Why Artificial Intelligence (AI) Projects Succeed or Fail?
A very interesting article in the Jan-Feb 2018 Harvard Business Review (HBR) is entitled ‘Artificial Intelligence for the Real World’.
The authors studied 152 Artificial Intelligence (AI) projects and categorized these into three groups;
- Robotics and Cognitive Automation (47%)
- Cognitive Insight (38%)
- Cognitive Engagement (16%)
What resonated with us is that FlexRule’s sweet spot is in the largest group – Robotic and Cognitive Automation. We are also well on our way to addressing the second group with our latest development in Predictive Analytics, due to be released this year. We are also in discussions with cognitive engagement providers, the third group, regarding the integration of virtual assistants and chat-bots into back end systems.
The HBR article covers the reasons why Artificial Intelligence (AI) projects succeed or fail, and provides a list of challenges identified by customer executives as shown in the table below:
We thought it would be interesting to add our own real life customer experiences to these and explain how we have helped address these challenges.
1. It’s hard to integrate cognitive projects with existing processes and systems.
The second area is overcoming the problem of integration with existing processes. A common issue here is that the existing process is not always well understood or formalized. At a recent CIO meeting we explained that when implementing processes in FlexRule, the customer’s project team would often say something like, “Oh we need Jim here. He is the only one who knows how this works”. At which point the CIO nodded in agreement and said, “That would be Ian in our organisation!”
When we demonstrate FlexRule’s graphical interface for mapping processes, which then becomes the execution model, customers can see how they can use FlexRule as a collaboration whiteboard, process diagramming tool and implementation platform all in one. On seeing these capabilities, the same CIO exclaimed, “I have a BA who will love this!”
2. Technologies and expertise are too expensive.
It is a sad but true fact that technology and consultant’s pricing tends to follow the same curve as the Gartner’s Hype Cycle. At the peak of the “Hype of Expectations” we also see the peak of pricing, only for it to come back down to realistic levels later. This is because technologies have been oversold in the marketplace (see point 6 below), which typically occurs when sales reps claim that the reason for high pricing is due to the miracles the technologies can deliver. The example in the HBR article was a $62m project to diagnose cancer that was shelved without ever having been used on patients.
At FlexRule, we are trying to break this norm. Our licensing model is aligned to the business outcomes achieved. We keep it simple, understandable and in line with delivering business value. We don’t price based on multiple metrics, like the number of users, servers, processes, cases, or any other rapidly multiplying cost driver. Instead, we offer a flat fee per business area.
3. Managers don’t understand cognitive technologies and how they work
The main culprit driving this issue is that we get caught up in technology buzzword speak instead of business common sense. The fact is that we are using the Artificial Intelligence technology purely to help with our business processes. Therefore, we need to start with the business processes and not the technology. It may seem obvious, but we often have to ask customers to stop and explain the business problem they are trying to solve. In other words, what are their current processes, business logic, and decision requirements information sources, and what is the desired outcome?
We show customers how to map this using FlexRule’s graphical logic layers of Process Flow, Business Logic, Decision Logic, and Data Logic. Suddenly, they realize that the understanding they need is simply their business knowledge, of which they are the domain experts.
4. We can’t get enough people with expertise in the technology
One of the most common questions that we are asked is ‘What skills are required to implement Robotics and Cognitive Automation, and how steep is the learning curve?’ The answer is that there are three different layers of skills that are required and most exist within our customers’ organisations today.
- As mentioned above, at the top level we need business expertise on how the business processes work, as well as the business logic required to make the right decisions
- In the middle are the Business Analysts and Business Process people that are needed to map out the processes and logic layers.
- IT people are needed to implement the data and application integration layers. These are the people with DBA and .NET skills. Typically, they are needed in the upfront setup.
Regarding the learning curve, it is almost the opposite – people need what we would call an ‘unlearning curve’. What we mean is that often, especially amongst developers and technologists, we need people to avoid thinking in coding terms, algorithms and etc, but rather to think in business logic terms. We explain how to take their business logic and turn that into the steps necessary to make a decision. Once they understand that it is all about making decisions, they hit the ground running. By the way, we have found that referring to this as “making decisions” makes it far more understandable than using a buzz-phrase like “Cognitive Automation”, even though it is the same thing!
Our goal is to make our customers self-sufficient as soon as possible. This includes enabling business people to make their own changes to the business logic.
5. Technologies are immature
Many of these technologies have been around for a long time. It is just that some of them are making promises which are just not there yet. This also resonates with the point below about technologies being oversold. As the HBR article explains, projects which are going for the “moon shot” are less likely to succeed than those that focus on the immediate problems of the real world.
At FlexRule, we sometimes describe ourselves as the boring part. We are the middleware layer that pulls the apps, cloud data, processes and business logic together. While we can justifiably claim that we can be used for a “moon shot”, we try to help customers solve more down to earth issues first.
6. Technologies have been oversold in the marketplace
There is an old joke that asks; “What’s the difference between a used car salesman and a computer salesman?” The answer is that “At least the used car salesman knows when he is lying!” The point here is that everyone, and we mean everyone, gets caught up in the hype. That includes the sales people, product folks and customers alike. Just search on Artificial Intelligence (AI), Machine Learning, RPA, Robo-Advice, FinTech, Blockchain, and more, and you will get the picture.
Our advice to our customers is to bring it back to the “real-world” by asking “What is the business problem you are trying to solve?” From there we can start to address the business needs, not the technology hype.
We can help you with that, please fill the below form and we will get back to you to setup a discovery workshop and identify areas that your business can improve using Decision Centric AI.