Addressing the Elephant in the Room: Building AI Capability from the Ground Up
by John Hurt August 2019
In magazine articles and summits, the equipment leasing and finance industry has begun to move beyond a discussion of emerging technologies and onto practical implementation. John Hurt interviews Fusemachines founder Sameer Maskey and examines how companies can take on what can often seem like a daunting task when they decide to integrate new technologies into their platforms.
John Hurt, Director, Business Advisory Practice, The Alta Group
In magazine articles and summits, such as the upcoming ELFA Operations and Technology Conference, the equipment leasing and finance industry has begun to move beyond a discussion of emerging technologies — such as blockchain and artificial intelligence (AI) — toward concrete examples of how companies in the industry can use this technology to devise strategies to kick start a digital transformation.
To better understand those opportunities, I spoke with Dr. Sameer Maskey, who holds a Ph.D. in Computer Science from Columbia University and is the founder and CEO of Fusemachines. He has more than 18 years of experience working with and developing technologies in the areas of AI, machine learning (ML), natural language processing (NLP) and data science.
Fusemachines is a company with more than 250 employees whose mission is to “make artificial intelligence accessible to everyone through education, software and services.” The concept of AI-as-a-service is what I find most intriguing and what we spent most of our time discussing. One of the impediments to AI/ML projects is the expense and difficulty of obtaining the talent. The prospect of “renting” a data scientist to explore the possibilities in one’s own organization can make that initial step much easier to make.
How Do You Eat an Elephant? One Bite at a Time
The initial challenge of adopting AI can often just be getting started, as the prospect of implementing an AI platform can seem overwhelming from a non-tech perspective. Leasing companies are typically either a small independent or a small portion of a bank or captive, so such projects are typically modest (by necessity) in comparison to a corporate mandate, and getting approval for a major initiative can be challenging.
Fusemachines’ AI-as-a-service allows customers to “rent” a data scientist to guide them through the discovery phase. These modest, proof-of-concept projects are often a great way of staying off the corporate radar while testing the waters and getting results while developing a larger initiative.
I asked Maskey, “How does an engagement start, and how sophisticated does a client need to be to take advantage of AI?”
He responded, “Once an opportunity is identified, we do a meet and greet over the phone and try to understand the problem. Then in the first meeting, we bring one of our Ph.D.s to come and walk through potential solutions when the problem is being described. That gives the client an overview of possible projects, the low hanging fruits, and how much it’s going to cost. If that leads to a follow up, then we have a more technical meeting.”
That leads to the next question: how technical does the client need to be?
The answer? “It depends.” If there are a lot of technical people on the client side already, then Fusemachines can fairly quickly deploy a team, knowing the client can feed them data.
“I think for us, more than the machine learning sophistication, it’s about knowing how to extract data from the different databases and having one person who knows how to give us the data,” said Maskey. “Without data, you can’t do much. We have worked with clients who already have 10 people on a machine learning team, and they want more senior talent. We have also worked with clients who have had no machine learning team, and they wanted the whole team set up.”
We went on to discuss how well defined a customer’s project has to be to get started.
According to Maskey, “Some people come with a semi-developed idea or a general theme, such as wanting to sell more or improving customer service by some amount. They may also come with a concrete problem: ‘If we have the CRM data and historical transaction data on 40,000 people who have bought from us before, and also have 80,000 potential new customers, can we predict when people are probably about to buy?’ We problem-solve with them.”
The power of AI lies in both its ability to process vast amounts of data and to identify relationships that are not immediately apparent to humans. Early adopters of the technology in the equipment finance space have often been involved in support of risk/credit analysis. Sales and asset management in particular are great candidates for the use of AI given the complexity of the decision-making process and the amount of data involved. Additionally, unrealized relationships between data points is an area where AI can reveal opportunities. Maskey discussed discovering these unexpected relationships.
“That’s what deep learning is,” he said. “Think about it as looking at all possible combinations of relationships that are impossible for a human to think about. Having a machine to do it gives more weight to relationships that exist but humans might miss.”
The greatest challenge to date is how to leverage all available data. The data about a typical lease is comprised of both structured and unstructured data. The structured data is contained in database fields, like “customer name,” “equipment brand” and “cost.” Unstructured data is all the other information often contained in contract documents. I asked Maskey how Fusemachines addresses this and was impressed with how much the capabilities have advanced.
“Processing unstructured data is the power of machine learning,” he said. “From a computer perspective, the kind of unstructured data that would go into the model could be the description of the equipment or the description of conversations that got transcribed and stored in a CRM. Machines are pretty good at making sense of documents at the level that you need for business purposes, good enough to do various kinds of categorization.
“Being able to process unstructured data is actually one of the fundamental shifts in the power of machine learning. The ability to process text, images and videos so that the machine can make sense of them and then being able to throw it into a machine learning model for better prediction is very powerful.”
A Second Helping of Elephant
The possibilities of ML are boundless given the accessibility to data we currently enjoy. Returning to asset management, Maskey talked about how AI/ML could leverage other data sources, such as smart assets using the internet of things to collect real-time usage data. He also discussed external data sources like auction results for pricing or news feeds for regulatory changes, such as the change in diesel sulphur regulation and how it impacted the residual market overseas for used equipment.
On this subject, Maskey noted, “One good example of that application of AI is the company called Council that SMP bought. Council does something similar, not exactly for equipment financing, but predicting how the stock market might move based on geopolitical development and so forth. They do simulations to figure out what the risk is going to be on a particular thing that they are about to buy.
“The problem [in the equipment sector] is similar on some level except you’re not trying to predict stock prices. You’re trying to predict the resale value of a particular kind of equipment. So, you could do something similar where you take historical data such as what happened to a tractor that has these kinds of features and its change in value when this kind of situation appeared.”
It doesn’t take much imagination to see how such a platform could be used by an asset management group or syndication team to develop a force multiplier to their expensive human assets. This requires investment from the users of the platform. Knowledge transfer about the company’s business is critical for data scientists to accurately model any human decisions.
Maskey added, “When I was teaching a data science and entrepreneurship class, I wrote an article called ‘Three-Headed Data Scientist’ about finding one person who knows machine learning, data science and business. One way to address that is by having training programs where you enable the business people to talk the data science language and have the data science people talk the business language. Then bring them together so they learn each other’s lingo. The most successful data science teams don’t only have data scientists. They have business people working with data science and machine learning engineers very closely.”
The applications of AI are boundless and present a significant opportunity for equipment finance companies to explore how to take the first step toward developing their own digital capabilities. •
Vice President of Financial Services,
Corcentric Capital Equipment Solutions
The first step in developing a long-term equipment financing strategy is to identify all of the fixed and variable costs associated with operating your current fleet. Patrick Gaskins of Corcentric recommends developing a spend analysis to identify current and future potential purchases.
Fintech Gen 1 may have smoothed out processes on the back-end of equipment finance, but its successors may prove to be truly disruptive on the front-end. Patricia Voorhees examines the ways fintech may come to define and change the customer experience.