Most of us have heard of the snake oil salesmen from the Gold Rush era. These salesmen, after seeing the rise in popularity of Chinese water-snake oil products, wanted to jump on the lucrative bandwagon. The only problem was, there were no Chinese water-snakes in the U.S.
In light of this problem, snake oil salesmen began passing off a combination of red pepper, beef fat, and turpentine as "rattlesnake oil" - although the "elixir" contained no snake oil of any kind. These snake oil salesmen were eventually discovered after many unhappy customers revealed the so-called elixirs had no healing properties. The entire event turned the description of a "snake oil salesmen" into a negative characterization of those who offer phony products.
Well as Mark Twain said, "History doesn't repeat itself, but it often rhymes."
"Artificial Intelligence" has been the biggest buzz word in the healthcare industry for the past year, leading companies to adopt this phrase lightly in order to get attention from the hype. Among all the noise, many companies are claiming to offer "solutions" to your biggest problem using AI. But, be aware of those that are trying to sell you technology that over promises, but under delivers.
You may be asking yourself: "How do I know which companies are really using pragmatic AI to solve healthcare back office problems?"
Well, there are a few key questions you should ask a potential AI partner. I've highlighted a few below, and you can download our full checklist HERE.
- "What kind of model does your solution use?" They key word here is "model"–whether the company uses a neutral network that transcends various data sources, or a single machine learning model focused on one source and one use case, you want to hear the them tell you that the model uses a cognitive feedback loop so that it is constantly learning over time.
For example, if you're talking to a prior authorization solutions company and they're describing a rule set that guides their auth submission criteria, you'll want to ask how frequently that rule set is updated and scrutinized over time. The reason? As everybody in the prior auth space knows, the rules that payers employ to review cases can change quickly and often—and any model worth its weight should adapt to those changes over time. (A static model that lacks a feedback loop is incapable of adapting... so that solution grows less effective the longer you use it!)
With a feedback loop, every subsequent time a task is completed, the tech is taking the what has been learned and applying it to the next task. This also enables your technology to appreciate in value as it grows smarter over time.
- "What percentage of the workload for [task X] do you fully automate?" A predictive and feedback-based model isn't enough, however, any A.I. solution must be just that i.e. an actionable solution to a clear problem. Plainly stated, the tech needs to actually do something. No brainer, right? But there are a number of A.I. solutions that can make predictions or send out alerts, but they lack the muscle and the interoperability to drive an action or outcome. You want to be sure that you're buying an actionable asset, something that can relieve burdens from your team. So ask the question: how much work will you take off my plate?
(Caution: if an A.I. company tells you "we take 100% of the work off your plate", walk away. A.I.-enabled tools have the potential to be highly valuable augmentation assets for your team, but there is no silver bullet).
Click here to learn about the differences between AI, Machine Learning, and automation in greater detail.
- "What kind of ROI do you guarantee? You don't just want to hear the blanket "we'll deliver you 400% ROI"... you want to hear "we deliver 400-500% ROI, based [criteria X], [criteria Y], and [criteria Z], and that's proven out across 3 customers like you." If you're reading this, you've bought software or services before. You know how to scrutinize a contract and a commitment. A.I. solutions are no different. You should see a measurable ROI, it should be based on the metrics that you care about—in the case of prior authorizations, metrics like labor savings, auth rate improvement, etc—and it should be confirmed based on experience.
- "What problem does this solve? One AI solution is not going to transform your entire revenue cycle successfully, no matter what you hear. You know the amount of variability across of the processes included in your rev cycle. However, pragmatic and narrow approaches go in to the necessary detail and data required to transform specific areas within the revenue cycle. Make sure what you're reviewing is appropriately narrow (i.e. focused on "coding" or "prior auth" or "eligibility", not "automating the entire revenue cycle").
- "Who are the people before the product?" Dig into who makes up the team behind the marketing. Are there data scientists and developers behind the scenes? Has the team built products like this before? Speaking on panels is one thing—building products and solving problems is another. Don't settle for "oh, we have a data science team back home"—get names, confirm bios, and use your internal professional radar to gut-check if you feel comfortable with the team behind the AI.
AI is a hot topic. You may come across legitimate companies offering useful, automated AI solutions for specific use cases. However, you may also be approached by companies selling AI snake oil. Think about these five questions the next time you speak with a potential AI partner.
Better yet, download our checklist so you don't forget the essential questions to ask.