Updated: Apr 28, 2020
Actually, it's "PAS," not "pass" - and no, your company should not be passing on Artificial Intelligence. What's PAS, you ask?
Proving It. What's the Business Case for AI?
Unless you are doing pure R&D, most technology initiative require clear ROI or strategic business alignment to get funding.
The challenge with AI?
Enterprise stakeholders often times do not have enough experience with AI to understand the paradigm shift it brings to their operation. This makes proving cost savings and revenue opportunities far more challenging. Educate your stakeholders on the benefits, risks, and impact to the organization.
What is the best way to select a business process you want to augment with Artificial Intelligence? Here are a few Do's and Don'ts:
Select a mature business process with some tolerance for errors
Make sure data is well-defined and supports the process
Have a defined metrics in place measuring performance and failures
Select a business process at least partially automated
Make sure there is room for improvement in performance characteristics of the process
Attempt to fully automate a manual process with AI for the first time
Attempt an AI implementation or integration without the right skill set to support it
Try to implement AI using bad, incomplete or unorganized data
Try to "fix" a poorly modeled process with AI
Managing feedback and risk of failure will be easier for an internal process; however, external processes may offer a broader training data set to improve your probability for success. Don't be afraid to get started by looking at your customer interaction and customer service processes. These can be great candidates for AI augmentation.
The goal is to set yourself up for success by picking a business process that is already enhanced by technology and easily measured at different points. We want to create a more intelligent process between people and technology.
Adoption: The Collective Intelligence
The AI strategy must result in improvements in the human-computer interaction and net gains in total output, efficiency and satisfaction.
Your staff should not be forced to make significant changes in business process behavior for the AI. Even if tempting, don't reinvent the process during an AI project. If the process itself is flawed, then improving the process is a prerequisite for adopting AI.
It's the augmented intelligence between the people and the machines that will make for a more competitive and efficient business process.
Adopting AI means that your people, process and the technology are working together for mutual benefit. Let that sink in for one moment. Mutual benefit....the technology has to benefit the people, and the machine process must benefit as well. Adopting AI is much more than simply standing up a couple of chatbots or rolling out a neural net. The machine processes must have a learning interface to improve. As the human element of the process improves, so must the artificial elements of the process. A chatbot that doesn't learn new phrases is not different than a static set of software instructions and isn't an intelligent system.
During the adoption phase, you should see:
New interfaces and methods of interaction with systems
Data repositories and structures changing and growing
Process execution generating new business insights
People understanding the value of training the AI with new data
A new view of the organizational roles aligned around the AI
Here again, your specific business case will help determine the right kinds of intelligence to add to the process. Make sure to allow for an extended "live trial" so that the system is properly tested with live data.
It's critical to make sure the feedback and training processes are working correctly. The training data is what adds the most value to the AI. Make sure people take the time to learn the enhanced processes and for some actual live training data to be fed into the system. By the end of your adoption phase, you should expect to commit regular resource hours to the training and review of the data in the cognitive repository.
Scaling the System: It's Not About Infrastructure
Traditionally we think about scaling as an increase in the capability of the system through the addition of more hardware and software. While this remains true with an AI system, scaling data and data scenarios are more critical. Scaling the baseline data (the neural net) means that the data is scrubbed, modeled, and routinely added back into the repository with human review.
While human review is essential in the early (adoption)phases of training, scaling means we have automated the feedback loop and established some level of machine learning. This means that new scenarios are actively sent into the AI as a post of how the process works. Without active feedback on false-positive scenarios and fine tuning, the learning component cannot take place and expected benefits will never be realized.
Scaling AI is a challenge! This is the phase at which management expects to see significant benefit while much training is still going on. It's important to benchmark performance and scenarios over multiple cycles of your business process. This will allow the stakeholders to validate the true performance improvement and value of the AI. Make sure to look at performance metric improvements relative to scale of the data in the AI. Limited data scenarios will always mean limited benefit.
Key success factors for scaling include:
A fully integrated human-in-the-loop training process
Automated learning (machine learning) is in place
Data scenarios increasing week over week
Predictions occur regularly with a declining number of false positives
Focus moves away from understanding and learning to growing the cognitive repository
Building an enterprise AI strategy doesn't need to be a daunting task. In fact, it should be part of every enterprise strategy going forward. PAS (Prove-Adopt-Scale) is a common-sense approach to implementing an AI strategy and can help you succeed. Throughout each step, remember the following:
Educate, Educate, Educate yourself and your stakeholders
Fix your data and processes before you implement AI
Know your goals and desired outcomes
Be ready to change and adapt
Implementing AI Technology can be a bit overwhelming to the organization, by a clear AI Strategy will reduce the risks, create clarity around the goals, and establish a vision for what AI will mean to your organization.