Main content

SuiteWorld 2023 - Oracle NetSuite's vision for business AI built on Cohere and OCI

Phil Wainewright Profile picture for user pwainewright October 18, 2023
At SuiteWorld this week, Cohere's Martin Kon spoke about its approach to generative AI for the enterprise while NetSuite's Evan Goldberg outlined the expected impact on business users.

SuiteWorld 23 Martin Kon speaks to Evan Goldberg - Gamma Nine Photography
Martin Kon speaks to Evan Goldberg (Gamma Nine Photography)

While many vendors are teaming up with ChatGPT creator Open AI to develop new features powered by generative AI, Oracle has partnered with Cohere, the AI startup co-founded by one of the Google researchers who came up with the original concept of Transformers, fundamental to the Large Language Models (LLMs) behind generative AI. At this week's SuiteWorld conference, Cohere's President and COO, Martin Kon and NetSuite CEO Evan Goldberg discussed Oracle's use of the technology and its expected impact on business processes and the user experience. Goldberg told journalists later that he foresees a fundamental change in the NetSuite user experience:

What we want to do with this powerful AI technology is put it right there at your fingertips, in your everyday workflow, re-engineering the user experience of these everyday workflows. It gives us the opportunity to insert AI in a more natural way than our current UI, which works out well for our customers, but it's much more record-based.

What you saw in the vision was not as much a record-based UI but more of a workflow-based UI. We think about the whole thing that a user is trying to do when they are looking at different records and tying it all together. That to me is a more natural way ultimately to integrate this AI. We think the hugest benefit for NetSuite customers is that it can tie together all the disparate data in your company and help you make sense [of it].

All of this work on AI is being carried out in close collaboration with other Oracle product teams. He added:

It's absolutely part of a bigger Oracle initiative. Our team that does AI is working extremely closely with multiple teams across Oracle. There is a larger vision. There are other business applications at Oracle that have many of the same needs. We're working with those teams to make sure that we're leveraging AI in the most efficient and effective way, but if we work together, we're just going to be able to deliver more to customers faster.

How Cohere combines language models

For his part, Kon explained some of the underlying technologies that allow Cohere to ensure that its models can provide reliable results for enterprise use. This comes from combining generative AI with other measures. He explained:

What some people maybe don't realize is that there are different species of language models. There are generative models and that's generally what we think about — generate text, summarize text, this conversational, very natural interface.

There's another set of models called embeddings models, and they understand semantic meaning. So for example, if I say, 'What's the best falafel in town?' there may be something that says 'Middle Eastern'. That's not the same word. Lexical search wouldn't find that, but the transformer and embeddings labels enable that to be found because it's semantically relevant. When you tie those together, it becomes incredibly powerful.

This is combined with a methodology called retrieval augmented generation which searches across internal enterprise information to fine-tune the results when delivering an answer. He went on:

What we do through retrieval augmented generation is tie the generative models and the embeddings models together. You enable the models to ground themselves in authoritative information — from, for example, your very sensitive, very valuable, NetSuite data — retrieve the right information from there, semantically similar, and generate the response based on that, and then cite where it's from.

So you basically do a couple of things. One, you get around the hallucination issue, the misinformation issue. But secondly, you enable your conversational agent to have the most up-to-date information. What are our latest discounts? Well, they changed yesterday. Something that was trained two months ago isn't going to work. But also, most of the data you have very securely in your NetSuite or broader systems, you don't want to train the model with, even if it's your own model. Because you don't want that in the weights of the model, but you still want to be able to retrieve from it.

That combination of these generative models, the embeddings allowing search retrieval, and generation on very secure data environments, is going to be a massive asset online for productivity. And that's something we're very excited about. I think NetSuite's the perfect platform to do that on behalf of your 30,000+ customers and we're looking forward to being the 'dog food' for that together.

Since Cohere is deployed directly on OCI's high-performance GPU clusters, there's a cost of ownership benefit too compared to other models where the processing is done remotely. Kon explained:

A lot of these models, hit through APIs for big model deployment, will cost an enormous amount of compute to deploy ... That's going to be very important, because if something costs 10 times as much as to deploy, it might be nice in a lab environment, it's not going to be as good at scale.

'Pick something and get started'

With NetSuite celebrating 25 years since its foundation, there were inevitable comparisons between the advent of today's AI technologies with the advent of cloud computing and SaaS a quarter of a century ago. Kon argued that there are lessons to be learned from those times for early adopters who want to get ahead of the curve today. He said:

If you look at, for example, the digital revolution. I think there were some companies that spent a lot of time, enterprises that spent a lot of time analysing and — I'm a former consultant — maybe hiring consultants and prioritizing what to do first and not really moving. And others have said, 'Let's just start and do things, even if it's not necessarily going to change our entire business tomorrow morning. But we need to get experience by actually deploying.' I think the way that [NetSuite] is doing this is exactly what we think companies should do now. What we saw companies do back in the late 90s, and let's say late 2000s, with mobile — pick something and start with it.

In NetSuite's case, one of the first use cases in production will use the technology to add a conversational interface to FAQs and how-tos in the NetSuite application. This first step, he explained, doesn't involve working with sensitive data sources but provides an opportunity to get used to the technology. The next step will be to add an intelligent assistant that can access finance and HR data and make suggestions, for example, which customers would be the most useful to visit on a forthcoming sales trip. And then the third step is for the models to power intelligent agents that autonomously take actions in NetSuite. He summed up:

Those are three pretty exciting steps, I would say, in the journey. But starting with something that you do now, at scale in production, learning how to do this in your own data environment, plugging into your systems and so on, to actually get something going, that is the right thing that's going to separate the winners from the losers. I think we saw the same thing in cloud — you obviously were a pioneer — that's a good example of people on the right side back then.

In Kon's view, it's imperative for companies to get on this path because the impact of AI will be as significant as the advent of the Web. He said:

Fifteen years ago, the iPhone came out. Thirty years ago, the Mosaic browser came out. In terms of the interaction between humans and computers, this is by far the biggest change, certainly since the smartphone, I would say since the Mosaic browser. And every customer, every employee, will just demand to interact this way.

Companies that are early with it will do well, just like in the late nineties, the set of companies that were early with digital, or early with cloud, or early with mobile. Those who are lagging will see that they will have some problems. The opportunities are absolutely immense in enterprise ... I think the productivity gains, the ability to reach new markets, the ability to tap into data — which is something that NetSuite obviously is built on and the broader Oracle is built on — how do you enable employees and customers to access data and information very quickly? To get answers to questions, to generate insights, make better decisions, and ultimately have what we would term as being the impact on intellectual labor that the steam engine had on physical labor.

Goldberg agrees that data is the key to making the most of the opportunity with AI. As he told journalists:

We feel very strongly that the best AI comes from the best data in your business — and the best data in your business is represented by NetSuite and having this system of record across so much of your operations. We'll see how that plays out. But I think it actually accentuates our advantage. That's what we're trying to show today — it's the interconnectedness of all these processes within your business and if there's an AI that can really help you tie everything together it becomes this incredible assistant chief-of-staff, whatever you want to call it, to help you really supercharge your growth. That's our approach and our view and we're going to be investing very heavily to bring that vision forward.

My take

NetSuite's first steps into generative AI are cautious — humans are kept in the loop at all times in the features announced this week, rather than the autonomous agents that Kon speaks about as the final step in this initial journey. But by working in step with Oracle, there's scope to make rapid progress because of the sheer scale of the numbers involved — hundreds of thousands of customers and millions of users across every variation of function, industry and geography. NetSuite adds an important dimension of businesses with limited internal IT resources, where it becomes essential to make the technology as simple and easy to learn and consume as possible. It will be important to be on guard for mis-steps along the way, but the vision of bringing data and processes together across an enterprise and making them more naturally accessible as part of people's everyday workflows is very appealing.

A grey colored placeholder image