Almost every firm that outsources runs a pilot first, and almost every pilot is designed to succeed rather than to inform. That is an understandable instinct and a expensive one, because a pilot that cannot fail tells you nothing about what happens in month eight.
Here is the problem in its usual form. A partner picks a client — typically a tidy one, with clean records and a patient owner. The provider assigns their strongest accountant. Both sides watch the work closely for a fortnight. The output is excellent. The firm signs, scales to thirty clients, and discovers over the following quarter that the median experience looks nothing like the pilot.
Nothing dishonest happened. The pilot simply measured the wrong thing.
Design for the median, not the showcase
The single most useful change is to pick a file in the middle of your book rather than the top of it. Ideally something with a moderate mess: a few unreconciled months, an owner who answers questions slowly, a chart of accounts that grew organically rather than by design.
What you are testing is not whether a good accountant can do clean work. It is whether the provider’s process copes with the conditions your book actually contains. A pilot on your best client tests the accountant. A pilot on a median client tests the system.
Test the handoffs, not just the output
Most outsourcing failures are not accounting failures. They are communication failures: a query that sat unanswered for four days, an assumption nobody surfaced, a deliverable that arrived in a format your reviewer had to rebuild.
So instrument those things explicitly. During the pilot, log:
- Every query raised, when it was raised, and when it was resolved
- Every assumption the provider made without asking — and whether it was correct
- How much time your own reviewer spent on the file, honestly measured
- The number of round trips before the deliverable was accepted
That last one is the number that matters most and the one nobody records. If a deliverable takes three round trips in the pilot, when both sides are paying maximum attention, it will not take fewer at scale.
Insist on the documentation deliverable
A good provider will document your process before they take it. Make this an explicit, separately assessed pilot output rather than a background activity.
Read the SOP they produce. Is it a genuine description of how the work is done, including the judgement calls, or is it a generic template with your client’s name inserted? The answer tells you a great deal about whether they understood the work or merely completed it.
This document also has standalone value. Several firms have told us the documentation exposed single points of failure in their own practice that they had not seen — one partner discovered their entire payroll bureau process lived in one person’s head, undocumented, after eleven years.
Run it long enough to include a bad week
Two weeks is the standard pilot length and it is defensible, but be aware of what a two-week window excludes: a sick day, a public holiday in the delivery location, a month-end that collides with a filing deadline, a staff member on leave.
If you can extend to a full close cycle, do. If you cannot, at least ask the provider directly what their cover arrangement is and who specifically would step in — then check that person’s name appears in the SOP.
Decide the criteria before you start
Write down, before the pilot begins, what would make you say no. Not vague standards like “quality must be good”, but specific thresholds:
We will not proceed if more than two deliverables require a third round trip, if any query sits unresolved for more than 48 hours, or if our reviewer spends more than 90 minutes on a file that should take 30.
The reason to write this down in advance is that you will not want to say no at the end. You will have invested time, built relationships, and started planning around the capacity. Pre-commitment protects you from that.
What a good pilot result actually looks like
Counter-intuitively, a pilot with no problems at all is a slightly worrying result. It usually means the file was too easy or the scrutiny too heavy to be representative.
What you want to see is problems that surfaced early, were raised by the provider rather than discovered by you, and were fixed without needing a partner to intervene. That pattern — self-reported, self-corrected — is the best available predictor of what month eight will feel like.