Beating the Market: How to Evaluate AI
Over the course of the past month, we’ve shared a series of blog posts telling you everything you need to know about leveraging AI technology for finance and risk. Missed it? Start at the beginning of the series and then come on back.
Today we conclude the series with some practical advice from several leaders on our team with decades of experience in the AI industry on how to evaluate and select the right AI solution for your business and needs.
Do Your Research
Carl Hoffman, Basis Technology CEO
“Evaluating commercial AI products is no different than evaluating any other complex software product. Read reviews, check references, and study applications. Also, you want to make very sure that their is a clear mapping between what the product does and the business decisions you have to make. Always ask yourself, ‘where will this help me make a decision?’”
Have a Viable Practice Dataset and Test Scenario
Declan Trezise, Basis Technology Director Pre-Sales Engineering
“A vendor’s demo data is not the same thing as field data. Ideally, you want to evaluate a vendor’s technology by having it perform a well-defined task using a realistic dataset for which you know the results. If you’re set up this way, you can ensure a quality evaluation.”
Set Clear, Reasonable Performance Goals
Steve Cohen, Basis Technology COO
“Perfection is often not a useful standard. Your performance goals should be based on actual alternative—not a hypothetical one. If you’re comparing an AI solution to human analysis, make sure you have reasonable benchmarks human performance. If there are no available standards around a given task, the next best thing is to see if the vendor’s solution can improve over successive trials.”
Make Sure They Can Explain It
Chris Biow, Basis Technology SVP
“The first thing I do when evaluating an AI technology vendor is ask them to walk me through their solution step by step. I need to know what they do, how they do it, why they believe their approach works, and what they are training their algorithm on. I record their claims, then I check to see if those claims are substantiated. For example, one of the problems with applying AI to low-frequency risks is that there may be very few historical “positive hits” on which to train your models. If a vendor dodges my questions around training data—or provides vague replies—warning bells go off.”
Make Sure You Can Explain It
Chris Mack, Basis Technology VP Product
“No matter how well a product performs, its explainability to regulators is paramount. Make sure the vendor can provide you with a clear compliance strategy, because the last thing you want is to buy great technology you can’t use…or get fined for.”
Put it to use
Despite the hype cycles of enthusiasm and disillusionment throughout the history of AI technology, there are several powerful applications of AI to the financial industry that your organization can utilize today, allowing you to better understand your customers and business partners, reduce AML costs, and detect fraud sooner.
You’re now armed with the necessary expertise both to find and evaluate AI solutions, as well as leverage them to improve your business’ risk and compliance processes. Ready to get started? Talk to our team to discuss your business’ needs and pain points.