We know how we want our companies to work. Enterprises ought to be customer-focused, responsive, and digital. They should deliver to each employee and customer exactly what they need, at the moment they need it. The data and technology to do this are available now. And yet, all too often, investments in artificial intelligence (AI) fail to deliver on these promises. Why?
The problem CEOs need to understand is that AI built without discipline fails. However, when AI “understands” what’s going on within a business, it becomes more than an incremental improvement. It becomes a key contributor to overall efficiency and effectiveness.
Consider the case of Allstate Business Insurance. Business insurance is challenging to sell. Compared to home or auto insurance, business insurance protects against a variety of risks (both legal and physical) for a variety of types of companies, from hairdressers to retailers to aircraft parts manufacturers. This creates a problem, because the ten thousand small insurance agencies spread across the US have to sell a product that raises countless questions. The sales agents in those companies frequently called Allstate’s support lines looking for answers.
All those calls were eating into Allstate’s profits. Each rep needed 16 weeks of training to get up to speed, but high turnover meant a continuing investment in new training. Answers weren’t always consistent, which generated an additional problem – agents calling back a second or third time, hoping to get the answers they preferred.
This seems like the kind of problem that AI could excel at, since it spans complex challenges, demands consistency, and depends on a large set of complex information in the form of policy documents. So Allstate executives, including its president at the time, Mike Barton, launched a project to create ABIE – the Allstate Business Insurance Expert. This AI resource was intended to generate a consistent, scalable answer machine for insurance agents with questions about the company’s business policies.
In theory, you could dump all the documents full of answers at Allstate into a big hopper and tell the AI to look up whatever it needed. But that idea ignores a key step. Any AI consuming that information needs an internal model of Allstate’s business – a model that understands, for example, that “clients” meant the same thing as “customers,” and that banks are financial institutions but car dealers aren’t, and neither are financial advisors. The model would also have to include knowledge about types of insurance products, types of risks, and regulations that varied by state or city – and how all those variables related to and interacted with each other.
The key to creating ABIE in such a way that it could understand all that was the creation of a set of classifications called taxonomies – taxonomies of business types and risk types, for example. Those taxonomies were linked together into a structure called an ontology. The ontology is the model of the business, and it is what allows a system like ABIE to ingest a policy document or other information, to represent it internally as a collection of valuable content, and, most importantly, to be able to surface that content when it is applicable to an agent’s question.
ABIE took a year to build. But once launched, it became a key resource. Call center volumes went down 10% as agents selling to potential clients realized they could get answers quickly from the online system. Callbacks plummeted because agents were getting a consistent, right answer on the first call. The call center staff themselves started using ABIE as a resource, and were able to get productive in just 12 weeks instead of 16. “We look at ABIE as a gigantic success factor for Allstate Business,” Barton said. And Allstate eventually put ABIE online where anyone – including potential Allstate Business Insurance customers – could get access to the answers themselves.
AI cannot start from zero. It builds on information structures and architecture. Artificial intelligence begins with information architecture. In other words, there is no AI without IA.
AI works only when it understands the soul of your business. The ontology is the key to that understanding. It is a representation of what matters within the company and what makes it unique, including products, services, solutions, processes, customer descriptors, organizational structures, methods, and every imaginable type of data and content. If you build it correctly and apply it appropriately, it makes the difference between the raw promise of AI and delivering sustainably on that promise.
An ontology is a consistent representation of data and data relationships that can inform and power AI technologies. It includes elements that have previously been described with the terms data models, content models, information models, data/content/information architecture, master data, or metadata. But it is more than each of these things in themselves.
You cannot buy an ontology from a technology vendor, because it is unique to your industry and your company. Building it requires a systematic process of the kind that Allstate Business Insurance and my company undertook to classify and organize all the information within that company. Building it is a step-by-step process that starts with observing how the company solves problems (like the questions asked of Allstate Business call center agents), imagining better and more productive ways to organize those solutions, identifying who uses those solutions, and building detailed use cases. With that work done – and it is, for any decent-sized company, a major effort – you can develop organizing principles for data and content.
It’s also possible to take a bottom-up, data-centric approach, observing all the data that the company uses and organizing it into a master set of taxonomies that, together, comprise the basis of the ontology.
Whatever approach you take, you will be creating an asset that pays off in multiple ways. As Allstate Business Insurance found with ABIE, a system for organizing and understanding all the data becomes valuable to multiple audiences, and you can evolve it to encompass more problems and solutions. And with an ontology to build on, you can apply AI in ways that are sustainable as the technology improves and the data becomes richer.
It’s certainly possible to apply AI without doing all this work. But AI point solutions nearly always fail – sometimes right away, and sometimes a little later when it becomes clear that they are just adding a layer of complexity to a set of already complex problems. You can’t build a modern building on an old, rotten foundation. And you can’t really leverage the power of AI unless it’s built on the foundation of an ontology that models what matters in the business.
This article was originally published in CEOWorld Magazine.