Working (Safely) With Wearables
What problem is VIT trying to solve and how is it solving that problem?
Yeah. So for us at VIT, we’re looking to eliminate injuries in the workplace before they happen using predictive analytics. One of the largest problems our clients face is that besides payroll, workers comp is the second-highest cost. And with current best practices, their employees are still getting injured. So we’ve developed this suite of products to monitor employee posture throughout the day and prevent people from staying in unsafe postures for prolonged periods of time. We can then take this data to intervene and provide posture and lifting coaching.
And how does your product do this at a high level?
So we first talked to a lot of ergonomists in the industry and noticed that there are these telltale signs of bad posture including poor form and improper lifting mechanics. So we looked at these two metrics, the frequency of lifting and quality of form, and integrated them to develop a risk model that’s deployed on our wearable device.
Interesting, so how did you come upon this problem? Did you know someone in the workforce that had back injuries?
No, we had no notion of lifting mechanics in the workforce, it really just started with Connor Young (Andy’s co-founder) and me thinking about how interesting it would be to use wearables in an athletic setting to track workouts and prevent injuries. After our sophomore year of college at Carnegie Mellon, we were accepted into an accelerator in Pittsburgh. Later, after we did some market discovery, we noticed there was this huge untapped market in lifting intensive settings to address workplace injuries that really hadn’t been addressed using technology. And then off we went.
How has your product evolved over time?
Initially, the MVP was a lot of Arduino boards, so no custom hardware just completely off the shelf parts, and the software was pre-written for us. We were just loading into our demo product to take to our pilot customers. It wasn’t only until after our first year that we decided that we needed to make a custom PCB board. The off the shelf boards at the time were not working at the level of performance we needed. The boards themselves were actually running out of memory as our program was getting so big that the Arduino flash memory couldn’t handle the logic. That’s when we decided to build our own custom hardware.
So I’m assuming that as your product has changed the amount of information your product is absorbing into your model has changed as well, how has that data changed over time, what information are you gathering now that you weren’t then?
So we initially just looked at the four signs of bad lifting mechanics, which are bending forward, bending side to side, twisting, and prolonged posture. Those are our bread and butter metrics, but now we’re including survey data from our users. So at the end of a shift, a user can actually give us direct feedback. We also are now sending information to our users in the form of the top three tips they individually could use to improve their lifting posture.
I know you mentioned that you’re using a model on the back end. I’m assuming that’s probably running domestically on the device, not in the cloud.
Yeah. So we worked for years to determine what happens in the cloud and what happens on the device. We need to do the edge computing on the device because it needs to be in real-time so the device can let the user know if they’re in a bad position. At the same time, we don’t want to overload our device and so we let the back end take care of some of the computational heavy lifting. It was an interesting discussion that took place over the duration of our product lifecycle really. At this point, we’ve figured out what the minimum amount of computation that needs to happen on the device to detect bad posture and alert the user.
Can you give us a sense of what the model is looking for and how it’s interpreting the data?
So essentially, our device uses accelerometers and gyroscopes, very similar to your phone or any wearable device. We take that data and merge it using sensor fusion. And then the algorithm that’s onboard our device interprets that data and looks for forward bending, side to side bending, twisting, and prolonged posture. This algorithm will look for values that are too high or too low and will trigger a flag, that flag is time-stamped and will also note the severity and type of movement.
When you think about severity, how did you establish your baseline? Did you bring in a pro-athlete and get them to lift heavy things for you?
We worked very closely with different ergonomists and with their feedback came up with a movement model that we deemed low risk and another model for high risk. We’re now able to update that model over the air if we want to make any changes.
What are a few of the challenges when it comes to collecting the data from your users?
User error is a big issue. Simplifying the workflow to be as simple as possible is also a challenge. We had users miscalibrating the device so we had to revisit how we thought about the calibration process. However, even with that users are moving around and sometimes are wearing more clothing which can affect its positioning and the data output. We tried testing different form factors but also worked on making sure the algorithm was more intelligent and understood different types of user behavior
How do you ensure your clients act on the data you deliver to them?
Good question. It’s all about finding the right partners, there’s only so much as a vendor you can do. It’s really up to your clients if they use the tools you give them. We happened to get lucky in getting clients that have rockstar safety teams. They were also the clients that held our hands and pushed us towards solving the problems that would let us scale to larger contracts. And I think that’s the relationship you need from a client, you need them to be actively involved if they’re going to take action.
How do you track the value you deliver to your clients?
Right now, a lot of the value tracking is on the client-side. We ask our clients for reporting data and we use that information to model our own numbers based on the behavior we’ve seen from the users. Ultimately, we’re trying to build out our platform so we can integrate into our client’s H.R. and worker’s comp systems.
How do you communicate the value of your system to investors and potential clients?
Our platform gives clients the ability to track and log how safe their workforce is, which informs how they train their employees and allocate their resources.
Shifting gears, given data is very intangible, what advice do you have for startups or students, on communicating the value of a data-based solution to another party?
I think what you want to focus on is actually not the data itself but the value it can deliver to a client. For clients that are not data-savvy, they probably don’t care about the data or maybe they value it differently. They might not even know they’re data is worth anything until you tell them it is. So I think it’s important to tell the client why the data is meaningful and how they could access that value. Most clients don’t realize that, so just verbalizing that can help them arrive at their “Ah-hah” moment. Also, it’s important to carry as much of the burden as possible and let the other party know they don’t need to do anything. Do the hard work for them.
What excites you about the future of VIT?
So we’re getting to the point where our clients are now using the safety score that we developed as part of their KPI daily dashboard. So they now actually report these scores to their managers as a metric in addition to lost days and other metrics which is very exciting to see!
Lastly, what’s a data myth you’ve had to dispel when talking with a client?
That our product will solve all of their safety problems, our product can be part of a larger solution but it won’t fix everything on its own.
