The engineer’s lens in a clinical organisation
I trained as an engineer, and the training never left. I see a hospital group the way I was taught to see anything: as a system with inputs, throughput, losses and feedback. Demand comes in through search, listings, the contact centre and referrals. It is converted, or lost, at a series of stages. What converts produces revenue at a margin. Everything can be measured, and what can be measured can be improved. This lens built my career, and it is the reason a hospital group hired me to run its digital front door.
It is also the reason clinicians were suspicious of me for the first two years, and on reflection they were right to be. The lens is powerful exactly where a hospital behaves like a system, and it is quietly wrong where it does not. The trouble is that an engineer cannot tell the difference from the inside. You have to be told, usually by a doctor, usually in a meeting that goes badly.
This is my account of where the lens has helped, where it has got in the way, and what I have had to learn about the parts of a hospital that no amount of instrumentation will explain.
Where the lens helps: the front door
The patient’s journey up to the consultation room is a system, and it is a badly run one in most Indian hospital groups. Search listings for hundreds of doctors across a dozen units, maintained by nobody. A contact centre that answers in two languages in a city that speaks four. A booking journey that breaks on the payment step. A CRM that records the enquiry and forgets it. Callbacks that depend on who is on shift.
An engineer looks at this and sees leakage at every stage, and is correct. This is where measurement, process and systems thinking earn their place without apology. Instrument the funnel. Find where the enquiries are lost. Fix the largest loss first. Repeat. Nothing about this offends a clinician, because none of it touches clinical work, and the results — more patients arriving, fewer enquiries lost — are results the department heads want.
The same applies to the data. A hospital group’s patient identity is fragmented across the hospital information system, the CRM, the app and unit-level records. An engineer’s instinct to reconcile it into one model is the right instinct, and the years of unglamorous work it takes is the most valuable thing a digital function can do. Nobody resists it. They just do not fund it, until someone shows them the number it produces.
Where the lens helps: the commercial case
The second place the engineer is right is in the P&L conversation. Hospitals have historically made growth decisions on reputation, relationships and the persuasiveness of a senior doctor. An engineer brings a different discipline: what does the demand data say, what is the conversion, what is the cost, what is the payback, and what is the sensitivity if the assumption is wrong.
Executive committees respond to this. A capital case for a new format built on ninety days of measured demand carries more weight than one built on a clinician’s conviction, and it should. The engineer’s insistence on evidence before capital is the single habit that most improves a group’s growth decisions, and it is the one I would not give up under any pressure.
The trap is believing that because the lens works in the commercial conversation, it works in the clinical one. It does not, and the reasons are worth understanding rather than resenting.
Where a hospital does not behave like a system
A booking funnel has a stable structure. A patient with chest pain does not. The same presentation can be indigestion or a cardiac event, and the doctor’s judgement about which — made in minutes, on incomplete information, with the patient’s life in the balance — is not a process step. It is the whole point of the hospital, and it does not have a conversion rate.
I learnt this the slow way. Early on, I proposed instrumenting the outpatient consultation the way I had instrumented the funnel: time per patient, follow-up rate, conversion from consult to procedure. The numbers were easy to produce. The medical director looked at the dashboard and asked me what a low conversion from consult to procedure meant. I said it might mean a doctor was not recommending treatment the patient needed. He said it might mean the doctor was the only one in the department not recommending treatment the patient did not need. I had no way to tell the difference. Neither did the dashboard.
That is the general shape of the problem. In a system, a measured variation is a signal to be reduced. In clinical practice, a measured variation may be judgement, or it may be error, and the number alone cannot say which. An engineer who treats every variation as waste will, sooner or later, optimise away the thing that was keeping patients safe.
There are other places the system model fails. Capacity in a hospital is not a fixed input; it flexes with who is on shift and what walked in overnight. A department’s willingness to release slots to the digital front door is not a configuration setting; it is a negotiation about trust. A doctor’s reputation, which drives more demand than any channel I run, is not a brand asset the group controls. It is a person’s, and it leaves when the person does.
Where clinicians are right to resist
I used to read clinical resistance to digital initiatives as inertia. Some of it is. Most of it, I now think, is a reasonable response to an engineer proposing something that would work in a system and does not work here.
They resist measurement that does not account for case mix. A surgeon with the worst outcomes in a department may be the one taking the cases nobody else will. A dashboard that ranks doctors by a simple outcome metric and is visible to management will, within a quarter, change which cases get taken. Clinicians know this. Engineers learn it after the damage.
They resist process that removes discretion. A standardised patient journey is good for the patient who fits it. The patient who does not — the elderly one with three conditions, the child whose parents are frightened, the one who cannot read the form — needs a human to step outside the process, and a process designed by someone who has never been in the room tends to make that step harder. The contact centre scripts I wrote in my first year were tight, measurable and wrong for a third of callers.
They resist tools that make the doctor serve the system rather than the reverse. Every additional field, every mandatory step, every workflow that exists so the dashboard can be populated is a minute taken from a patient. Doctors are right to ask who the minute is for. When the honest answer is the digital function’s reporting, they are right to refuse.
And they resist the assumption that because the engineer has measured something, the engineer understands it. This is the one that stings, and it is the one that is most often correct.
What an engineer has to learn about judgement
Judgement is what a system does not have. It is the capacity to decide well on incomplete information when the rules do not fit. Clinicians are trained in it for a decade before anyone lets them practise alone. Engineers are trained to eliminate the need for it. Neither is wrong. They are different jobs, and a growth leader in a hospital group has to hold both without pretending to be either.
What I have learnt, mostly from medical directors who were patient with me.
Measure the journey, not the medicine. Everything up to the consultation room and everything after the patient leaves it is mine to instrument. What happens inside it is not, and the only measures I bring into it are the ones a clinician has designed with me and agreed to be judged by.
Treat variation as a question before treating it as a fault. When a number is different across units or doctors, the first step is to ask why, in person, of the person whose number it is. Half the time there is a reason the dashboard cannot see. The other half there is a problem, and the conversation has earned the right to raise it.
Build process that has an exit. Every journey, script and workflow I now ship has an explicit point at which a human is allowed to step out of it without justification. The measurement records that it happened and does not penalise it. The frequency of exits tells me where the process is wrong, which is more useful than a process nobody is allowed to leave.
Accept that the most important inputs are not in the data. Trust between a department and the front door. A doctor’s reputation in a city. Whether the medical director believes the growth function is on the patient’s side. None of these is a column. All of them determine whether the system works.
Where the two lenses meet
The productive place is the boundary. A clinician who has seen a well-instrumented front door starts asking for the same clarity in the parts of the journey they control, on their terms. An engineer who has been shown where the system model fails starts building tools that leave room for judgement, and those tools get used. The relationship between the growth function and the medical director is where this happens or does not, and it is a relationship, not a process.
I still see the hospital as a system. I have simply learnt that the diagram has a large box in the middle labelled judgement, that I do not get to open it, and that everything I build has to work around it rather than through it.
If you are bringing this lens into a hospital next year
- Instrument everything up to the consultation room. Nothing inside it, until a clinician asks you to and designs the measure with you.
- Before proposing any clinical-adjacent metric, ask a medical director what a bad number would mean. If there are two plausible answers, do not ship the metric.
- Design every script and journey with a documented exit. Measure the exits. Do not penalise them.
- Take every unit-level variation to the person whose number it is before it reaches a dashboard anyone else can see.
- Sit in a department’s morning meeting for a month. Say nothing. Learn what they argue about and why your dashboard does not show it.
- Keep the engineering discipline for the commercial case, where it belongs, and be willing to be told where it stops.
An engineer in a hospital is useful for exactly as long as he remembers that the system exists to serve a judgement he is not qualified to make.
Questions people ask
The front door and the commercial case. The patient’s journey up to the consultation room is a system, and a badly run one in most Indian hospital groups: listings maintained by nobody, a contact centre answering in two languages in a city that speaks four, a booking flow that breaks on payment. Instrument the funnel, find the largest loss, fix it, repeat. And in the P&L conversation, insisting on demand evidence before capital is the single habit that most improves a group’s growth decisions.
Inside the consultation room. A booking funnel has a stable structure; a patient with chest pain does not. The doctor’s judgement about whether it is indigestion or a cardiac event, made in minutes on incomplete information, is not a process step and has no conversion rate. Capacity flexes with who is on shift. A department’s willingness to release slots is a negotiation about trust, not a setting. And a doctor’s reputation, which drives more demand than any channel I run, leaves when the doctor does.
Usually because they are right. They resist measurement that ignores case mix, because the surgeon with the worst outcomes may be taking the cases nobody else will, and a visible ranking changes which cases get taken within a quarter. They resist process that removes discretion, because the elderly patient with three conditions needs a human to step outside it. They resist every mandatory field that exists so a dashboard can be populated, and they are right to ask who the minute is for.
Not without a clinician designing the measure. I proposed it early on. The medical director asked what a low conversion meant. I said a doctor might not be recommending treatment a patient needed. He said the doctor might be the only one in the department not recommending treatment the patient did not need. I had no way to tell the difference, and neither did the dashboard. Pressure on that number moves it, and the wrong patients get operated on.
Everything up to the consultation room and everything after the patient leaves it. Search, listings, contact centre, booking, callbacks, the CRM, discharge follow-up, reviews. What happens inside the room is not the growth function’s to measure, and the only measures worth bringing into it are ones a clinician has designed with you and agreed to be judged by. Before proposing any clinical-adjacent metric, ask a medical director what a bad number would mean. If there are two plausible answers, do not ship it.
Every script, journey and workflow I now ship has an explicit point at which a human is allowed to step out of it without justification. The measurement records that it happened and does not penalise it. The frequency of exits tells me where the process is wrong, which is more useful than a process nobody is allowed to leave. A standardised journey serves the patient who fits it; the exit serves the one who does not — the frightened parent, the patient who cannot read the form.
The contact centre scripts I wrote in my first year were tight, measurable and wrong for a third of callers. I read clinical resistance as inertia when most of it was a reasonable response to something that works in a system and does not work here. And I assumed that because I had measured something, I understood it. That is the assumption that stings most when a doctor corrects it, and it is the one that was most often correct.
Treat variation as a question before treating it as a fault. Take it to the person whose number it is, in person, before it reaches any dashboard others can see. Half the time there is a reason the data cannot show — case mix, a visiting arrangement, a referral pattern. The other half there is a problem, and the conversation has earned the right to raise it. An engineer who treats every variation as waste will eventually optimise away the thing keeping patients safe.
About two years, in my case, and the suspicion was earned. Trust is built by instrumenting only what is yours, by taking variation to the doctor before the dashboard, and by sitting in a department’s morning meeting for a month saying nothing. Once a clinician has seen a well-run front door, they start asking for the same clarity in the parts they control, on their terms. That is when the two lenses meet, and it is a relationship, not a process.
That capital should follow measured demand rather than a senior doctor’s conviction. A case for a new format built on ninety days of demand data — what the city searches for, what converts, what it costs, what the payback is and what the sensitivity is if an assumption is wrong — carries more weight than one built on persuasion, and should. The CFO should also recognise where the discipline stops: a clinical-adjacent metric on a P&L slide is a liability, not evidence.
Leave room for judgement. Every additional mandatory field is a minute taken from a patient, so if a field exists only to populate reporting, remove it. Build the exit into the workflow and record it without penalty. Reconcile the fragmented patient identity across HIS, CRM and app — that is engineering work nobody resists, they just do not fund it until someone shows the number it produces. And never ship a clinical-adjacent metric to a shared dashboard before the clinician whose number it is has seen it.
The engineering discipline for the funnel and the commercial case, and the humility to know where it ends. Concretely: the ability to instrument a journey, build a demand case a CFO will fund, and then sit in a medical director’s office and be told the dashboard is wrong without defending it. The people who fail in this seat are not the ones without technical skill. They are the ones who cannot hold a box labelled judgement that they are not qualified to open.
Whenever the most important inputs are not in the data. Trust between a department and the front door. A doctor’s reputation in a city. Whether the medical director believes the growth function is on the patient’s side. None of these is a column and all of them determine whether the system works. It is also the wrong tool for anything where a measured variation could be either judgement or error, because the number alone cannot say which, and acting on it as if it could does harm.
