Redesigning Pharmacy Operations with AI: Inside Horizon Pharmacy

Natalie: Thank you, everybody, for joining us today. My name is Natalie. I am a pharmacist by training and the Chief Product Officer at TJM Labs. We have Megan and Jason here from Horizon Pharmacy, who will be sharing their journey with AI. We recently had an RxInsider article come out about their experience, and we have been getting a lot of questions from pharmacies, so this is an exciting opportunity to hear directly from them. Megan, let’s start with introductions.

Megan: I am Megan McCaskill. I am a pharmacist as well. Here at Horizon Pharmacy I am the Vice President of Compliance and Strategy. Day to day, I oversee quality improvement, inventory management, training of new staff, and onboarding new businesses. I also take care of our accreditation. I have been with the company since we opened 12 years ago.

Natalie: Jason, could you share your role, and for those who are not familiar with Horizon, tell us a little about who you serve and what your operation looks like today?

Jason: Happy to. Hello, everybody. I am Jason Dana Costa, President and CEO of Horizon Pharmacy. I am not a pharmacist, so everything I have learned has been at the grace and education of my colleague Megan. I have been in this role about three and a half years. Horizon is unique. We are a social enterprise, a closed-door pharmacy owned by three community mental health centers. Every dollar of profit we generate goes back to the community to subsidize free or low-cost behavioral health and substance use services. Over 1,000 people a year get additional care beyond pharmacy care because of our organization.

To give you a sense of size, we are in Rhode Island and serve just the state. On any given day we see about 1,600 patients. Because we do a lot of substance use services, annually we might see 2,300 or 2,400 individuals across 21 or 22 partner companies, and we deliver to anywhere from 130 to 140 places in Rhode Island. We send out packaged medication four times a day. We serve behavioral health, substance use, assisted living, developmental disability, and IDD, in both adult and children’s services. Our logo is “Your pharmacy, our community,” so we really try to make sure what we do has a positive impact on the people we serve.

Natalie: That is a really unique setting and a meaningful mission. Megan, before introducing AI, where were you seeing the biggest opportunities to improve how work was getting done?

Megan: As with any pharmacy, my team, especially my data team, was having a lot of trouble keeping up with day-to-day tasks. We have a very high volume of incoming scripts, and reconciling them for the different groups, getting them ready for multi-dose or single-dose packaging, we were struggling. And God forbid someone had a sick day, or when we have two people on maternity leave like we do right now. Even when we could plan for it, anything like that threw a wrench in our day and we would fall behind. The morale was missing. And hiring is very difficult right now for certain roles, so we were struggling there too.

Natalie: That is the classic pharmacy operations story we hear from practically every pharmacy we meet. Jason, what led you to start seriously considering AI, and what were you hoping it could do for Horizon?

Jason: Like any businessperson, I am always looking for efficiency and ways to grow the organization. I had been monitoring AI for a couple of years, the same way most people do through the news and conferences. About a year and a half ago I started hearing from some pharmacies that were beginning to have conversations about automating certain tasks. Megan and the team have been very forward thinking with technology, which emboldened me to reach out to a few organizations.

What I have really been pushing since I arrived is getting everybody to the top of their scope of service. I assume many people here are owners or leaders of independent or LTC pharmacies. I wanted my team doing the most important, toughest, highest-value work, the work at the top of their license, and to let something else handle the monotonous work. So I started reaching out to several companies, not just TJM Labs, about 18 months ago. It seemed like the technology was not fully matured yet, but mature enough that our team was courageous enough to start trying it.

Natalie: Megan, when you started discussing bringing AI into prescription processing, what was your reaction, and did you have concerns?

Megan: I was very skeptical. I do not consider myself very tech-savvy, especially compared to Jason and my counterpart Justin, who runs day-to-day operations. I did not fully understand what the bot could do. I use Copilot and ChatGPT occasionally, but that was the extent of my AI journey. I was concerned for our staff. You hear all those things, like the robots are going to replace us, and I did not want my staff to feel that way. I wanted them to feel this was something that would help them, not replace them. We have been working together since March, and the opportunities of the partnership have opened up a lot more ideas than I imagined.

Natalie: That skepticism is understandable and a natural part of the journey. A lot of it comes from fear of the unknown, and webinars like this are one way to demystify the process. Jason, when you were evaluating partners, what were you looking for, and what gave you confidence that TJM Labs was the right fit?

Jason: I want to reiterate that TJM Labs was not the only provider we looked at. As CEO I am looking for something affordable and cost-effective, but honestly that was a small part. In prior organizations, and here at Horizon, the IT and service providers I value most are true partners, ones where we can influence how their product impacts our business. We come to them with a problem to solve and figure out how to get there together. We still have regular calls with TJM Labs and a couple of other providers.

So I was looking for a strong partnership, someone who understood the core value of what we are creating at Horizon, and what they are trying to build too, so we were solving the problem together. A couple of things stood out about TJM Labs, specifically the team. Natalie is a great example: she is a pharmacist. I am not, so I depend on experts who share the language pharmacists speak. Some technology companies had heard the word SIG and could say it, but it did not mean much to them. To Megan and to the TJM Labs team, it had real meaning. That shared language reduced the barriers and a lot of my team’s concerns.

I also liked their model. It is software as a service. A few organizations we spoke to would essentially build us a bot, and then it was done, it was ours, and that was it. TJM Labs treated it as something we would keep iterating on, which aligned with that idea of partnership. We hoped to influence how they design their bot, and we thought that was really important.

Natalie: TJM’s products were inspired by and built by pharmacists, so our team has deep empathy for pharmacies. We have lived the challenges. I have called patients, done data entry. I like to think of us as a true operating partner. It is not just automating what pharmacies do today, it is helping solve operational problems and meet business goals. Sometimes that means suggesting a pharmacy change a workflow, or think carefully about what to automate and in what sequence. The practical benefit, as you said, is that it is a bit like hiring a technician who has worked at many other pharmacies, so there is no learning curve on basic pharmacy operations.

Now, one of the fun parts of implementation was naming the bot. Tell us that story.

Jason: We have a history of naming our automation. We have a SynMed machine we lovingly call Philomena. We have a TruePack system named D. Spencer, one called Checkers, and my favorite, Sir Isaac Newton, the camera that photographs the pills. So naming this one seemed natural. We landed on Hal, inspired by 2001: A Space Odyssey. We do not think it is a threat, and we did not think it would lock us out of the pharmacy. Everybody refers to Hal as part of the team. We have a standup meeting every morning, and it was just easier, and more humanizing, to talk about it that way. The technology can feel threatening to some people’s jobs, so we wanted to humanize it as a tool, and in some respects a colleague, that people would work with.

Natalie: That tells me two things: you have a number of automations, and you treat them as part of the team. Megan, you were closely involved in training and implementing Hal. What did that look like, and what did it take to translate Horizon’s processes into an automated workflow?

Megan: Hal is definitely a member of our team. We have a standup data meeting every morning and Hal comes up. He does not reply, but we give an update on how he is doing and what he is up to. The training process was surprisingly similar to training a newly hired technician who has worked in a pharmacy before, just not in my pharmacy. So we had to go over our processes.

We started in March. First we decided what Hal would do for the first step. We picked one high-volume facility, about 100 to 200 scripts a day, very routine and repetitive, the same four or five doctors, an inpatient facility. It started with the team recording exactly what I was doing as I walked them through it: how to enter a new patient, how to search for insurance, how to enter a new prescription, the quick SIG codes we use. Every week they reviewed it and came back with questions. That went on about four to five weeks. Then we moved into testing, setting aside five or ten prescriptions a day for Hal, reviewing them, and making adjustments. Things come up in training that you do not anticipate, a pop-up or a scenario, and when Hal did not know how to resolve something we would record that and show him what to do. That lasted about six weeks, and we set Hal free in the middle of April, worked into our day-to-day for that one facility.

Natalie: I want to highlight that you started with high-volume, repetitive tasks. That is exactly what automation is great at. Its focus does not drift. People doing repetitive work can naturally skip a step, but once the workflow is configured, the bot consistently follows it. Customers love that the bot always puts in patient notes, always checks the address. It also makes change management easier: when you change part of your process, you change what the bot does, rather than retraining multiple staff. Jason, how did you introduce AI to the broader team, and how did employees respond?

Jason: Pharmacy staff are a skeptical bunch, and change gets a raised eyebrow. Some people were genuinely afraid it meant their job was on the line. It was not, and we said that explicitly, right out of the gate. We brought it up first in broad terms at the standup, since it would affect the data team specifically. We positioned Hal as a tool and as an opportunity to expand people’s skills.

Most LTC pharmacies are small and growing. Our data team worked in little pods, a one-to-one relationship between a facility and a data team member. We wanted Hal to be the platform to upskill everybody and let them learn other parts of the pharmacy they had not had time to learn. We spent a lot of time talking about it, sharing what we were teaching Hal, when, and why, and how it would affect the team, and we keep talking about it. We only launched in March and it is the middle of September, so it is still an iterative process.

Natalie: Open communication is critical in change management, and I love that you had upskilling plans for the team. People sometimes ask whether AI will replace pharmacists, but when you automate, you focus on specific tasks, not everything a pharmacist or technician does. And there are many areas where human judgment is essential. Megan, when Hal does not recognize a scenario, what happens?

Megan: At Horizon every prescription goes through two pharmacist verifications before it leaves. The first, QV1, happens right after data entry. The second, QV2, is the final verification before the medication leaves the door. So every prescription entered by our data team, including Hal’s, gets a QV1 review for accuracy: directions, quantities, day supply, the right person. If an error is found, the pharmacist rejects it to our error resolution queue for a technician to correct.

I monitor any errors Hal makes. When a prescription is rejected, our system shows the technician’s name, and Hal is entered as a technician, last name TJM, so we know when it is his. Just like with our technicians, we treat errors as opportunities for learning, not blame. I document the issue, share it with the TJM team, they review it, and they implement a fix so it does not happen again.

When Hal finds something it does not recognize, this part is really cool. We use QS1 PrimeCare and DocuTrack. We created a “bot skip” key. Hal puts a blue question mark on the prescription and sends me a message through a continuous Google Sheet, telling me why he did not recognize it enough to type it. Every day I review the bot skips, email them to the TJM group, and we work to reduce them.

Natalie: That is the theme of continuous learning and iteration. Scenarios change, pharmacies change workflows, so Hal has to learn new processes, and like a more experienced team member, the automation gets better and better over time. That is why continuous support matters. It is not “build one automation, here you go,” and then it cannot adapt as your operations change, which they will. Now that AI is part of daily operations, how has it changed the way you think about staffing capacity?

Jason: If there are independent and LTC pharmacies on the line, this is what you want to know: is this having an impact? I do not want to give you platitudes, I want to give you specifics. We run about 10,000 to 11,000 scripts a month, and Hal is easily touching and doing something with 30% of that, either a bot skip or putting it into the queue. Megan has managed the bot-skip rate down from the mid-20s at the start to under 10% in the last 60 days. That is about 2,700 scripts a month the team does not have to touch directly. It still hits QV1, which is why we can depend on it safely.

It is really allowing us to cross-train. Three years ago, if a staff member servicing a facility called out on a Friday or Monday, it was a full stop. That pod model worked well but was starting to break as we grew, because so much relationship memory lived with one person. Hal handles the basic work in the background, which lets us cross-train and build relationships with multiple staff members. And it is only running 20 to 30% of our scripts right now, because we have taught it two skills. We have identified five or six, and we iterate on a new one every 90 days. I think it could eventually reach all 11,000 of our scripts if we train it enough.

Natalie: That resilience, business continuity, is a great insight. Automation is not only doing the tasks, it is creating space for staff to grow expertise. Megan, I know you have redirected roughly 1 to 1.5 FTE of capacity away from repetitive data entry. Where has that gone?

Megan: We have reallocated toward more customer-focused tasks rather than routine data entry. Our technicians are more front-facing now, talking to providers, patients, and nursing staff. We shifted the data team toward where they have the greatest impact: answering phones, responding to Mediprocity messages. It has improved workflow efficiency and the level of service to our facilities, and reduced response times. Facilities get human interaction instead of a voicemail. That improvement in outgoing communication is where I have seen the biggest benefit.

Natalie: Jason, where do you see the next opportunities for AI at Horizon?

Jason: First, we will keep growing Hal’s skill set, adding the three to five other data-entry tasks we have identified. Beyond that, there is agentic AI that can do voicemail and call work. Today we are closed-door and B2B with facilities. I see an opportunity to do more business-to-consumer work. LTC-at-home is one area. The assisted living model is strained, and many people would prefer to age at home, so I see us extending into the home. Automation is an excellent way to do that: reminders that go out to patients, “did you get a new script this month,” text-message systems, quick calls. We also have a healthcare collaborative and a self-insurance program through our owners, which is another B2C opportunity unique to us. I genuinely think we can double the number of patients we support over the next two to three years, and extend more profit back to our nonprofit owners, not by cutting employees but by upscaling what they do. Improved service and turnaround time increase satisfaction and help people stay adherent. Packaged medication is a lifesaver, especially for the behavioral health and substance use patients we serve, who make up about 40% of our business.

Natalie: These are not only possible, they are happening today with our customers. We have an LTC-at-home pharmacy where we automate patient outreach. They were always reaching out to ask about recent hospitalizations, medication changes, PRN needs, and deliveries. Automation increased their touchpoints, so they check on patients more frequently, and whenever a patient says yes to one of those questions, they get to talk to a pharmacist. On the B2B side, LTC pharmacies answer calls 24/7, so we automate inbound calls around the clock, helping facilities with refill requests and status updates. Megan, what advice would you give another pharmacy team preparing to implement AI?

Megan: A few things. As with training a new technician, be prepared to invest time in teaching, testing, and giving feedback. The more clearly you define your workflows, the more successful implementation will be. AI is extremely literal, it follows the process exactly as taught, so keep that in mind. We set a goal of giving Hal a new task every 90 days, which let the team give feedback, get used to Hal’s role, and maximize his efficiency before moving to the next task. We are actually working on a new task this week. Most importantly, view AI as a tool that supports your staff rather than replacing them. If you do that, it will be well received by your whole team.

Natalie: We are almost at time. A question came in about KPIs. Beyond FTE capacity, what KPIs do you care most about? Jason?

Jason: Why not more than 30%? The 30% is based on the uptime and number of tasks we have taught it. So one KPI is whether we are teaching it a new task every 90 days, and how many facilities we can engage it with. A second is the bot-skip number, whether it is efficient and effective with what it can do. A third is total volume against uptime. Right now we run it about 18 hours a day, and ideally I would get it close to 24. We want it iterative and intentional. And I will make one very personal: how fast we answer the phone. We track that as a KPI, and it is influenced by the bot. We have gone from missing one or two calls a day to zero, and it has been zero since March, because time was freed up. Anyone who wants to talk offline, connect with me on LinkedIn and I am happy to.

Natalie: One more question: with prescribers sending different forms of Rx, does Horizon try to standardize the format so it is more consistent for the bot?

Megan: The first facility we used actually did not send prescriptions like our other facilities. They used a faxed internal form, typed and electronically signed, a different format from the one that comes through our computer system. Hal was able to learn both without a problem. We did the same process both times: reviewed what it looked like and where the data was. The prescriptions have the same information, just in a different spot, and once you show the bot where it is, it learns it.

Natalie: Amazing. Jason also shared a great RxInsider article Megan wrote, so we encourage you to read it. Thank you, Jason, Megan, and everyone for joining. We will send out the recording shortly, and we look forward to seeing you at our next webinar.