Hg's 2024 Digital Summit in Amsterdam brought together over 120 technical leaders to explore emerging use cases and the latest advances in AI. During the summit, Hg's James Cope, Business Systems and Transformation Specialist, speaks with Inbal Budowski-Tal, Senior Director of AI/ML at Pendo, on the different levels of AI adoption in respect to product development.

The power of AI can be alluring to software companies as they push the innovation curve. This can easily lead to innovation becoming purely tech-driven, such is the desire to roll out the latest greatest AI-powered solution.

The risk to doing this, however, is that it overlooks the needs of the customer, potentially diminishing its value. Rather than be seduced by the technology, the bigger value arises when thought and effort are given to making something that people actually need, and will use; this is product management 101. Customer first. Technology second.

Introducing AI in to product development has evolved over the years. As businesses become increasingly familiar with its benefits, and AI models become more sophisticated, the world is moving towards greater automation, where the AI will be in control, with no human involvement needed. There are limited examples of this today, the most obvious being the self-driving vehicle.

Four Levels of Adoption

The first level is no AI at all, where the process is driven entirely by humans. The second level is akin to having a digital assistant, where the human drives the process and the AI provides insights and curates certain signals. Level three represents a step change, as the focus shifts to the machine leading the process, and the human editing and approving what is being suggested; they, in effect, become the assistant to the AI. Level four is full process automation, with the AI in control.

Level one is clearly no longer an option for any company – let alone software company – with serious growth ambitions. Equally, the world is not yet ready for companies to be operating at level four. Not only are the regulatory guardrails needed to protect consumers (in terms of data privacy) and laws updated to take into account myriad ethical considerations, but also the AI models themselves are not yet mature enough.

Product leader maturity

Today, most companies would typically fall between levels two and three.

Example of products operating at level two AI maturity would be chatbots, which companies have grown accustomed to deploying in recent years. These are used in areas such as e-commerce, customer services, sales and marketing to better serve end customers. With regard to level three maturity, this is still typically the reserve of big tech companies. For example, the AI used in Netflix, which is entirely machine-driven as it learns viewers’ content preferences and makes recommendations. It is a curated experience.

AI specialists like Budowski-Tal, who joined Pendo to establish a dedicated AI department, are investing a lot of time thinking about how AI can make that transition from giving insights to providing suggestions; and by doing so, help humans to improve their performance. Such a jump will depend less on the maturity of AI models and more on the maturity of product leadership. Understand the commonality of users’ actions and then develop a product around it.

“At Pendo we have an analysis platform and a guidance platform where we can see if users are using the app correctly,” said Budowski-Tal. Pendo is using AI to analyse and make suggestions to automatically guide users and help them complete their tasks.

The days of level four AI maturity are still some way off. Self-driving or autonomous cars have been under development for years yet the world is still not ready for mass adoption of this technology. At BMW, they define five stages of automation. Level five, known as “full automation”, where AI is the driver and people are merely passengers in the car, is still in the testing phase. In the not too distant future, robo-taxis will become a common feature of the city.

New ways of thinking

Strict regulations will be needed to support level four AI maturity. Equally, the algorithms in AI models will require significant accuracy in their decision making to foster collective trust. While today it is okay for the AI to be wrong occasionally, as the human can step in and make adjustments or further enhance the information being provided, in a level four world the risk of failure will need to be close to zero. Otherwise, the consequences could be disastrous.

Companies who embrace a high rate of experimentation in their product design are likely to be tomorrow’s winners, as they set about collecting feedback from customers and using AI to add further value to the customer experience.

Indeed, as more AI models are created, unknown use cases will manifest. This will inevitably lead to new ways of thinking. And new ways to create value.