Why Enterprise AI Isn't Delivering ROI | Ahmed Bashir, DevRev

Craig S. Smith (0:00)

And the problem, and I've heard this from a lot of people, is that Enterprise AI demos work amazingly, but once you get them in the Enterprise and the complexity of the Enterprise, everyone's disappointed.

Ahmed Bashir (0:12)

While everybody claimed to have industry-leading AI., CIOs were betting on AI heavily, and very few were actually unlocking real R1. There was a gap in the level of governance, the level of orchestration, the level of semantic understanding at scale. And that's exactly what shared memory provides. I think that the major observation is that a more sophisticated model cannot always make up for not having a sophisticated memory layer.

Craig S. Smith (0:37)

Swapping in a smarter model didn't win you that much, that what mattered was the plumbing of the system. Is that right? And if so, did that surprise you?

Ahmed Bashir (0:47)

If you think about computers with CPU and RAM, just getting a computer with a really, really high-performing CPU cannot counter the ill effects of not having enough RAM.

Craig S. Smith (0:57)

We talked on our first call about redigitization, and you had an analogy or an example of how music digitized for CDs, but then had to be redigitized when it moved onto the Internet or into applications. Can you talk about what redigitization means in this case?

Ahmed Bashir (1:26)

Yeah, absolutely. So obviously, we went from vinyl to tapes and then ultimately CDs. And when you talk about the Compact Desk, it was a digital format, right? I mean, we went through a format. I mean, these were optical drives, but they were showing digital content. And so to some degree, we had satisfaction of knowing that we had moved to the digital era, but it wasn't going to support the iPod.

It wasn't going to support the shuffle. It wasn't going to support Spotify playlists, for example. I'm just using those as examples. So we needed to fundamentally rethink what digital music needed to be in order to support all the new scenarios that we were going to be able to support with the portability that was to come with the mobile era, right? So with the mobile era, you had the iPod shuffled, the one that you could clip on to your at leisure apparel and on you go. And of course, now you have AirPods and you have music on your watch and everything else that we've seen over the last 20 years.

And it's all because we were willing to re-digitize. And of course, this has happened several times. When we moved to the cloud, we didn't just take databases as they were on premise and just move them over. We created an entirely new data interface through web applications and APIs and microservices. But then also, we took that information, we created a change data capture mechanism so that we can move the same exact data every few minutes, every few hours over to create a reservoir, what we call the data lake, which then gave way to a snowflake, which gave way to a Databricks, which gave way to a BigQuery. And so we created this entire alternate universe where analysts could work and scientists and engineers, data engineers could work and define new features and find new opportunities and pattern match, and understand trends and usage, and run their business without disturbing the customer and the customer interactions, right? So we've gone through this time and again, and I think we're going through the next one right now.

Craig S. Smith (3:37)

Flagship product is called computer, is that right? And it's a conversational quote unquote teammate that syncs structured and unstructured data from existing tools and builds a knowledge graph.

Ahmed Bashir (3:56)

Yeah, I mean, absolutely, Craig. We'll start with the name. Craig, I mean, I think it's a nod to the fact that this is, we're going through sort of a re-digitization at this point.

We've done this before, you know, when we went from, you know, having, you know, accounts and books to spreadsheets and, you know, spreadsheets to, you know, more sophisticated accounting software to Cloud and web software. And then, of course, from Cloud to, you know, what we see with Databricks and ML engineer and data scientist and business analyst type software, which needed its own, you know, data format. And again, you know, we need yet another data format that is AI native, something that can structure and index and join and relate and negate information so that it can provide really, really high value search results, answers, and of course, take the right actions. And so, you know, this new expectation, this new demand from AI is not just about being able to speak in natural language or, you know, make up very interesting and compelling sentences. But we also have to combine it with truly understanding the work that we do, the role that we play within the business, the customers we serve, the products and services we create. These relationships have to be understood. There has to be a meaningful governance and orchestration layer on top of it. And fundamentally, it has to help us with our search, with our answers, with our actions.