That ‘shiny object’ syndrome. AI has become that desirable, must-have trinket. Leaders want to have AI and ensure that employees and customers know that they placed their company at the forefront of technology. 

Let’s illustrate this with an example: When working on a project with a major insurer, we implemented and supported the adoption of a classic NBO/NBA (Next Best Offer/Action) model that helped segment customers who were most likely to purchase an insurance product. Everyone, from leaders to engineers, were understandably excited by the inclusion of AI, and collectively, we focused on the key success factors that were set: improve business outcome and reduce risk without disrupting the user.

Hence the design of the overall solution took into consideration minimal disturbance to ways of working and existing business processes, minding people, processes and technology and potential value. While the solution was a complete redo of the business logic which entailed replacing rule-based logic with an AI model, the front-end remained the same. Same way of working, same team, same process, just smarter and focused, rather than triggering end-to-end confusion.  

So, what was the outcome? Nearly an 18% uplift in conversions across the board without additional training for the call center agents, with no transition downtime. 

It was a big deal. The solution incorporated a complete redo of the NBO/NBA business logic, replacing rule-based logic with no changes to the user interface. It was the accuracy of the data and model predictions that made all the difference. 

While this anecdote serves as a clear example of AI done right, it’s fair to ask how one can achieve such an outcome, repeatedly. 

Navigate successfully to get AI done right  

We’re in a transitional period where applications and systems are being infused or replaced with like-for-like purposes. Powered by the flexibility and accuracy of AI, this period will soon be followed by native AI systems and the emerging technology of physical AI: robotics. 

While there’s still some ways to go with respect to these advanced use cases and Agentic AI, the application of AI today offers much promise. It’s well-known that hyper-personalized recommendations to customers using AI lends accuracy and timeliness to online shopping. As a result, customers are likely to hit the “Add to Cart” button more often. With an increase in sales, the inclusion of AI serves as a positive disruption to business but never disturbs customers with a new journey or additional “clicks”. It provides better business outcomes, leveraging the same digital experience with greater customer satisfaction. 

On the contrary, what if customers had to switch to a new application to receive such recommendations? This is not very different from a brand-new automotive design that requires the use of additional buttons and functions to activate adaptive cruise control when one switch works just fine. Because at the end of the day, it’s not about how “cool” your AI implementation is. It’s about whether anyone wants to use it without needing a tutorial. 

As you can imagine, in both instances this will not sit well with customers whose emotional reactions will range from mild annoyance to utter anger. While those loyal to such a retailer or automobile manufacturer might adapt to such changes, most people will defect to a competitor. 

So, how can you tell whether you’ve adopted AI the right way? The criteria would be the disruption of business outcomes vs. disturbance to users. During the AI-infused period, getting AI deployed in a streamlined, almost invisible way is a key success factor for many user-centric applications. We can say that no one likes changes, but it is safer to say no one likes unneeded changes. So, if a business process works well, a seamless and native implementation of AI would be a winner. In fact, no one really needs to know that such an AI model is in place. If you absolutely must, it should only be made visible when counting the beans at the end of the month. 

Doing AI right the first time is possible 

Before we address this challenge, let’s examine how such businesses worked prior to the inclusion of AI. If we revert to the above insurer example, rule-based legacy systems did little when it came to segmenting high-value customers from others. We’re talking about machines that ran within the confines of programmable rules, and that could not cope with changing flavors or adapt to new data sets and reality. AI provides the flexibility to achieve these capabilities, that, in turn, can become a competitive edge, if handled right. 

Disrupt, don’t disturb: When it comes to the interfaces used by calling agents, everything remains exactly as it is. There is no change in the way callers go about their business nor is it necessary to reskill them to use a new system. In fact, callers can focus on customers who have a higher probability to buy, so that outbound calls will be more impactful. In similar fashion, the online shopping experience can continue using the same interface, while ensuring that the upsell and order value is optimized. 

Benchmark, don’t just test: AI needs to be checked and tested against the performance of legacy systems, to see whether it meets or exceeds performance. It also must be benchmarked against other models and ensembles. A superior model is most likely to offer a competitive edge, not just technological advancement and flexibility.  

Apart from these two factors, it’s just as important to have your customers and users play a key role when building and making business justifications for AI investments and projects as it will most likely optimize business benefits. This can be achieved by understanding user psychology and providing them with experiences that are streamlined with their expectations, avoiding digital stress.   

Final thought: Begin with the end in mind  

With its predictive and generative capabilities transforming the workplace, infusing AI into our digital ecosystem is already happening. As a guardrail, it is key to concurrently couple it with a business case and problem that will be solved. Yes, AI is transformational but it’s important to remember that it is merely a means to an end.   

In a nutshell, it’s vital to identify a clear problem statement we are addressing, a desirable business outcome while also creating leadership consensus around it, before taking the first step of diving into the technology and the excitement of innovation that comes with it. In other words, begin with the end in mind.