Страховая компания IAG внедрила ИИ-ассистент Now Assist для автоматизации ответов на вопросы сотрудников

Key points - IAG has moved from ServiceNow's conversational chatbot to its Now Assist AI suite to boost self-service deflection rates for HR queries. - The insurer is reformatting about 730 HR knowledge base articles to be AI-optimised, with the work due to wrap up in September. - Now Assist-based virtual agents delivered an average 67 percent deflection rate in their first month, more than double the earlier chatbot's 30 percent. IAG is using artificial intelligence to meet self-service goals for employees seeking HR-related information, The insurer has long had ServiceNow as part of its HR technology stack, but in recent times has sought to use virtual agents to help connect staff to relevant information and responses. Digital planning and engagement manager Carli Jackson told the ServiceNow World Forum in Sydney that an internal desire for more self-servicing of HR requests emerged about three years ago. Initially, this was delivered through ServiceNow’s employee service center platform which, according to product notes, incorporates an “automated, conversational chatbot” that scans knowledge articles for answers. The insurer’s HR team used the capability to build “five conversations” for common HR questions, including around benefits, leave, pay, and time and attendance. Jackson said the results were promising, headlined by a “30 percent deflection rate” of common requests being automatically fulfilled, rather than requiring a HR team member to get involved. “We knew that we were heading in the right direction,” Jackson said. “We can start to change the way people engage with us, and start to pivot to [a] more chat-based kind of platform.” More recently, IAG has started to make use of Now Assist, ServiceNow’s native AI suite, to increase the range of questions for which responses can be self-served, raising the deflection rate. The company was given a “proof-of-concept instance” of Now Assist and, at least from a HR perspective, was able to gain internal approval “to put our actual real knowledge base in there so that we could test” it, Jackson said. “We did get an endorsement to enable Now Assist live in our environments for 12 months and see what it would actually do.” Through testing, the insurer identified that its knowledge base articles, used as the foundation for the AI to work from, required reformatting to be parsed by the AI and to improve the accuracy of results. Part of the challenge from a HR perspective was getting the AI to correctly delineate between guidance prepared for Australia versus New Zealand, and to reflect that accurately in its responses to staff. “We hadn't formatted our knowledge in a way that was AI-optimised,” Jackson said. The reformatting of the HR knowledge base is being led by a small squad that uses Copilot to rewrite and prepare the content, and then asks domain and subject matter experts to run final checks of the material before allowing AI to access it. “We started that [work] towards the end of last year, and we're aiming to wrap it up in September,” Jackson said. In total, about 730 knowledge articles need to be checked and potentially rewritten, although Jackson indicated that number is likely to be higher at the end of the process. “We're probably going to come out of this with more articles, as some get merged, some get split, and [to cater for slight differences in] the Australian and New Zealand side of things,” she said. Now Assist is live, despite the ongoing checking of the remaining portions of the underlying knowledge base. IAG has Now Assist-based virtual agents, again targeting some of the more common HR questions that staff ask. While there is some variability in the metrics, deflection rates using Now Assist-based virtual agents, compared to the older conversational chatbots, are significantly higher. “We have highs and lows,” Jackson said. “We have days where we're over 70 percent [deflection], but in the first month, we had an average of 67 percent. “That is more than double what we had just with our five conversational workflows.” Jackson said that IAG hoped to consistently achieve a deflection rate of around 80 percent, and that this was coming into view, although it is still early days. She added that in the immediate future, IAG will look at its first agentic AI workflows in the HR domain. Ry Crozier attended ServiceNow World Forum in Sydney as a guest of ServiceNow.