As the VP of Adaptive Intelligent Apps - Oracle’s AI applications division - Clive Swan has experience and insight into what enterprise customers are demanding, when they’re considering the use cases for AI in the enterprise.
Oracle’s AI strategy can be defined as having three pillars:
It sells AI-embedded apps, which are functional solutions that drive up recommendations (smart actions). They typically take first party application data and merge it with third party data, feeding that into predictive models.
Oracle is also doubling down on AI UX, which aims to move the enterprise away from a ‘one size fits all user experience’ to one that is personalised and optimised for each user.
Finally, Oracle is investing in digital assistants. Because both smart actions and AI UX are exposed as REST services, any of it, if appropriate, can be surfaced as a digital assistant.
Unsurprisingly, given the hype around artificial intelligence, Swan has experience of buyers approaching Oracle with demands for ‘AI everywhere’, thinking it will solve all of their problems. However, Swan is frank in his acknowledgement that mostly what customers really want - when he gets into conversations about what problems they’re trying to solve - is basic automation. Automation to solve use cases that rules-based software can’t tackle.
Firstly, most people come in and think that AI is going to do some unbelievably magical stuff - “I want to remove all bias in my HR processes”. In my experience, and I think this will play out in the majority of cases, I think folks in 80% of cases are looking for pretty boring automation. Automation that can’t be done by rules-based software.
For example, let’s say you get an invoice in and there’s no PO. There’s no effective rules-based system that can solve that problem. So every AP department in the world typically gets people, usually offshore, entering those by hand.
Swan said that at Oracle, for example, 65% of the company’s invoices are handled automatically with rules-based software. The remaining 35% fits into a variety of buckets, one of which includes ‘no PO’. As a result, it has developed a solution to default the account code where there’s an invoice with no PO. Swan said:
I’ve been staggered at how excited all our financials customers are with that. And now we are looking to build on this with incremental capabilities. For example, we are looking at solving for three invoices coming in with only two POs. It’s impossible to do that with rules-based software.
So whilst they may come in and say crazy things like “remove all bias”, when we actually start talking they get very excited about those capabilities. I think a good 80% of the solutions in AI will be pure automation.
Swan gave another example, where Oracle has developed a recruitment solution that matches CVs to job descriptions. He said that it isn’t trying to select candidates, it just trying to solve the problem of ranking the best fit. Swan explained:
I would argue that if we do as good a job as a human being, but no better, we will do a better job. And that’s just a standard software argument. If you give a recruiter 100 CVs and ask them to find the best 10, they’ll find the ‘best 10’ in the first 50 CVs.
If you give a piece of software 10,000 CVs, it will find the best 10 in 10,000 CVs. Software doesn’t sleep, software doesn’t get tired. But that translates into a genuine uplift in outcome.
Oracle’s AI priorities
Swan also took time to outline some of Oracle’s AI priorities within the apps teams over the next 12 months. Both of which provide interesting insight into some of the challenges vendors such as Oracle are having to think about scaling AI for their customers.
Firstly, Swan said that the use of AI models across applications provides an interesting operational challenge for Oracle, and others. He said that if Oracle can solve for this, it could be a huge differentiator for the vendor. Swan said:
Our biggest priority over the next 12 months is getting customer success at scale with our solutions. There are operational challenges of, how do I manage an AI model? The same AI model across 5,000 customers? The whole idea, historically, has been a manual activity for data scientists to tune and tailor that AI model to the shape of each customer.
More critically, to maintain that solution over time as data shapes change, characteristics to the problem change. That’s great when you’ve got one single instance for HSBC for customer churn and you can afford data scientists. When you deliver one singal AI solution to 10,000 customers, how do we do that?
So what we’ve done under the cover is we’ve spent a huge amount of effort automating that capability. So I have half our data scientists for the past 18 months applying machine learning to the operational maintenance of the models. We will see how well that works. I see other vendors are pushing that out to the customer now. If we get that right, that will put us heads and shoulders above everybody else.
Secondly, Swan is thinking about how he and his team can scale the work they’ve done developing AI solutions across Oracle’s portfolio. Part of this involves centralising data science capabilities as a resource, as well as providing frameworks for development. He said:
My team is pivoting from being a build team to an enabler team. The objective was always that we’d build a few solutions, come up with common patterns, common framework approaches - making sure that all the other app teams build to a common pattern, a common framework. And we’re providing a centralised data science service.
I’ve got an aspirational plan - it is aspirational at this stage - I want to get somewhere between 15 and 20 solutions built this year by apps teams working with us with that model. That gives us the ability in 2021 to completely explode.