More than thirty years ago, Fred Davis developed the Technology Acceptance Model (TAM) as part of his dissertation at MIT. It’s one of the most widely cited papers in the field of technology acceptance (a.k.a. adoption). Since 1989, it’s spawned an entire field of research that extends and adds to it. What does TAM convey and how might today’s AI benefit from it?
TAM is an intuitive framework. It feels obvious yet powerful and has withstood the test of time. Davis started with a premise so simple that it’s easy to take it for granted: A person will only try, use and ultimately adopt technology if they are willing to exert some effort. And what could motivate users to expend this effort?
He outlined several variables that could motivate users, and many researchers have added to his list over the years, but these two variables are the ones that were most important: 1. Does it look easy to use? 2. Will it be useful? If the learning curve doesn’t look too steep and there’s something in it for them, a user will be inclined to adopt. Many researchers have added to this foundation over the years. For example, we’ve learned that a user’s intention can also be influenced by subjective norms.
We’re motivated to adopt new tech at work when senior leadership thinks it’s important. Perceived usefulness can also be influenced by image, as in, “Does adopting this tech make me look good?” And lastly, usefulness is high if relevance to the job is high.
TAM can be a powerful concept for an AI practitioner. It should be front-of-mind when embedding AI in an existing tool or process and when developing an AI-first product, as in, one that’s been designed with AI at the center of its functionality from the start. (Think Netflix.) Furthermore, AI can be used to drive adoption by levering TAM principles that increase user motivation.
Making AI more adoptable
With the proliferation of AI in sales organizations, AI algorithms are increasingly embedded in tools and processes leveraged by sales representatives and sales managers. Adding decision engines to assist sales representatives is becoming increasingly common. A sales organization may embed models that help determine a customer’s propensity to buy or churn, recommend next best actions or communications and more. The problem is, many of these initiatives don’t work because of a lack of adoption.
TAM can help us design these initiatives more carefully, so that we maximize the chances of acceptance. For example, if these models surface recommendations and results that fit seamlessly into reps’ tools and processes, they would perceive them as easy to use.
And if the models make recommendations that help a sales person land a new customer, prevent one from leaving and help them upsell or cross-sell when appropriate, reps would perceive them as useful. In other words, if the AI meets employees where they are and offers timely, beneficial support, adoption becomes a no-brainer.
We also see many new products and services that are AI first. For these solutions, if perceived ease of use or perceived usefulness are not high, there would be no adoption. Consider a bank implementing a tech-enabled solution like mobile check deposits. This service depends on customers having a trouble-free experience.
The Newark airport’s global entry system uses facial recognition to scan international flyers’ faces. It’s voluntary, and the experience is fantastic. The kiosk recognizes my face, and a ticket is printed for me to take to the immigration officer. Personally, I find this AI-first process a better experience than the previous system that depended on fingerprints, and now I will always opt for the new one.
Using AI to drive adoption
And perhaps counter intuitively, what if AI was used to drive elements of TAM within existing technology? Can AI impact perceived usefulness? Can AI impact perceived ease of use? Consider CRM. It has been improved and refined over the years and is in use within most sales organizations, yet the level of dissatisfaction with CRM is high and adoption remains a challenge.
How can AI help? A machine learning algorithm that uses location services can recommend that a rep visit a nearby customer, increasing the perceived usefulness of their CRM solution. Intelligent process automation can also help reps see relevant information from a contracting database as information on renewals are being entered. Bots can engage customers on behalf of the representatives to serve up more qualified leads. The possibilities are numerous. All these AI features are designed to ensure that CRM lives up to its promise as a source of value to the sales representative.
Outside of sales, consider patients. In the past few years, many new technologies have been introduced to help diabetics. Adoption of this technology is critical to self-management, and self-management is critical to treating the disease. For any new technology in this space, patients need to see that it’s useful to them.
AI can play a role in gathering information such as glucose levels, activity and food intake and make recommendations on insulin dosing or caloric intake. Such information gathering could go a long way toward reducing the fatigue that diabetics feel while they make countless health and nutrition decisions throughout the day.
AI’s algorithmic nature makes it easy to forget that it’s another technology and that it can aid technology. Its novelty can convince us that everything about it is new. TAM holds up because it’s intuitive, straightforward and proven. While we boldly innovate a path forward in the world of AI, shed convention and think like a disruptor, let’s keep an eye on our history too. There’s some useful stuff in there.
Arun provides strategy and advisory services, helping clients build their analytics capabilities and leverage their data and analytics for greater commercial effectiveness. He currently works with clients on a broad range of analytics needs that span multiple industries, including technology, telecommunications, financial services, travel and transportation and healthcare. His areas of focus are AI adoption and ethics, as well as analytics organization design, capability building, AI explainability and process optimization.
Source: What AI Practitioners Could Learn From A 1989 MIT Dissertation