The Key to Success in Enterprise AI: Chain of Values
AI Masterminds
The most important business lessons that I learned as an AI solution architect working with a finance enterprise
I wrote this article to share the most important business lessons that I learned working with enterprise clients as an AI solution architect. These lessons may help some of you become more successful in delivering solutions to enterprise clients and securing business with them, especially if you are still in the negotiation process to sign a contract.
AI solution architects face many challenges when helping enterprises adopt AI. Enterprises may lack clean and structured data to start with or may not even be familiar with AI’s potential. Many of these challenges are widely known, and you may already be prepared for them. However, there is one common challenging demand across every enterprise that can open all the doors if you properly address it: an end-to-end business solution with actionable and prescriptive insights. In this article, I share lessons that I learned in this path through my experience working with a finance enterprise.
Finding a single use case for a finance enterprise is not hard, but the challenge is how to create an end-to-end business solution. Here, an end-to-end business solution refers to a sequence of use cases that sit back to back and solve a real problem. Each use case must come along with actionable and prescriptive insights to generate business value; otherwise, raw final results would not be that effective for the enterprise client. Before I forget, please don’t confuse “end-to-end business solution” with the “end-to-end AI solution,” which means a complete solution from the data prep step to the model deployment step. The business values generated in the sequence of use cases, along with actionable and prescriptive insights, is what I refer to as a “chain of values”!
— Single Use Case: Business Value with Actionable Insight
The first and foremost challenge in the AI enterprise world is to determine how to create business value. Your potential clients will not purchase your product or service if you can not show how they can gain from AI. The finance enterprise that I was working with was interested in the charge-off prediction use case as the main problem. They wanted a solution to predict the likelihood of customers' charge-offs.
When a creditor writes off a debt as a loss and decides not to go after collecting this debt, it is a charge-off case. [Investopedia]
The business value of a charge-off prediction solution was clear to the finance enterprise, so I didn’t need to convince them of that. However, they were not satisfied with a single use case, so they started expanding the project to other use cases. They also needed some actionable insights based on the AI solution to reduce the total loss rooted in the charge-off cases. For example, they wanted us to determine the most important factor for customers to get charge-off. That helps the finance enterprise to adjust its financial services and reduce the total loss due to the charge-off. We offered this through Explainable AI tools that we had: (a) feature importance, (b) partial dependence plot, and (c) subpopulation analysis. You can read more about these tools in the article below.
How to Use Explainable AI Tools
— Multiple Use Case: End-to-End Business Solution
As said above, the finance enterprise wanted to go beyond the “charge-off prediction” use case and, for example, predict “the chance of self-correction” as well. Why? Because they didn’t have enough resources to dedicate to the collection processes for all the potential cases. If customers most likely would correct their behaviors, the company prefers not to intervene. Do you think that they stopped here? No.
They told me, “Let’s say we find out whether customers would self-correct. For those who wouldn’t self-correct, how should we intervene?” So, now the question is, what is “the most effective intervention” for the collection team to have a maximum ROI? Do they only need to send emails or mail to those customers in danger of charge-off? Or does the collection team need to be involved further? These are the questions that they need to answer for their daily activities.
After many back-and-forth discussions with the client, we agreed on a list of use cases: [1] detect charge-off cases (classification), [2] estimate the amount of loss (estimation), [3] determine self-correction likelihood (prioritization), and [4] determine the most effective intervention (personalization). These use cases work perfectly in a chain and generate tangible value for the business. This was how we convinced them to sign a contract with us.
Last Words
After we defined the chain of values around the charge-off use case for the enterprise client, they agreed to sign a contract with us. This was just the starting point, though. We still had to answer many other questions, such as “Do we have access to the relevant data for each use case in the chain of values?” or “Do we have records of intervention strategies (mail solicitation, settlement offer, or litigation) in past charge-off cases?” Most finance enterprises have large structured datasets that are helpful but not necessarily enough for all the use cases.
If you are interested in this concept, you can read how the DataRobot team tackles the “loan default” use case, which is similar to the “charge-off” use case. Check out their official documentation below.
Predict the likelihood of a loan default: DataRobot docs
This article is part of a series titled AI Masterminds, which helps you think strategically in AI development. You can check other articles below.
- How to Improve Single LLM Performance with Multi-Agent Systems
- How to Create a Perfect AI Strategy
- How to Create a Perfect Data Strategy
I have also published a book that discusses the best practices in AI development. I called it a handbook to success in the AI world. If you are eager to deepen your understanding, you can find it here:
Artificial Intelligence: Unorthodox Lessons: How to Gain Insight and Build Innovative Solutions
Thanks for Reading! ❤️
If you like this post and want to support me…
The Key to Success in Enterprise AI: Chain of Values was originally published in Towards Data Science on Medium, where people are continuing the conversation by highlighting and responding to this story.
from Datascience in Towards Data Science on Medium https://ift.tt/DtbcevZ
via IFTTT
