Many organisations get stuck at the proof of concept stage when they come up against the task of trying to change the way the enterprise operates to embrace AI, according to research from Gartner.
Speaking at the Gartner Symposium on the Gold Coast this week, Bern Elliot Vice President and Distinguished Analyst with Gartner Research, explained that AI projects are very different to traditional IT projects.
Gartner has five levels of AI maturity from awareness to transformation. Organisations face many different hurdles while trying the scale the maturity curve, with many struggling to scale the benefits across the business.
Often at the proof of concept stage every project is a one-off Elliot said.
“It’s hard to get from one offs to scale. So you do a project, a virtual assistant and it’s good, but you didn’t really retain anything, you don’t have a process to do the next one. And the next project we do so starting from scratch.”
Another major barrier organisations face is putting a model into production requires working with other parts of the organisation, such as IT or the data science group and the part of the business such as marketing or logistics which will own the product.
“The processes and methods for maintaining AI projects is different than the processes and methods for maintaining more traditional software projects,” Elliot said.
“Most software projects, the expense is up front, you put the development in and then it goes into production. You may need to tweak it but the handoff between development and production is clear.”
“The handoff for AI projects currently is not very clear in most organisations, because these projects need to go back and forth for production and tuning as the model matures.”
Many companies stay at the AI centre of excellence level because the fourth step requires changes to the organisation and how it operates.
However, Elliot noted if organisations can implement what Gartner calls the hybrid AI-model they will see a real effect on the workforce as it starts “really changing what it does and how it does, because they’re starting to leverage these new sets of tools.”
This level also requires a stricter level of governance, with a policy of how data can be used.
“The governance has to be able to look across all the data that’s being used across the different projects, because a group that’s working on sales, if they’re using marketing data, they may get a hold of personalisation information, and they don’t really know that they are not allowed to use this kind of data for this kind of a project,” Elliot said.
Some companies choose a different option, leapfrogging the fourth level and spinning off a new business that operates with “continuous AI orchestration.”
“A number of organisations actually go straight from three to five because it’s too hard to change the organisation, but they really can envision a new line of business or a real change.”
Getting started with AI
Elliot recommended always starting with the objectives and business results, then identify use cases which are the “lingua franca of AI projects.” From there, organisations should move on to assessing the applications and core technologies to accomplish the use case.
That process should be revisited and repeated as AI maturity improves, Elliot said.
“You revisit your plans, your priorities and your use case descriptions regularly. In some cases, you may iterate just on the use case objectives.”
Elliot recommended having several AI uses case, maybe 20 — three to five in each of the major areas of a business. Although mature organisations will have as many as 200 defined use cases.
“Companies that are really have really progressed relatively far in their high execution will have a backlog of AI projects, which means a backlog of use cases.”
For organisations just getting started with AI, Elliot recommended prioritising the projects which can be addressed with off the shelf tools, rather than strategic AI projects which are expensive and require large amounts of customisation.
For example, chatbots and virtual assistants are an example of ‘AI through the enterprise’ which can be purchased as add-ons to your existing applications.
“All of the customer service vendors, HR vendors, marketing vendors are working hard to build have pre-integrated learning and insight capabilities into their models.”
According to Gartner, these projects should deliver ROI is less than two years.
“This is the most efficient way to start benefiting from AI in your organisation,” he said.
On the other end of the spectrum are strategic AI implementations that require large amounts of customisation such as identifying trends and buying patterns by segments and products. These strategic projects will require a multi-year investment before you start getting results.
While these projects can deliver a competitive differentiation, Elliot cautioned “You don’t want to any of those projects.”