To The Moon

Q&A（Question-and-Answer Session）is a session after the company's prepared remarks where institutional investors and analysts ask management questions. In this dialogue, you may find some valuable information that might affect the stock price in the following weeks.

Now let's look at some key points from  $C3.ai, Inc.(AI)$ Q4 FY2023 Q&A Session Transcript 

Q:Hi, thanks for taking the question. So Tom, you said that sales cycles were down to 3.7 months from five months last year. Why do you think that is? Is this entirely due to the consumption model? How much of this is due to general excitement around the potential in the space? And even potentially increase sales force productivity?

A:No, I think -- thanks Kingsley. I think it's all of that. I mean, clearly, AI is on everybody's mind, the consumption based pricing model that we have makes it much easier to adopt our technology. We made it in the old days, one and two years ago to do business with us was $5 million, $10 million, $20 million, $50 million to open the door. And now the transaction is pretty much, you know, we'll bring the application live in six months or $0.5 million. If you like it, keep it and pay $0.55 per CPU hours to be CPU hour, so we're pretty easy to do business with. And so we're seeing the number of transactions increased dramatically as we'd expect.

The ease of contracting with us. As you know, we have largely reconstituted the sales organization in the last 1.5 years to a sales team that is candidly much more productive and effective than our other sales organizations. So I think all of those are contributing to increased pipeline, increased business, increased business activity by which we're quite optimistic.

Q：Thanks, Tom. That's really helpful. And then one for Juho. When I think about the timing of the transition. So -- if it is the case that the vast majority of existing customers are not necessarily migrating to the consumption model, how should we think about the contribution of consumption over time and particularly in the back half? Because I think that you said revenue could accelerate as consumption increases in mix?

A:Yes, Kingsley, thanks for that question. So that's exactly as we sign and initiate more pilots within the quarter. The pilots are generally two quarters long, and then you start to see the consumption revenue kick in. As we finished the quarter with 19 pilots last quarter, we had a good increase in pilots with 17 pilots as well. You can start seeing those layer on to the revenue by Q3 and Q4 of this fiscal year.

Now to your point about renewals, we do expect our existing customers with the large enterprise agreements to continue to remain on those types of agreement structures, but you will see the RPO trickle down as these contracts enter into renewal phase, and then we would expect to see a pickup as they renew.

Q：Thank you. [Operator Instructions] Our next question comes from the line of Pat Walravens of JMP Securities. Your line is open.

A:Oh, great. Thank you. Tom, can you talk some more about the opportunity with National Security and the Department of Defense? And then also, you said something I thought was interesting about a version of generative AI that doesn't hallucinate, if you could maybe comment a little more on what hallucinating is? And how you prevent it from doing that, I think that would be really interesting? Thank you.

A:DoD, well, Pat, you asked kind of many times about the -- we have two basically authorities to operate contract vehicles once were $100 million and once were $1.5 billion in DoD that are associated. It could be applicable to what we're doing at RSO, that was the rapid sustainment office and the predictive maintenance application that we're doing for the United States Air Force for F-15, F-16, F-18, F-35, KC-135, et cetera. And what we made a proposal to the Secretary of the Air Force to take that into full production for all the aircraft in the Air Force, which is 5,000. I think the proposal would have increased aircraft availability for the Air Force by 25%. And I think decrease their cost of maintenance and readiness by about $6 billion.

So he considered that as did his Chief of Staff, General Brown, and they went off on their own for a few months, while you were asking the questions, and we didn't have the answers and these guys go into their start chamber the way they do. What they came out with was not -- was a selection of C3 as the standard -- as the system of record, not only for aircraft in the United States Air Force, but for all AI-based all predicted maintenance okay, in the United States Air Force for all assets. So this is genuinely a big deal, okay. Now we have the opportunity to make this a line item in the budget. So this is -- it's hard to over describe the impact of the overestimate the impact of this. And then not only do we have it in Air Force, we can talk now to other services like the Army and the Navy and the Marines and the National Guard, what have you. So this is a big one.

The second one has to do with generative AI. So one of the problems with generative AI is the -- is you're limited to the number of data sources that you can use with these large language models typically is text, HTML and sometimes code. And the large language model will interact directly with the data. But one of the problems is you get, kind of, random answers. Every time you ask the question, you get a different answer. If two people have an question and get a different answer.

And the -- there's no traceability. It doesn't tell you where the answer came from okay? And finally, if it doesn't know the answer, it makes one up. This is what they call hallucination. So it doesn't know it just kind of wings it makes up an answer. So we've leveraged -- we're using the entire C3 platform. And the way that we do that is we incorporate -- as you, I think, all know, we're very good at aggregating enterprise data, extra price data, code, images, text, sensor data, what have you, into a unified federated image. When we do that, those data are read by a deep learning model and they happen to be stored in a vector database that we have a kind of a firewall between that and the large language model. Now our customer uses any language model they want, be it ChatGPT, be it Home, be it hard, be it [Indiscernible] 5, whatever it may -- whatever comes along next.

Now, but we built a firewall within the large language model and the data. So it will -- every time -- I mean what's really -- every time you ask the question, it will give you the same answer. Okay, if two people ask the same question and they have the authority, they will both get the same answer every time. Associated with the answer, it provides you traceability to see if they click on it, you can see exactly where the data can come from, okay? And a very importantly, there's no risk of LLM cost data exfiltration, see Samsung for details where they find out that all of their proprietary information is not published on the Internet, okay?

And finally, there's no risk of LLM cost hallucination. It doesn't know the answer, it tells don't know the answer rather than making one up. So for these, you think would be kind of table stakes, and they are table stakes for any large commercial or government installation, and this is something that really distinguishes the C3 generative offering. And one of the reasons that we're seeing very high levels of interest.

The above Q&A are highlights that are edited for brevity. Click here for the full C3.ai Q4 FY2023 Earnings Call Transcript.​​​​​​​​​​​​​​​

If you want to know more details, you can click here to re-watch the C3.ai Q4 FY2023 Earnings Conference Call