A hospital administrator in Kanpur ran a six-week referral incentive programme — ₹200 credited to a referring GP’s account for every OPD patient who converted to admission. At the end of six weeks, admissions were up 18%. The programme was declared a success and rolled out across all departments.
What nobody noticed: a 40-bed private hospital three kilometres away had been closed for most of that period due to a licensing dispute. When it reopened, admissions returned to their original level. The incentive programme had cost ₹6 lakh. It had proven nothing except that patients go somewhere when their usual option is unavailable.
This is not a growth experiment. It is a coincidence mistaken for a result.
The previous piece in this series argued that the operators who grow consistently are not the ones chasing the largest market — they are the ones who have defined the smallest market they can actually win. TAM/SAM/SOM is a framework for market clarity. But knowing your SOM is the beginning of the work, not the end of it. Once you know which patients are realistically available to you, you still need to win them. And winning them requires learning. Learning, in a business context, means running experiments.
But most healthcare operators don’t run experiments. They make bets.
The distinction is not pedantic. A bet is a single action with an uncontrolled outcome reviewed in hindsight. An experiment is a designed test with a specific question, a defined success condition, and a controlled environment that allows you to trust what you observe. Bets generate anecdotes. Experiments generate knowledge. The difference compounds: a practice that has run fifty rigorous experiments in five years knows things about its market that a practice that made fifty bets in the same period simply doesn’t.
So, in this article I’m teaching you how to run experiments, and how to run them better.
Define what failure looks like before you start
Healthcare operators write success-biased intentions, not hypotheses. “We will invest in digital marketing to improve new patient acquisition.” This statement cannot fail. Acquisition might rise by 2% and the initiative gets declared a success. Acquisition might fall for unrelated reasons and digital marketing gets the blame. Neither outcome teaches you anything, because you never defined what you were actually testing.
A hypothesis is a falsifiable claim with a number attached to it. “If we run targeted Facebook ads for our diabetes OPD to a 5-km radius around the clinic for thirty days at ₹15,000 total spend, we expect at least 12 new diabetes consultations that cite the ad as their referral source. If we see fewer than 8, the channel does not justify the spend at this scale.” That hypothesis can fail. That is the point. When failure is defined in advance, any result — positive or negative — tells you something.
The test of a good hypothesis is simple: write down what would convince you that this idea was wrong. If you cannot answer that question before you start, you are not running an experiment. You are running a bet with paperwork.
Change one thing at a time
A nursing home in Meerut decided to improve its patient retention rate. Over the same quarter, it introduced automated appointment reminders via WhatsApp, replaced its front desk manager, and launched a post-discharge follow-up call programme. At the end of the quarter, retention had improved by 11%.
Which of the three changes caused it? The operator assumed it was the WhatsApp reminders, because those had been the most visible initiative. The new front desk manager was the more plausible cause — she was significantly better at managing patient complaints than her predecessor. The follow-up calls had almost no uptake. But because three variables changed simultaneously, the operator couldn’t know. Twelve months later, the WhatsApp vendor contract was renewed, the follow-up call vendor was renewed, and nobody noticed that the actual driver of improvement had quietly resigned and been replaced by someone less effective.
A scientist never changes the temperature, the pressure, and the chemical composition all at once. If the beaker explodes, they won’t know why. Healthcare operators routinely change the referral incentive, the sales script, and the marketing channel in the same quarter — and then conduct a retrospective that tells them nothing except what they already wanted to believe.
Isolate the variable. Change one thing. Observe the result. Then change the next one.
Decide when to stop before you start
The most common error in healthcare growth experiments is stopping when the early results look good. A practice launches a GP liaison programme, checks the numbers after three weeks, sees referrals up 30%, and declares the programme a success.
Three weeks is noise. Referral patterns are seasonal, relationship-driven, and lumpy. A single doctor who sent three patients in one week because of a personal conversation will inflate a three-week result in ways that have nothing to do with the programme design. A conclusion worth acting on requires enough data points to rule out randomness — which in most referral development contexts means a minimum of six to eight weeks and a defined comparison group.
Define your required sample size before the experiment starts. Write it down. Commit to it. If you decided you need eight weeks of data, do not declare victory at three because the early numbers are flattering. Early results are frequently noise. The discipline to wait for signal is harder than it sounds when the pressure to act is strong.
Don’t scale what you haven’t replicated
In science, a result isn’t a fact until another lab can reproduce it. One successful experiment is an anecdote. The same result across multiple trials, in different conditions, with different samples — that is a pattern worth acting on.
Healthcare operators scale too fast and replicate too rarely. A clinic in Chandigarh runs a patient satisfaction initiative — a structured post-consultation feedback process combined with a follow-up call at 48 hours. Retention improves. The owner has two other clinics, in Ambala and Patiala. The initiative gets rolled out to both within the month.
Six months later, it is working in Ambala and has no measurable effect in Patiala. The owner cannot explain why. The answer, on closer inspection, is not complicated: the Patiala clinic has a fundamentally different patient profile — older, more rural catchment, lower smartphone penetration, different expectations of the doctor-patient relationship. What drove retention in Chandigarh was a digitally-mediated engagement loop that simply doesn’t apply to that population. The initiative wasn’t wrong. It was right for one context and wrong for another. Scaling it before replicating it cost six months of operational distraction and a meaningful amount of management attention.
The rule is straightforward: if a tactic works in one quarter, re-test it in the next before making it permanent policy. If it works in one location or one patient cohort, pilot it in a second before assuming it generalises. Context changes — by season, by patient mix, by who is running the programme on the ground. What produced the result the first time may not be what you think it was.
One success is an anecdote. Two successes under different conditions is evidence. That distinction is worth the discipline of waiting.
Log your failures as rigorously as your wins
Healthcare practices have institutional amnesia at a rate that would be alarming if anyone measured it. The health camp that produced no new patients. The Google Ads campaign that drove enquiries but no conversions. The specialist tie-up that generated two referrals in a year. These experiments were run, they failed quietly, the conclusions were not recorded, and three years later a new manager proposes the same initiative to a room of people who have forgotten they tried it before.
Failure is data. The reason scientists publish null results is that knowing what doesn’t work is as valuable as knowing what does — it prevents others from wasting time and money repeating the same dead ends. A practice that documents what it tested, what it expected, and what actually happened is building a body of knowledge that its competitors don’t have. “We tried GP outreach in that colony eighteen months ago — eight referrals from twelve meetings over six weeks, conversion didn’t justify the field staff cost at that density” is worth more than any positive case study you can read about someone else’s practice in a different city. It is specific. It is yours. It tells you exactly what not to repeat.
Five rules. Most operators follow none of them consistently, because the pressure to act — to add the department, run the camp, hire the manager — is immediate, and the cost of undisciplined action only becomes visible later, when the practice is busy but structurally unable to grow.
The operators who grow consistently don’t have better instincts than the ones who plateau. They have better information. Not industry reports or market research on healthcare trends in tier-2 cities — specific, hard-won knowledge about which interventions work in their specific market, with their specific referral base, at their specific price points.
That knowledge is not available to buy. It can only be built, one properly-designed test at a time.
Running well-designed growth experiments is one of the pillars of the growth strategy guide for private clinics and hospitals in India.
Leave a Reply