How Smith+Nephew surpasses staff expectations with ServiceNow’s Moveworks AI assistant

When Thomas James Smith founded his chemist business in Yorkshire, England in the mid-1800s, its main specialism was supplying cod liver oil to hospitals, and then providing bandages and surgical dressings for soldiers fighting in the First World War. 170 years on, Smith+Nephew is now a global medical equipment manufacturer, with 18,000 employees providing advanced technologies like skin graft dressings and artificial joints. As it continued to make advances in the medical world, the business wanted to apply the same advances internally by offering a more modern employee experience. According to Sebastian Pasch, Head of End User Services at Smith+Nephew, the company wanted to bring a more consumer-like experience into the enterprise environment, moving from search results to generative AI answers, in order to meet staff expectations: Google-like search experience is no longer good enough, people just expect a ChatGPT-like response, specific to the question asked. The company originally tried out a Ritual agent, but found that to be old-fashioned technology that required the team to program its own workflows. Pasch notes: Understanding human language should no longer be a problem, that's already solved by AI. Smith+Nephew then moved onto testing out ChatGPT and built its own proof of concept. But it soon realized the technology was much more complicated than it looks from the outside, and so it decided to purchase a product instead of building the tool itself. After looking at the available options, the firm quickly arrived at Moveworks, an AI assistant platform that automates workplace tasks and answers employee questions. Use case A major factor in convincing the firm to pick Moveworks was being able to talk to a customer that had already undertaken a similar project, Pasch explains: At the time, we spoke to Honeywell, and we saw their bot and how it works. It was making the real difference.” Honeywell has been working with Moveworks for over five years, using early transformer models like the BERT model through to ChatGPT architectures. During the selection process, discussions were already underway for ServiceNow to acquire Moveworks. As the former was already in use at Smith+Nephew, this was a good sign, as the two platforms would get closer together once a deal went through. Smith+Nephew made the decision to roll out Moveworks initially within its internal network so it could start using the service quickly. It chose IT services as the first project. Pasch says: We know how good our knowledge is. Knowledge has been at the core of everything in IT for service desk operations, ITOM, for years now. So, we know our service knowledge is actually not too bad. Considering that we are going to provide responses to customers with the AI, we also knew we need to focus more on the self-service knowledge specifically. You don't want to provide instructions to users that only an IT professional can actually follow. Not wanting to lose more time in terms of debates inside the company - we all know how company politics work often enough - we said, all right, IT, go forward.” During this process, Pasch was already being approached by staff from other departments, who had heard about the project and were interested in participating. The business also had some gray areas where the initial deployment covered areas that weren’t exactly IT, for example devices like VR headsets being used in medical education: Those were not originally in scope, but there were specific challenges with people going out on maternity leave. And we said, give us your knowledge, as long as you can provide this, no problem. We can incorporate that. That's not real effort. This is where we started and landed first. What’s in a name? When it came to developing the company bot, choosing the name proved a rather contentious issue. Many people came forward with their own idea of what the name should be, as Pasch explains: You should call your bot X, Y, and Z. 20, 30, 40 ideas. It was none of those. It was user research that we did, and we also did the same research in terms of what's the brand looking like, and taking this extra step in the end in terms of making it corporate-brand compliant took us another month with marginal differences. With the help of the user experience team, the company finally settled on the name Ellie. The AI assistant technology has proved popular with staff, with an 80%-plus rating for positive feedback. Pasch noted: There's a thumbs up and a thumbs down message on every single message within the conversations. I did not expect we were quickly achieving 80% plus and were able to sustain this. That's one of the biggest [surprises]. AI naysayers That’s not to say everyone at the business is passionate about using AI. He added: There obviously are a few people who do not like AI. They just say, keep away with this, I will just stick to my processes, I don't want AI. That's especially when you then try to make changes on top where you remove redundant notifications. We are shutting down email notifications now, and people must use Ellie, our bot, now, otherwise they no longer can follow what's expected of them. While there’s an expectation for staff to use the new technology in certain instances, the company has tried to be responsive to feedback. Pasch recalls: We reacted to every single feedback that we received in a written form. If people just clicked thumbs up, thumbs down, fine. But if you took the time to actually give feedback, especially negative ones, we came back to the guys in person, in email communication, and reacted to this. We have been demonstrating, your feedback is not for nothing. We are actually working on improving this. Sometimes the feedback proves valuable, identifying additional knowledge that can be added or outdated knowledge that has previously been undiscovered: Sometimes it's people not understanding it. Although all of our communication was 100% on, 'This is IT support only, this is Ellie, your IT help first stop', people still asked it an HR question, procurement question, whatever, and were frustrated and gave negative feedback on, this bot doesn't respond. Pasch attributes this to the consumer experience: Everybody's expecting, you have just put in ChatGPT and a nice user front end, and it's now a corporate thing. No, it's not. We have created knowledge that's actually giving correct responses that you can cite. Pipeline The interest from employees to use the AI assistant for more functions has prompted Smith+Nephew to begin expanding the tech to other service lines outside of IT. The company has built a pipeline of potential use cases from the people who were approaching the project team at the early stages. From that, it has been able to develop a clear priority list that it’s still working through. Pasch explains: In the interim step, and right now, what we plan to do is expanding by knowledge. So what's already on the ServiceNow platform anyway across finance, procurement, and so on. Make it accessible to Ellie, and here you go. For other businesses wanting to get similar AI deployments right early in the process, Pasch shared some advice based on his own experience. First was to get the knowledge right – make sure the organization understands what knowledge it wants the AI to consume, and that s elf-service knowledge is up to date and easy to digest. Second, avoid a sprawl of tools. Pasch liked the aspect of the Moveworks products that means the company can do many things from just the one platform. Third and final tip - build a pipeline of use cases and apply them according to what’s easy and quick to achieve. He advises: Take the wins. Make sure you launch, even if this is a smaller bucket, where you have confidence on. Because what we really have seen is a big benefit is, it does work, we have proven it, and now everybody is coming to us asking, I want to be next, I want to do this here and there. So we have hundreds of use cases already lined up, and we decided low-hanging fruit first.