Slowly but surely, businesses are starting to realize the importance of offering more data-driven, analytics-driven techniques in order to render more real-time decisions for their customers.
No longer is the old “set-it-and-forget-it” approach to decisioning enough.
Across various industries and use cases, the interest in data analytics and decision management as a tool for keeping pace with the competition and delivering on customer expectations is growing.
Joe DeCosmo, chief analytics officer at Enova International, shared that using predictive analytics and data to build automated decision flows is a service that many businesses want but aren’t able to build.
It’s a hard job to do right, and it can be difficult to have the infrastructure in place to do it well, he added. It also requires having the right talent and teams in place to build the right models and the infrastructure.
This is exactly where Enova has observed an increasing interest in third-party solutions.
The Rise Of Analytics As A Service
If a business doesn’t have the robustness in terms of data sets, the data scientists needed to analyze that data or even the technology infrastructure to run analytics properly and perform real-time decisioning in house, then they may be better off leveraging the scale of a platform that does.
As DeCosmo explained, it’s hard to have all of the pieces needed to do those things well, which is where analytics as a service (AaaS) comes in.
Many companies have good analytics and good data scientists that can analyze the company’s data to build good models, but they lack the technology and technical infrastructure to be able to turn those into automated decision flows, he noted.
In these cases, the company has the analytics talent but not the IT or infrastructure capabilities to make the most out of the analytics they have.
And there are some companies that are lacking on all of those fronts.
“With technology and analytics being our core DNA and our heritage, we provide solutions that help companies move faster in that respect when it comes to model-driven, automated decisioning,” DeCosmo said.
Whether it's decisioning models used for fraud, credit risk, customer operations or marketing, AaaS can be a powerful tool in helping a company address whatever it is trying to solve.
The “Buy Vs. Build” Decision
Though leveraging a third-party solution sounds like a good option, companies need to know that the AaaS solution they are getting will actually create a decisioning model that is both helpful and meaningful in the automated decisioning they are looking to deploy.
Understanding whether to build in-house or buy is significant decision for any company, but especially when it comes to decision analytics and management.
DeCosmo explained that Enova’s approach is to always recommend that a company work backward in order to first determine the outcome it hopes to gain in the particular decision it is solving for.
If it’s solving for credit risk or underwriting, he noted, then the outcome is likely seeking to have a better credit decision with customers, ideally one that is done in an objective, automated and real-time way as opposed to a manual, slow approach.
An institution or business must first know the outcome it wants and then think about all the data it will need in order to render that decision. DeCosmo also pointed out the importance of understanding any existing rules or policies that will also go into the decisioning model and could possible impact the outcome.
With a clear understanding of those policies, a company can then look at the decision flow and identify where a model in automation would improve the outcome.
“Once you have a good understanding of those three pieces, then you can really understand the value of automating that process in either a platform you build or a platform you buy from a third party,” DeCosmo explained.
But he warned that this process is also one where many people can get ahead of themselves by jumping right to wanting a new methodology or decisioning platform without taking the time to think through the entire decision flow to understand what they need to either build or buy.
Cutting Out Decisioning Bias
Last month at PYMNTS Innovation Project, author and Harvard PhD mathematician Cathy O’Neil spoke to the audience about how the biases that can be built into models can lead to a very dangerous situation when the algorithms become WMDs: weapons of math destruction.
According to O’Neil, an algorithm goes full WMD when it has widespread (used often), mysterious (poorly understood outside of technician circles) and destructive (do recordable damage to people’s lives) results.
While these biases are not intentional in most cases, because they reflect the state of an organization and a desired outcome, the usage of these algorithms can perpetuate decision making that may not always be in the best interests of consumers or businesses and, in fact, sometimes not even the institution itself.
DeCosmo said that this is a very valid concern but that there are things that can be done on the front end through sampling methodologies and other approaches to make sure a company is not introducing bias into models when they are built.
On the back end, he noted, there can also be good controls put in place to make sure that the result doesn’t start to generate some unintended bias as well.
The biggest takeaway is that these models need tests and controls, on both the front and back end, to ensure the decisioning is balanced. As models become more machine learning–driven and automated, the kind of tests performed on a scheduled basis now will need to be done more frequently, he added.
“You need to be monitoring your models to make sure that your model is not picking up some underlying bias that is now being institutionalized into the model through a machine learning algorithm,” DeCosmo explained.
By placing good monitoring and controls in the process and ensuring they are taking place on an ongoing basis, companies can be confident that they are using a well-built model that makes a decision, takes it out of a subjective process and makes it more of an objective decision.
This also enables more controls over that process and the ability for a company or a third-party provider to make sure the model is built in a way that doesn’t incorporate bias at the beginning and then doesn’t begin to be biased over time.
“With algorithmic-driven decisioning, you still get better, more objective and more consistent decisions, but they take work,” DeCosmo emphasized.
With the need for consistent and constant monitoring of the model and its outputs, there’s no longer any room for putting a model in place and forgetting it, he added.
What’s Next In AaaS?
Whether that decision flow is around marketing, credit risk, customer operations and retention, etc., he said the AaaS can be used to optimize those decisions.
The company has also found a need for this service across all types of industries, including banking, telecommunications and insurance.
DeCosmo noted that the insurance industry in particular in being disrupted by a number of startups on both the brokerage and carrier side that are doing interesting things online. Traditional providers as well as new entrants are looking for ways to be more efficient and automated with the way they interact with customers.
“The analytics and the data is just one piece of it — the decision flow and the hosting of that decision flow to make it available through APIs and render an automated real-time decision is really key,” he explained.