What does pharmacy automation actually include?
The word covers several distinct things which are sold together and should be evaluated separately.
Storage and retrieval robots hold packs densely and fetch them on demand, which is primarily a space and picking-speed technology rather than a dispensing one. Automated pouch and blister systems assemble compliance packaging, which is a different problem with different failure modes. Conveyor and sortation moves items between stations. And the software layer ties it to the patient record, which is where most implementations actually succeed or fail.
A pharmacy evaluating automation should be clear which of those it is buying and which problem it is solving, because the business cases differ entirely. Storage density suits a constrained premises. Picking speed suits high volume. Compliance packaging suits a care home contract. A vendor answering all three at once is describing a product range rather than your solution.
Why throughput is the wrong headline number
Every vendor leads with packs per hour, and it is close to useless as a purchasing criterion because it describes the machine's performance on the work it is good at. What determines whether an installation improves a dispensary is the opposite figure, being the proportion of your prescriptions the machine cannot handle and what happens to them. Fridge lines, controlled drugs depending on configuration, liquids, appliances, awkward pack geometries, split packs and anything needing assembly rather than picking typically sit outside. If a fifth of your items are exceptions, you have not automated dispensing, you have automated four fifths of it and created a second parallel process for the rest, and the second process is now the one that sets your actual pace. The question to put to every vendor is therefore not how fast it goes but what share of this pharmacy's product mix it would exclude, followed by how those items reach a person, and whether the exception route is designed or is simply what is left over.
What happens to the items the machine cannot handle
Three questions establish whether the exception route was designed or inherited.
Are exceptions identified before or after picking begins? A system which discovers an item is unhandleable partway through an order has already consumed time and created a partial basket, which is worse than routing it correctly at the outset.
Do exceptions and automated items reunite reliably? A prescription split between a machine and a bench has to come back together, and that reconciliation is where errors enter. Ask to see it happen rather than hear it described.
Who checks what? The professional check does not disappear because a machine did the picking, and the arrangement needs to be explicit about what a pharmacist is verifying and how they can verify it when they did not watch the item being selected.
What happens when it stops
It will stop, and the difference between an inconvenience and a closure is decided by three things which belong in a contract rather than in a conversation.
Can you get your stock out? This is the question that matters most and is asked least. If the machine holds a substantial proportion of the pharmacy's inventory and cannot be opened manually during a fault, the pharmacy cannot dispense at all. Manual retrieval should be demonstrated rather than asserted.
What is the guaranteed response, and when? A four-hour response on a Tuesday and a next-working-day response on a Saturday are different products, and a dispensary needs the weekend answer.
Can you still trade? The fallback process should be written, resourced and practised, in the same way the downtime arrangements described in the clinical safety article are, because the day it is needed is not the day to invent it.
How it connects to your PMR
This is where implementations most commonly disappoint, and the failure is rarely mechanical.
A robot which does not know what was prescribed has to be told, and the quality of that connection determines whether the installation saves labour or relocates it. Four things are worth establishing precisely. Whether the integration is bidirectional, so the record knows what was picked rather than only the machine knowing what was requested. Whether stock levels reconcile automatically or require counting. Whether an exception raises a flag in the PMR or only on the machine's own screen, since staff watch one and not the other. And whether the audit trail of what the machine did lands on the patient record or in a separate system nobody consults during an investigation.
A pharmacy which cannot see, from the patient record, that a given item was picked by the machine at a given time has an automated dispensary and a manual audit trail, which is the worst combination available.
Who is accountable when assembly happens elsewhere
Automation within your own premises and hub and spoke are separate decisions which vendors frequently present as one journey, and conflating them is the most consequential error in this area.
Automating your own dispensary is an operational decision about space and speed. Moving assembly to another site is a strategic and regulatory decision carrying its own governance obligations, supplier concentration risk and a reversibility problem, all of which the hub and spoke analysis examines and none of which follows automatically from buying a machine.
What does not change in either case is where the professional responsibility sits. The pharmacy supplying the patient remains accountable for that supply, and the regulator's expectations concerning oversight follow the arrangement rather than being discharged by it, which is the finding the thematic review reported. The practical consequence is that the governance work increases rather than decreasing, because a pharmacy now has to assure a process it no longer watches.
The data questions nobody asks until later
Four, and each becomes expensive at a different point.
Who owns the operational data? Pick rates, error rates, stock movements and utilisation are generated by your operation and are commercially informative. Establish who can access them and for what.
Can you export it? Stock and transaction history should leave the system in a usable format, for the same reasons set out in the data export guide.
What happens at end of contract? Not merely to the hardware but to the data and the integration, and whether a successor system can be connected without the incumbent's cooperation.
Does any patient data reach the vendor? Frequently it does, through support access or remote diagnostics, and that requires a written processing arrangement rather than an assumption.
How to work out whether it pays
On your own numbers, over five years, with the costs vendors omit included.
The benefits side is usually stated well, comprising labour hours released, storage density, picking accuracy and, where relevant, the capacity to take on volume the premises could not otherwise handle. The costs side is where business cases are optimistic. Include installation and premises works, which for a retrofit can rival the machine. Integration, including any charge from your PMR vendor. Service and consumables annually. Training, and retraining as staff turn over. The parallel running period, during which both processes operate and neither is efficient. And the exception process, which is a permanent operating cost rather than a transition one.
Two structural points determine the answer more than any of those lines. Fixed costs do not scale down, which is why the arithmetic is harder for a single site. And released labour only becomes a saving if it is actually redeployed or removed, so a business case resting on hours freed should say what those hours will be doing, since hours which simply become less pressured are a benefit rather than a return.
How to run a site visit that tells you something
Vendors will arrange a reference visit, and most visits are wasted because the visitor watches the machine work.
Ask instead to visit at the busiest hour rather than a convenient one. Watch what the staff do rather than what the machine does, since that is where the labour actually went. Ask the reference site what they would do differently, which is the single most informative question available and one an honest reference will answer. Ask what still goes on the bench and what proportion that is. And ask when it last stopped and what happened that day.
A reference site which cannot answer the last two is either unusually fortunate or has not been asked before, and both are worth knowing.
Questions to ask any automation vendor
Eight, in order, and the first two do most of the work.
What proportion of our specific product mix would this exclude, and how do those items reach a person? What is the guaranteed response time on a Saturday, and can we retrieve stock manually during a fault? Is the PMR integration bidirectional, and does an exception surface in the record or only on the machine? Where does the audit trail of what the machine did land? What is the five-year total cost including installation, integration, service, consumables and training? What operational data do you access, and what happens to it and to our integration at end of contract? Which reference sites can we visit at their busiest hour? And what does this not do well, in your own words?
A vendor who answers the last question candidly is telling you more about what it is like to work with them than any of the others.
Key takeaways
- Automation covers several distinct technologies solving different problems, so be clear whether you are buying storage density, picking speed or compliance packaging.
- Throughput describes performance on the work the machine is good at, and the useful figure is the proportion of your own product mix it would exclude.
- A large exception rate means a second parallel process which then sets the dispensary's actual pace, so ask whether the exception route was designed or inherited.
- Settle downtime contractually, particularly whether stock can be retrieved manually during a fault, since a pharmacy that cannot reach its own stock has a closure rather than an outage.
- Integration decides whether labour is saved or relocated, and the machine's audit trail must land on the patient record rather than in a separate system.
- Automating your own dispensary and moving to hub and spoke are separate decisions, and professional accountability for the supply does not transfer in either case.
- Model five years on your own volumes including installation, integration, service, training, parallel running and the permanent exception process, and say what released hours will actually do.
FAQs
Automate the picking, not the audit trail.
Whatever assembles the pack, the record has to show what was picked, by what, and when, on the same patient journey as the assessment and the supply. Dataforge PMR keeps that trail intact whether the item came off a machine, off the bench or from a hub.
See Dataforge PMR