When Automation Makes a Practice Slower, Not Faster
Automation makes a practice slower when it is pointed at work that does not repeat, and it does so reliably rather than occasionally. That is not a vendor failing. It is what happens when a tool that needs consistency is applied to a process that never had any.
Most of the slowdown is invisible in a business case, because it lands on different people than the ones who got the saving.
If two competent people do the task differently, there are two processes. Automate one and you have made the other person slower.
Four ways a practice gets slower
Why the four happen
Nobody agrees what the process is
Ask three people in the same practice how a matter is opened and you get three answers that are each partly right. That is normal in a firm small enough to trust its people. It is also why automating “the” process fails: there is no single process to encode.
What gets built is a version. The person whose approach was not chosen now does the task twice — once the way they know and once to satisfy the system. Their day got longer so the report could say the firm saved time.
The 80% claim does not survive month end
“It handles 80% of cases” sounds like a win until you look at what the remaining 20% costs. In accounting, legal and medical work the exceptions are frequently the interesting cases, and they cluster at month end, at settlement, and at the point where someone is unwell.
The 80% disappears quietly. The 20% arrives with a timestamp, an owner and a queue. If nobody was reassigned to work that queue, the queue is a delay, and the delay is invisible until a client asks where their matter is.
Something has to look after it
A workflow is not a purchase. It is an asset with a version, a dependency on other systems, and a failure mode. When the practice management system updates, when a form changes, somebody has to notice and fix it.
None of that appears in the hours-saved estimate. In most practices the maintenance lands on the person who was most enthusiastic about the project, which is also the person with the least spare time. And for the first two months nobody fully trusts it, so lookups happen twice and the measured time goes up before it comes down.
The test that tells you which one you have
Write the process on one page, in the order it happens, with every branch you can think of. Not a policy. The actual steps, including the ones that only happen sometimes.
- If it fits on a page and two people would write it the same way, it is a candidate.
- If it needs three pages of exceptions, the exception handling is the job — automating the tidy part will not help.
- If two people disagree about the order of steps, stop. Fix the disagreement first; it is cheaper than software.
- If the volume is low, the maths rarely works no matter how clean the process is.
This is a different question from whether the build takes longer than the demo promised — it will. That is the delivery problem. This is about the running.
Four things not to automate
The case where slower is correct
Automate when the work is high volume, identical, and cheap to check. Document intake, invoice coding, matching bank lines, filing correspondence, assembling a standard letter. These are the jobs where the same thing happens two hundred times a week and a person can spot a wrong one in four seconds. There the slowdown is temporary and the saving is real.
A person can check that output faster than they could have done the work. That is the actual test. If verifying the result takes as long as producing it, automation has moved the work rather than removed it — and added a system to maintain.
What we build
What we build
We start by writing the process on one page with you. If it does not fit, we say so and we do not build. When it does fit, we build the narrow version that handles the repeatable part and hands the rest to a person with the context attached.
Pingback: Practice Website UX: Where Enquiries Actually Drop Off
6 November 2026