Weighing in on the Value of Big Data and AI: A Chief Technology Officer's Perspective
As Vice President and Chief Technology Officer of Nexis® Solutions, Stephen Iddings has been closely involved in driving the Artificial Intelligence (AI) strategy at LexisNexis. In an interview, he shared how LexisNexis® uses AI and Machine Learning (ML) to innovate for customers and find operational efficiencies. But he warned that AI is only as useful as the quality of the data going into it and questions it is trying to answer.
What are the main ways in which LexisNexis is using AI technologies?
I would break this down into three big areas. The first is customer innovation—we are putting AI in front of customers through our products. The second is operational efficiencies—we are using AI to help us operate more effectively, either through fabrication processes or our business system processes. The third is that we are exploring new frontiers—this is work that is not in production yet, but we are working on it in labs or innovation areas to be ahead of the curve for the future.
Can you give us an example of each of these areas?
AI contributes to improving how customers interact with products. When a user types a query into a search box, AI predicts and auto-completes the query. Then the query is mined out for semantic parsing and phrase extraction so we can understand user intent. Then after the search is done, AI models apply relevance based on previous searches to make sure the most pertinent material is at the top, and where a user has searched for a particular company it might bring up the company card associated with that company.
In business systems, Robotic Process Automation (RPA) is starting to play a role by automating menial tasks that use technology, and AI is being used in the business system space on customer retention models and identifying the likelihood of a customer to cancel. This gives sales representatives greater visibility prior to renewal time to understand the likelihood of a customer cancelling based on customer service calls or usage of the product.
One of the next big things is chatbots, or conversational agents which use Natural Language Processing (NLP) and ML to become smarter—they will be understand the user’s query, classify it by the type of question they are answering, know what they have researched in the past and what they are currently working on.
What area(s) of the business has AI and big data affected most, and how has it changed or improved the way you did things before AI?
AI has introduced a lot of opportunities for savings by offsetting or replacing manual efforts. In 2009, we put in place our first massive scale named entity recognition and named entity linking solution which was able to recognise people, places and things in content and link them to an authority.
When we first did this, people had to manually tag attorneys and in a three-year period we did 3.5 million attorneys and spent maybe a million dollars. But when we implemented our first big data solution to recognise entities, we tagged 24 million attorneys in our content in one 24-hour run. We did a comparison with human-tagged entities and found the quality was on a par and better in some cases.
That was the first really big, compelling realization that we can use AI to augment editorial tasks. These developments have allowed staff to take on higher value jobs. In the editorial space, RPA has freed up human intellectual property and resources to focus on things that are of higher value to the business.
What is the role of data in the effective use of AI technologies?
There is no AI without data, you have got to start with data. You need training sets and content to help train your Machine Learning models. You have got to have content to train your image recognition models. So, data is what drives it all and content is king.
We have spent a lot of money as a company to collect a large corpus of content and quality data. We have content from 47,000 different sources and licensed publicly available web content which we can normalise into a structured or semi-structured format like XML. A lot of companies have their own data on customers. When we start applying enrichments and knowledge graphs that span our data and theirs, customers gain valuable context for research, media intelligence and entity due diligence.
What advice do you have for companies considering using AI?
Think of your customer, who they are and how receptive they are to AI functionality. I have been involved with the due diligence in a lot of acquisitions and the pattern I have seen with successful start-ups that we acquire is not about how good they are at AI. Everyone pretty much uses the same stuff; I haven’t seen any secret sauce out there. But the best are not those who are best at answering questions, but the best at coming up with the questions that need answering.
You can typically solve anything with AI and a data scientist’s role is to start with asking: ‘what is the question I need to answer, what content and features are needed to answer the question and what does the model need to be to answer the question, iterate and refine.’
What advice do you have to for companies looking to acquire more data and do data integrations?
With software development, there are a lot of open source technologies, enabling companies to get hold of AI code and tap into available data scientists’ skill sets. But what I see is that getting access to big data in training sets is still a barrier and a lot of start-ups are making massive investments to buy the data or partner with a big company like ours. The reason that Nexis® Data as a Service is booming is that it’s an easier barrier to overcome for partners to get access to the data to add value to their business.