“How can we help you today?”
“Please select from the following options…”
We’ve all come across prompts like these when trying to get in touch with our doctors. They’re typically followed by a long list of categories or a minutes-long preamble about which numbers to push for a specific outcome. In most cases, the provided intent doesn’t actually match the need — we just select anything to start speaking to someone.
Needless to say, it’s a frustrating experience for patients and medical staff alike. This is why Klara is tackling this problem with artificial intelligence.
“We’re building a model to understand what patients want and to enable fast, easy communication between patients and care teams,” said Simon Bolz, Klara’s Co-founder and Co-CEO.
The model will be incorporated into Klara Assistant, a solution that automates routine patient outreach.
Alexander Bibighaus, Klara’s VP of Product and Engineering, continued: “We want to help practices understand what a patient is talking about right from their first message so we can automatically categorize their intent, assign it to the appropriate practitioner, and get the patient a quick response from the right person.”
Patient communications come from a variety of channels — text, phone calls, voicemails, web chats, etc. Triaging all this communication is a real burden on practice staff who are also juggling a host of other duties. As a result, patients often have to wait longer than they’d like for a response.
Some medical practices already make use of automation to handle some of this triaging, but in most cases they rely on blunt keyword detection. For example, the system might route all messages that mention “bill” or “invoice” to the billing team. This relatively simple approach usually means most messages still need to be triaged manually.
“To meet patient and practitioner expectations, we really need a much more sophisticated understanding of the patient’s messages — and we can achieve that using machine learning and other AI techniques,” said Near Privman, Google AI Resident at Klara who is building the AI model (and currently hiring engineers for the ML team — more on that later).
Klara’s AI model will take keywords a step further to also understand different types of language and be able to piece together different messages to figure out intent. This neural network approach simulates how the human brain analyzes and processes information, solves problems, and improves itself through self learning.
Bibighaus gives a great example of why keywords alone aren’t the answer: “Think about addresses. There are so many ways to type out an address, from abbreviations to even misspellings. Keyword searches wouldn’t be able to handle such variations, but machine learning would.”
Ultimately, Klara’s model will read and understand words, sentences and grammar to understand what a patient needs, route the message to the right party, and drastically reduce response times. Plus, it’ll have no problem making sense of any variations or misspellings in a patient’s communication.
But that’s not all. “The long-term vision for our AI assistant goes beyond predicting patient intent and routing messages,” said Privman. “We want the Klara Assistant to understand many aspects of patient-staff communication, as well as internal staff-to-staff communication, so that it can be helpful in other ways. For example, the Assistant could suggest convenient self-service options like our appointment scheduler, minimize unnecessary back-and-forth, and ensure that needed patient information is collected in advance of appointments.
According to Dan Malinovsky, Senior Product Manager at Klara, the use of AI will significantly improve staff workflows and the patient experience.
He said, “All of our solutions are designed to maximize practice efficiency and optimize the patient experience. Klara Assistant has already helped medical practices save time on routine patient outreach, and the built-in AI model will be able to do even more. Staff will save more time, patients will wait less, and practices will ultimately grow faster.”
Here’s what you can expect:
“Patients will no longer have to wait so long for a response or feel like they need to follow up numerous times or via multiple channels to get their question answered,” said Malinovsky. “For example, if a patient texts their provider, they wouldn’t have to wait two hours until their text is read, triaged, and responded to — all while the patient is trying to call, email, etc. to get what they need.”
Instead, Klara will securely read the original text, figure out the intent, and send it to the right person on staff. The patient will then get a direct response from that person. This goes for all communication channels as well — not just SMS.
Other patient benefits include:
“When we first took on this project, we knew we needed to build out the team,” said Bibighaus. “As a business initiative and part of Gradient’s Ventures Google Residence Program, we brought in a Google engineer and gave our team an opportunity to dig into AI and machine learning and develop new skills.”
The AI team at Klara, led by Privman on the engineering side and Malinovsky on the product side, is busy building and testing this first-of-its-kind model for patient communication, which will improve in performance over time.
The team is currently hiring machine learning engineers and will be rolling out phase one sometime in Q3 2021. If you’re interested in getting this solution for your practice, stay tuned to our blog and social channels for launch updates and more.