In addition to working with Deepak and the team at Sensal Health, (in my day job) I own and manage multiple pharmacies in Southern California. In my years behind the pharmacy counter and consulting with clinical teams, I have come to recognize a certain kind of patient. She picks up her refills on time, every time. If you asked her whether she takes her medication, she would say yes, and she would mean it. Her chart looks clean. Her refills are generally on time. And yet I hear from the clinical team that her INR is bouncing around, or that her endocrinologist cannot get her TSH to settle, or that the transplant team is puzzled by her variable tacrolimus levels. The differential diagnosis is wide and the conversation usually drifts toward drug interactions, dietary factors, or pharmacogenetic variability. Adherence rarely makes the list, because from what we can see, she is adherent.

This patient sits at the center of a problem that has bothered me for a long time. For most chronic medications, the precision of any given dose does not matter very much. A patient on a statin who occasionally takes 80% of the prescribed dose will not see a meaningful change in their lipid profile. But there is a class of medications, the ones pharmacologists call narrow therapeutic index drugs, where this forgiveness disappears entirely. Warfarin, levothyroxine, and tacrolimus are the three I think about most often. With these drugs, the question is not whether the patient took some medication, but whether they took the right amount. A 20 to 40 percent deviation in either direction can push the patient outside the therapeutic window, and the consequences are serious: stroke or bleeding for warfarin patients, cardiovascular morbidity or bone loss for levothyroxine patients, graft loss for tacrolimus patients. The literature on these consequences is extensive and quantitative. What has been missing is a way to measure dose precision directly at the patient level.

The instruments we have today measure dispensing events. A refill claim tells us the bottle was picked up. An electronic cap tells us the bottle was opened. A pill count at a clinic visit tells us, retrospectively, how many pills are left. None of these distinguishes between a patient who opens the bottle and takes the prescribed pill and a patient who opens the bottle and takes two pills, or none, or one every other day. When the Sensal Health team started developing solutions to fill the gap of providing intelligence on medication quantities dispensed, I could see its value for NTI drugs. Steve Feldman, our colleague, immediately saw the incremental value for topical medications, which are difficult for patients to manage. Steve has often discussed how difficult it is for patients to know whether they are following his instructions, and telling them if they are squeezing out too little or too much is exactly the kind of feedback that helps them stay compliant.

What I find clinically meaningful about direct pill count measurement, the approach Sensal Health has built into MyAide, is that it makes this distinction visible for the first time. A patient with stable pill counts and an out of range INR becomes a candidate for drug interaction or dietary investigation. A patient with divergent pill counts and the same INR becomes a candidate for adherence intervention. These are entirely different clinical conversations leading to entirely different interventions, and current measurement cannot tell us which conversation to have. For pharmacists in particular, who often hold the closest view of a patient's medication routine, having a direct measurement of dose accuracy alongside the existing biochemical markers changes what we can offer. It is the difference between guessing about a patient's behavior and knowing it. For the narrow therapeutic index population, where the cost of guessing wrong is measured in strokes, bleeds, and graft failures, that difference is the entire point.

Creating medication intelligence that actually works in the real world has always been a delicate balancing act. For patients and the clinical teams who support them, we viewed simplicity as the highest priority: the technology should feel intuitive and natural, fitting into routines rather than reshaping them. But given our goal of leapfrogging the proxy-event approaches that have dominated this category, a high degree of technical complexity across the system was always to be expected. Yet if that complexity complicated the patient or clinician experience, the product was likely to flop. This tension ultimately shaped everything that reached the people who use it, and navigating it occupied much of our team's thinking as we developed MyAide.

Some background helps frame the challenge. In an era of increasing reliance on telemedicine and remote patient monitoring, we saw a need for medication intelligence that could measure compliance with precision and put that information in front of clinicians and pharmacists in a form they could act on. What we built is a complete system rather than a single device: a smart dock and cap that measure medication use gravimetrically, a companion smartphone app for patients, and a set of dashboards for clinician and pharmacist workflows. Intelligence here is not just measurement. It is the whole loop, from what happens at the patient's counter to what shows up in a clinical workflow. For that loop to be useful, every part has to function together reliably and seamlessly.

What kept us honest throughout was a parallel commitment to simplicity of user experiences even while we welcomed technical complexity. Smart health products are nearly always grudge purchases. Patients only use them because they are dealing with a condition that already creates obstacles in their daily life, which leaves them with a very low tolerance for additional impositions. The same logic applies to the clinicians and clinical operations staff who interact with the system on the other end of the data. Our goal was to impose no further burden on either group. In practice, this meant that every time we added intelligence to the system, we measured the resulting experience against what performing that task would feel like without the technology at all. Sometimes complexity was clearly warranted, as when we built automatic data transmission so patients would not need to take any extra steps beyond using their medication. Other times, evaluating design ideas against real-life behavior made it immediately clear when we were asking too much of the user. Multiple rounds of feedback, from clinical experts at one end and everyday patients at the other, were essential to keeping that balance. Those additional iterations took time and effort, often delaying our updates by a few months. The pressure was not easy to take, for a startup, but we had no way to compromise.

The patient experience ratings we received at the end of a recent study are, frankly, gratifying. On a five-point scale, patients rated the system 4.7 for ease of use and for helping them take their medication on time, 4.5 for likelihood to use again, and 4.4 for being visually pleasing. Those numbers tell us we threaded the needle, at least well enough that patients found the intelligence helpful rather than burdensome. Qualitative feedback, comments during zoom calls, and other feedback from clinicians and operators have been no less important in giving us a shared sense of rewards.

The line from idea to product almost never follows a simple path, and balancing a patient's need for simplicity against the demands of building genuine intelligence is a significant challenge. But success is attainable with a steady vision and shared principles to guide the process.