Now that we have all got over the initial flurry of being astounded by AI chatbots and their very human way of handling conversations, a lot of clients and companies are moving on to the serious, practical side of implementation. Everyone seems to want to fit AI into everything, and once you reach this stage, the excitement quickly gives way to the realisation that most of the challenges you face in getting it to work are the same old ones we have always faced with integration and software adoption.
To be honest, I personally find chatbots, while amazing, don’t quite beat the real thing. If you have ever worked in a company with helpful colleagues, it is impressive that a computer can do that job, but the function itself is not new. What is amazing is doing it at scale. Instead of asking ‘Janice in Finance’, you ask an AI, and an AI can answer thousands of people at once.
And here is where we hit the same old problem; even ‘Janice in Finance’ isn’t a divine miracle, so how does she do what she does, and how can we get AI to do the same?
The first issue is data. Have we given the AI all the data it needs? Have we given it access to live sources, or are we expecting miracles from a one-time data load? Have we structured the data in a way that makes sense? Just dumping it into a database without context or rules will not work. AI needs to know how information connects in order to make useful deductions. and while there are incredible tools that can help with this, when it comes to making assumptions, particularly financial or legal ones, the wrong connections or lack of context can quickly become very dangerous. This has been a problem since long before AI.
The second issue is workflow. Once you have gone past the novelty of “oh, this is a chatbot”, if it does not fit into the user’s normal workflow, they will not use it. Make sure that any new AI features work within existing processes. This not only helps adoption and avoids unsettling users, but failing to do so actually takes away some of AI’s advantages. For example, if you are assessing whether an insurance claim is potentially fraudulent, that used to be done using a set of complex but static rules which produced a percentage likelihood. AI can now perform that calculation far more accurately, but you do not need to change the interface at all. It simply improves what the user already sees, which is exactly how good AI should feel: an invisible improvement that makes the job easier without disrupting needlessly.
The third issue is adoption. Forcing new features onto people drives them mad, even if those features are shiny and fashionable. It makes users nervous, and they start to see change as a threat. There are thousands of books and experts on adoption because it is a very old game. My least favourite part is when senior leaders believe the technology will replace staff and start talking directly to business teams about “efficiency gains”. This immediately puts people on edge. AI already carries an overtone of job replacement, so pushing it from the top without the right tone can do real harm. It needs to feel like an aid, not a threat.
Finally, there is the way we introduce AI. Making a big song and dance about any new technology might be great for publicity, but it rarely impresses the people that do the work day to day. Pitching AI as something that makes people’s lives easier and helps them do their jobs better works far more effectively. It should be seen as a tool that helps teams perform better and, ideally, be rewarded for it. Declaring “AI is here, the world will never be the same” is exactly how you generate resistance and backlash. Features will be viewed as threats rather than improvements, and that helps no one.
In short, make sure any AI you introduce actually adds value rather than simply being “the new hotness”. Even though AI is a fundamentally transformative technology, from your users’ and business teams’ point of view it is still just another software adoption process. Use all the lessons you have learned over the years to make it as seamless as possible.