OPD queue management in Indian hospitals: where it clogs and what fixes it

Estimated reading time: 5 minutes

Part of the hospital operations notes, alongside Selling AI to Indian Hospitals.

Walk into a busy Indian OPD at 10am and the problem is not that nobody is working. Everyone is working, hard. The problem is the queue: one long line where a patient who needs thirty seconds waits behind a patient who needs ten minutes, and nobody in the line can tell which one they are. Queue management is not about making staff move faster. It is about making sure the right patient reaches the right desk at roughly the right time, so the same number of people get seen with far less standing around. Here is where OPD queues actually clog in Indian hospitals, and what genuinely unclogs them.

At BeyondChats the pattern repeats across very different hospitals: the wait patients complain about is rarely the doctor’s time. It is everything around it.

Where the queue clogs, in one line each:

1. One line for many different needs, so quick tasks wait behind slow ones.
2. Registration collects at the desk what could have been collected before arrival.
3. Nobody can see their place, so everyone crowds the front to be safe.
4. Peaks are predictable, but staffing is flat.

Split the line before it forms

A single queue is fair in theory and slow in practice, because it forces a patient collecting a report to wait behind a patient registering as a new case. The fix is to sort people by what they actually need before they join a line: new registration, follow-up, report collection, billing. A patient who only needs a report should never enter the consultation queue. Even a simple triage at the entrance, a person or a screen asking one question and pointing each patient to the right desk, takes a surprising amount of load off the main line. You are not adding capacity. You are stopping the fast tasks from getting stuck behind the slow ones.

Move registration off the desk

The registration counter is where flow most often dies, because everyone must reach the same desk to give the same details before anything else can happen. Much of that work does not need to happen at the desk. Details captured when the appointment is booked, a patient who arrives with a reference number instead of a blank form, routine questions answered ahead of time: each one turns a two-minute desk interaction into a ten-second one. The counter then handles the exceptions, not the whole crowd. This is the same seam as the integration work: the more the system of record already knows before arrival, the shorter the desk queue.

Let patients see their place

A lot of the crowding at the front of an OPD is not impatience. It is fear of missing a turn. When a patient cannot tell whether they are next or fortieth, the safe move is to hover near the door and ask repeatedly, which adds noise and load for staff. A visible token number, a display, or a simple message that says “you are number 12, roughly forty minutes” changes the behaviour entirely. People sit down. They step out for water without panic. The wait feels shorter even when the clock says the same, and the corridor stops being a scrum. This is the “invisible wait” made visible, and it is one of the cheapest wins in the building.

Staff the peaks, not the average

OPD load is not random. It spikes in the first hours of the morning and around specific clinics on specific days, and anyone who runs the department already knows when. The queue clogs when staffing is set for the daily average instead of the morning peak. You do not always have more people to add, but you can shift breaks, open a second registration point for the first two hours, or pull a spare hand to the entrance triage when the rush hits. Reading yesterday’s arrival pattern to plan today’s is not sophisticated analytics. It is looking at a number most hospitals already have and choosing to act on it.

Where AI actually helps

None of this needs artificial intelligence to state. What software adds is doing the boring parts at scale: sorting patients by need before they arrive, collecting registration details in the channel they already use, sending a live queue position so nobody has to hover, and flagging the predictable peaks before they hit. The honest limit is the same as everywhere in patient flow: technology moves the information and communication around the queue, not the physical capacity inside it. If you have too few doctors or too few counters, no token system fixes that. But a large share of OPD waiting is not a capacity problem at all. It is a coordination problem, and coordination is exactly what this kind of tooling is for. The number to watch is average waiting time, which makes it a clean single metric for a pilot.

This is part of what I’m writing about here: building AI for Indian healthcare, in public. If that’s your world too, here’s why I started writing, and the monthly letter below is where the numbers and the messier lessons go.

Related: queues are one of four places patient flow breaks, the empty-slot version of the problem is the no-show, and shorter desk queues start with what the system knows before arrival.