Half a week of two people’s time, down to a thirty-second run
Paystone is a Canadian payments company that pays around a hundred sellers every Friday. It was paying a vendor to calculate that, and two people two and a half days each to make the answer payable. This is what changed on Commish, including the errors the switch found in the old numbers.
~100
people paid out of the commission book
5 days
of staff time a week correcting it, on top of the vendor’s fee
30 sec
to run a full pay cycle now
Paying twice for the same work
Paystone sells payment terminals and software to Canadian merchants. Independent agents do the selling, some of them with sub-agents underneath, and residual earners are paid out of the same book. Around a hundred people in total.
A vendor was already being paid to calculate all of it every week. The file it produced was not one Paystone would pay from, so two people spent about two and a half days each getting it into shape. One checked every merchant line against Salesforce: duplicates, clawbacks that had added back a fee Paystone does not take back, leases that were not funded yet. The other took the agents, who wanted to know why a deal had not paid, why a number had moved, and what a deduction was for.
Five days of skilled work a week, on top of a licence fee, to produce a payout file the software was sold as producing.
Why the checking never came down
The vendor’s system calculated, but it could not show its working, and a number you cannot open is a number somebody has to verify by other means, so the checking was never a temporary measure while trust was built up. It was the job.
Correcting anything cost more time than the run itself. Asking the vendor to re-run took three or four hours; a fresh pull of the source data took most of a day. None of that fitted the week, which does not move: deal data cuts off on Monday, the payroll file has to be uploaded by six on Wednesday evening, and agents are paid on Friday.
The arrangement had been designed as an engine with a quality-control team behind it. By 2026 the engine broke often enough that the quality-control team was redoing the work instead of checking it.
“So now it’s 90% human.”
Prove it on history before touching a live payroll
Paystone moved onto Commish on 1 August 2026. Standing it up took days: their plan, their rate cards and a read-only connection to their warehouse, running pay cycles well before anyone was paid from one. The condition for switching was proof rather than speed. Commish had to reproduce pay runs Paystone had already made before it was trusted with a live one.
So the engine was pointed at a year of their actuals: 830 merchant accounts, 3,712 payout lines. A September 2025 period came back to the penny. Across seven months of 2026, compared before tax, the two systems landed within 1.5% of each other, with 239 merchant accounts matching exactly.
What the reconciliation turned up
Every gap had to be explained before go-live, one deal at a time. This is the kind of thing that came out.
Merchants paid twice
Three accounts belonging to one agent had each been paid in full on two separate pay dates inside eight weeks, roughly $4,400 between them. Nothing in the old process compared a payment against the payments that came before it.
Merchants never paid at all
Nineteen merchants appeared nowhere in the payout history and should have been picked up in earlier runs. They surfaced from the opposite direction: the engine wanted to pay something the record showed no sign of.
A fee clawed back that was never charged
When a deal reversed, the application fee was being added back into the clawback, so agents were charged for something Paystone does not take back. Accounting had been catching it by hand almost every week for months. On Commish it is one setting on the plan and it applies to every agent at once.
First terminals paid like additional ones
A leased terminal that was the first on an account was being tagged as an additional terminal, which pays a lower multiplier. It had been invisible until agents could see the multiplier applied to their own deals. One agent brought three merchant accounts in a single week, with the arithmetic attached.
A fee an agent had been absorbing
One agent’s deals kept showing a set-up fee deducted from his payout. He was certain he had charged the merchant for it. The signed merchant applications said otherwise: he had been filling the form in wrong and eating the cost himself on every deal. It had never been laid out plainly enough for him to notice.
Two of the five were ours and shipped as product changes. The other three are records inside Paystone’s own systems, and putting those right is Paystone’s work. What changed is that they surface in the week they happen rather than never.
What the week looks like now
Running a cycle
Before
Request it, then wait. A fresh pull of the data was half a day
On Commish
Thirty seconds to a minute, run by Paystone whenever they want it
Making a correction
Before
Re-run at the vendor, back in three or four hours, reviewed again from the top
On Commish
Fix the record, re-run, look at it. The fix and the proof arrive together
Validating the run
Before
Every merchant line by hand, against Salesforce and against last week
On Commish
Read what the run flagged and decide
The weekly cost of that
Before
Two people, about two and a half days each, on top of the fee
On Commish
The same two people, on the exceptions
When an agent asks why
Before
Someone reconstructs the answer from Salesforce and memory
On Commish
The agent opens the deal and reads the arithmetic
The evidence
Before
Whatever somebody wrote down
On Commish
Every payout traces to the deal, the rule that fired, and each step of the sum
“This is probably the first Tuesday in a very long time that I’ve seen you smiling and not stressed about commissions.”
“I love how quick this stuff is. It honestly makes it so much easier. You have no idea.”
“Less issues than before, and we’re not even through week one.”
What this page does not claim
A case study that only carries the good half is worth nothing to someone deciding whether to trust the rest of it.
Commish did not make their data correct
A commission engine pays what the source systems say. Where those say the wrong thing the payout is wrong, and what Commish adds is that you find out in the week rather than at year end. Several of the corrections above are still working their way through Paystone’s records.
Not all hundred are on it yet
Upfront commission moved first and is live. Residual earners and sub-agents are the rest of the hundred, and that side is still being mapped. The book is a hundred people; the migration is not finished.
Two numbers are missing from this page
A case study usually quotes disputes per cycle before and after, and the full previous cost including the licence fee and the internal time. Nobody has measured either at Paystone, so neither is estimated here.
Week one had a list
Clawbacks including a fee they should not have, two headings the wrong way round on the agent screen, agents rendering as record IDs instead of names. All of it was fixed inside the week.
Bring a year
of your actuals
Commish will run them before you commit to anything. If the numbers disagree with what you paid, you will know exactly where, and so will we.