We have discussed the importance of having the right talent in place when it comes to AI adoption in enterprise quite thoroughly here at Emerj. The scarcity of data science talent and its price point are one of the main reasons small businesses are not likely to adopt AI successfully at this time.
That said, two types of talent are necessary for AI adoption across all industries. One is of course data science expertise: people that deeply understand the role of data in machine learning and the capabilities and applications of AI. The other, however, is domain or subject-matter expertise: people that deeply understand the business itself and the context surrounding business processes.
Through all phases of an AI initiative, data science and subject-matter experts are needed to work in tandem. Currently, it is rare to find both kinds of talent in the same person.
Of course, that may change in the next few years. Grant Wernick, CEO and co-founder of Insight Engines, a company that claims to allow security personnel to ask questions of their data, spoke to us about how AI itself is helping bridge the knowledge gap between data scientists and subject-matter experts. He claims:
We are starting to see tools that allow non-technical people to become a little bit of these wizards (data scientists). We have actually helped up-level people who have a lot of domain expertise, in this case their security guards, and they become part of the cybersecurity team in a matter of a couple weeks.
However, that situation is still mostly the exception rather than the rule, and so it is important to ensure that data scientists work closely with subject-matter experts on three critical areas when it comes to making AI work for a business:
- Determining the Business Problem an AI Solution Can Solve
- Determining How AI Can Solve That Business Problem
- Maintaining and Updating the System Once It’s Built
We’ll begin by diving first into determining the business problem:
Determining the Business Problem an AI Solution Can Solve
The role of AI in the enterprise is primarily to make processes more efficient. Data scientists often organize the data available to a business in data lakes or similar infrastructures so they might use that data to build machine learning models that drive business value. However, this is not a one-size-fits-all proposition. Each business will have unique needs and different priorities. Subject-matter experts provide that context to data scientists.
Subject-matter experts that work at a company know its real priorities and the challenges it faces. They also bring an understanding of their field, be it marketing, healthcare, or insurance. As a result, they can identify opportunities for where to apply AI to best solve a business problem or achieve a business goal. They have an idea of which processes would benefit from automation most. Perhaps most importantly, they have an idea of the business implications of implementing an AI solution.
For example, let’s say an insurance company wants to become more competitive. A high-level employee at the company, who has been in the industry for more than 20 years, may know from experience that clients often complain about the time it takes to process claims. They have the context to know that a data scientist might best spend their time for the company building a machine learning model that speeds up the claims process. They could also provide context to the data that the data scientist will be working with.
An experienced insurance employee may know general demographic information about the company’s customers and have an idea of how well customers have responded to different products and services in the past. Perhaps most importantly, the domain expert may understand which parts of the claims process are the most frustrating for customers; those are the parts that the data scientist may want to focus on the most when building the machine learning system. The employee might not have the technical knowledge to build the AI solution, but they understand aspects of the goal that need addressing.
Data scientists likely won’t know anything about how a company operates because they are on the outside looking in; they went to school for computer science, and that is where their expertise lies. They won’t understand the context for what happens in a company without the guidance of subject-matter experts at the company.
The insights provided by subject-matter experts define the business problems. This in turn provides data scientists with the context to determine the parts of the data required to solve the problem and the structure in which to organize it.
Determining How AI Can Solve That Business Problem
Data scientists also keep subject-matter experts grounded. Domain experts may have an overblown idea of the capabilities of machine learning. Most do not have any idea of what it can and cannot do. Data scientists do, and so it is their job to draw the line for subject-matter experts to manage the expectations of stakeholders.
For example, a marketer might tell a data scientist to make sure the model they build can predict the best price to bid for ad space on Google ad Network without realizing that such an application is nascent in practice and technically challenging. It might not be possible for most businesses and many data scientists. A data scientist would be able to explain this to the marketer, preventing them from promising their team and stakeholders too much.
Similarly, data scientists might point out that a company does not have enough data to meet a particular goal with machine learning. For example, subject-matter experts might want the system to optimize prices for a product, but they might not have the data on competitor prices to make it happen. The data scientist might recommend the type of data the system would need to solve that problem or the type of data it would need to proxy those prices, but they might put a brake on any expectations the domain experts have for the AI to do it in the moment.
Tim Delisle, CEO and co-founder of Datalogue, a company that helps prepare data for machine learning models, finds that often business leaders think they have the right data to build a machine learning model that will solve their business problem, but they are mistaken. He suggests that business leaders steer away from basing their data lakes on what he calls “the ground truth:”
The ground truth is overrated. There’s this underlying perception that ‘I’ve got the data to answer my business question,’ and sometimes it’s just not the case. It might be that [a business] didn’t have the processes that were recording the right data or that [a business] didn’t have the data that [it] thought [it] had. This concept of creating a data lake that contains the ground truth is just so far-flung in terms of the reality of the situation.
Data scientists can outline the bounded realities of AI to subject-matter experts. They can set expectations for the features of a machine learning model: what it can and can’t do for a business. This intricate balancing act will only be successful if both sides work together.
Maintaining and Updating the System Once It’s Built
Business leader might think that once the subject-matter experts tell the data scientists what they want, it is all up to the data scientists to build the machine learning model and maintain the system going forward. That is not the case. In fact, the role of the subject-matter experts become even more critical once the AI system is up and running. A test run of the model is necessary to make sure it delivers the desired results.
Subject-matter experts have the business context to determine if the model is effective in driving business value, whether it be reducing the time it takes to pay an insurance claim or predicting sales revenue of a certain product, and point out the area in which that value might improve. They can then relay this information to data scientists to further tweak the model.
Take for example a manufacturing company with a large number of heavy machinery to maintain. The subject-matter experts collaborate with data scientists to develop an AI system that gathers data from IoT sensors on the machines to predict the optimal time to do preventive maintenance on them.
During the test run, the data scientists check if the system is picking up on IoT data and if it is making the expected predictions. They may believe that the system is making accurate predictions on when to do maintenance on these machines, but they don’t have any context about those machines. They can’t interpret the numbers to figure out if in fact one machine should be maintenanced more often than the predictions are calling for. An employee at the manufacturing company has that context.
They may realize that traditionally certain machines require more maintenance than another, and so they might find it strange when the AI system is predicting significantly less maintenance than other machines. This might lead them to check where the sensors are placed on the machines, or they might realize that the machine learning model isn’t factoring in the machine’s fuel levels even if it really should.
The subject-matter expert would then be able to relay this information to the data scientists, and the data scientists could then tweak the machine learning model or weight certain factors over others so that it generates more accurate predictions.
Steering an existing system requires the feedback of a subject-matter expert about its effectiveness and a data scientist to use the feedback to make the required tweaks to the algorithm if the system is to drive value for a business.
Takeaways for Business Leaders
An effective machine learning system today will always need both data scientists and subject-matter experts to achieve ROI. In the future, this might not be necessary. In five to 10 years, more domain experts might also have enough data science context to speak to coders about what they want to see out of an AI system. They will likely have a better understanding of what is possible with machine learning.
Right now, however, very few subject-matter experts understand how machine learning works, and data scientists are not subject-matter experts. It will take the two experts working together to get an AI system up and running in the right direction for a business.
Header Image Credit: Britsafe