Artificial intelligence

Fast and easy implementation of AI medical coding: How we do it

When talking with potential clients, we often see negative preconceived notions surrounding the implementation process for AI medical coding, usually stemming from disappointing or stressful past experiences with other solutions. 

We get it; there's a lot of overpromising and underdelivering within the AI healthcare space. Naturally, this leads to skepticism or distrust, even if they recognize the importance and potential power of using AI and automation.

While many see the value in utilizing our AI automated medical coding solutions, they're hesitant to buy in completely because they believe the implementation process will require large amounts of time and resources. 

We're here to tell you that's a myth. 

Does implementation require a heavy lift from IT and coding teams?

Not at all. 

Many worry that the setup process will solely rest on the shoulders of internal coding and IT staff—a valid concern seeing as though these teams are usually already stretched thin. However, that’s not the case.

We do all of the model development and most of the implementation on our end, aiming to put as much work as possible on our plate instead of asking our clients to carry the heavy lift. To get started, all we require is minimal information from the coding and IT teams.

What we ask from the coding team is their ideal coding workflow and an understanding of their rules. It’s as if you just hired a new coder named Fathom—what would you tell Fathom on day one of the job? Getting a solid idea of their goals and rules is essential because, in the end, we build the coding model based on their requirements. 

On the IT side, we have two asks. 

  1. The first is sample data. We evaluate this data to help inform and train the model we're creating for their organization. Any data format is ok. We don't need adjustments or customization of the data they provide because we don't want our clients doing any additional work. The more sample data the IT team can provide, the better the model performance will be. High sample data volumes allow the model to become even more familiar with coding guidelines and unique data templates to further boost accuracy and automation rates. 

  2. The second ask comes once we build the new coding model. After implementation, we want to ensure that results flow into your system properly. To accomplish this, we ask for whatever is the most convenient file transfer method for our client—usually an SSH file transfer protocol (SFTP) dump of a CSV file, or we can leverage HL7 FHIR to access the data.

This process ensures that we do all the heavy lifting when it comes to setup, not our clients' internal coding and IT teams.  

How long does setup take?

Anywhere from three weeks to three months, depending on a few factors. 

One is the complexity of a client's coding guidelines. Some organizations follow standard CMS rules, which are already baked into our model. If that's the case, there's no work for us or the client. What makes the process longer is adopting custom coding guidelines.

Another factor that lengthens the setup process is highly customized workflows. If a lot of data needs to be put into multiple systems once we've completed an organization’s coding, we need to work with their IT team to ensure our output is mapped correctly within their system. 

From there, we go through iterations where we test data back and forth to optimize and finalize this process, which will never take more than a few months across all of your locations. All of our clients are ROI positive with high levels of automation from day one after going live.

How can we achieve such a quick turnaround?

The driving force behind our quick implementation process is our unique technology.

Other coding automation vendors mainly use a rule-based machine learning AI engine. So while their GPU or processor completes the automation, analysts are still manually building rules into the system to account for the variance in data format and coding guidelines. These models tend to break very easily.

We don’t have hundreds of analysts working around the clock—our deep learning AI model is doing all the work. It excels at taking unstructured data across multiple formats with many variances and interpreting it the same way a human would. It can do so because we’ve trained it on more than 400 million coding encounters.

We deliver on our promises 

We don’t overpromise and underdeliver. When we say we can get our AI automated coding solution up and running across all of your locations within weeks, we mean it. 

Still hesitant? Have any questions? Want to see how our technology can work for your organization? Let's talk.