over the first 12 quarters in terms of CPU seconds per new customer. We just did an analysis of, Juho, I don’t know, about 30 customers — or 12 customers. And those data that we predicted, I think 6 or 7 quarters ago and provided you, it’s uncanny in how accurate it is. It’s basically plus or minus 10%. And so if you look, as these things kick in, in quarter 5, 6, 7 and 8, the consumption numbers get pretty big. So you can expect that — we don’t really need to cut back on the investments to get to the point of cash positive and non-GAAP profitable. So, the top line kind of takes care of that.
Operator: Our next question comes from the line of Sanjit Singh of Morgan Stanley.
Unidentified Analyst: Great. This is Steve on for Sanjit. Tom, maybe starting with you. I mean, with a couple of quarters of the consumption model now under your belt, clearly, you’re seeing a lot of sort of quantity of deals and pilots. Is there any way that you can frame or give us a sense of the quality of those customers that went with the consumption model early on? I guess, any sort of scale in terms of spending or growth profile that they’re hitting now that you can kind of shed up some light and give us the quality piece where you’ve given us, I think, a lot on kind of the quantity piece of those yields? And then for Juho maybe, could you just give us some color on the subscription revenue versus the services revenue this quarter, and then also maybe the partner impact and sort of what that looks like on a go-forward basis? Thank you.
Tom Siebel: Regarding quality, I think there’s only two ways to look at pilot quality. It’s going to be what’s the conversion rate and what’s — and what are they going to consume. Based upon our best guess at this time, based upon looking at every pilot we have out there, going to look at what actually has converted and what we think we will convert, we think our guesstimate that we gave you 6 or 7 quarters ago, 70% is about right. So, there’s one indication of quality. The other indication of quality is how many CPU seconds are they consuming over — as you go from quarter 0 to quarter 12. And it’s tracking right in line. I mean, it varies a little bit from one quarter to another, but it’s basically right in line with what we told you.
The quality is pretty high. Now that being said, as we move now to mass markets and start dealing with hundreds of thousands of people just either kind of ordering this online and playing with it, you can expect that conversion rate from that level of pilot to be, I would say — I mean, the quality there will be much lower. And I think we need to measure quality by conversion rate and consumption levels. A lot of those people will try it for 5 minutes and drop off. And that’s just the way that it is with free stuff. Now the rest of the question, I think, goes to you.
Juho Parkkinen: Yes, right. So your second part about subscription versus services. So we were 9.3% professional services this period, which is a little bit lighter than our expected long-term model of 10% to 20% on professional services. We continue to expect that we will be at that range on a go-forward basis. And then, I think you were asking about how we feel about the partners in a go-forward basis. And partners are hugely important for us. And we continue to believe that they’re the key part of our go-to-market approach going forward.
Operator: It looks like we have time for one last question. Our last question will be from Pat Walravens of JMP Securities.
Owen Hobbs: This is Owen Hobbs, on for Pat. I guess first one for Tom. What would you say are the top one or two federal use cases for generative AI that you’re seeing with those new — those five new federal degenerative AI deals this quarter?
Tom Siebel: Our largest federal use case, as you know, is predictive maintenance of the United States Air Force. This was chosen by the Chief of Staff. And we now are doing — this is the PANDA system, which is the only AI system of record that we’re aware of in all of DoD. So, this is a system record for the Air Force for predictive maintenance for all assets. So far, we have loaded the data, I believe, from 22 weapon systems. F-15, F-16, F-18, F-35, KC-135, F-22, et cetera, into unified federated image. This is 100 terabytes of data. Some of it is maintenance data, sorted data, inventory data, flight data, flight history, telemetry and one aircraft like B-50 — each B-1 bomber has 42,000 sensors on it, admitting telemetry, and I’m not sure what hurt cycles but pretty fast.
