ISO comp plan modeling: test a commission change before you announce it

Comp plan changes are decided on a whiteboard and discovered on a statement. An owner works out that a 55% split with a small override is affordable, announces it, and three months later learns which agents it quietly cut, which tier everyone clustered under, and what it costs now that the book has grown. The fix is not more careful whiteboarding. It is running the new plan against months that have already happened, where every merchant and every agent is real, before anyone hears about it.

This guide covers what to model, what to look for in the output, and how to roll a change out once the numbers hold.

What a backtest is

A backtest takes a proposed set of rules, applies them to past months exactly as if they had been in force, and compares the result with what was actually paid. The inputs are the residual reports you already have and the agent assignments you already use. Nothing is estimated. If the last six months of reports are on file, the last six months can be tested.

It answers four questions that a projection cannot:

1. What would the total payout have been, month by month, against what it was?
2. Which agents gain and which lose, by how much, and does anyone lose enough to leave?
3. What happens to the house margin, and is it stable across months or only good in the best one?
4. Where do the edges bite: tier thresholds, minimums, override caps, the places where a small change in production makes a large change in pay?

What to model

The common changes, and what each one tends to break:

The structures themselves, and the arithmetic of each, are in ISO commission split structures explained.

A worked example

An office pays a flat 50% and is considering 45% with a tier: 55% on any month above $8,000 of residual. The idea is to reward the biggest producers and trim the rest. Backtested over one month, illustrative figures:

Three things the whiteboard missed:

• The plan costs more, not less. The two agents above the line gained more than the three below it lost. The office designed a cut and built a raise.
• Agent C is $400 a month from a $980 raise. At $7,600 they are paid $3,420; at $8,000 they would be paid $4,400. One small merchant flips them from loser to winner. That is a retroactive cliff, and C will either find that merchant or resent the plan. Marginal tiers, where only the residual above $8,000 earns 55%, remove the cliff. See the tier section of the split structures guide for the difference.
• B sits $300 above the threshold. One lost merchant drops B under it and cuts their pay by about $1,000 in a single month. A tier measured on the trailing three months would soften that.

None of this needed judgment to find. It needed the plan applied to one real month.

What to look for in the output

• Total payout, every month tested. A plan that looks fine on the average month may be expensive in the strong ones, which are the ones the agents remember.
• Per-agent change, sorted by dollars. Anyone losing more than a few percent needs a conversation before the announcement, not after. Anyone gaining a lot is worth asking why.
• Agents near an edge. List everyone within one merchant of a threshold in either direction. That is where disputes and behaviour changes will come from.
• House margin per month, and its lowest value across the months tested.
• Merchants over 100%. With overrides and shared accounts, check that no merchant pays out more than the house received.
• Liability as the book grows. If production rises 20%, does the payout rise 20% or 35%? Tiered plans are convex; the cost grows faster than the book.

Rolling a change out

1. Backtest at least three months, ideally six, including a strong month and a weak one.
2. Fix the edges before announcing: marginal rather than retroactive tiers, trailing-period thresholds, a cap on total allocation per merchant.
3. Pick an effective date in the future, the first of a month, and apply the new rules from that month only. Never restate a month that has been paid. Rates carry dates for exactly this reason.
4. Tell the agents who lose first, one to one, with their own numbers. The backtest gives you those numbers.
5. Publish the plan in writing, with the base it applies to, the tier rules, the measurement period and the effective date. The checklist in the split structures guide covers what to include.
6. Run the first live month in parallel against the old rules, so the first statement under the new plan can be explained line by line.
7. Review after three months. Compare what the backtest predicted with what happened. The gap is your model's error, and it is usually attrition or new merchants, which a backtest cannot see.

Frequently asked questions

How do I test a commission plan change before rolling it out? Apply the proposed rules to three to six past months of real residual data, using the merchants and agent assignments as they were, and compare the result with what was actually paid. Look at total payout, per-agent change, house margin, and agents near tier thresholds.

Why did our comp plan change cost more than we expected? Usually because gains for agents above a tier threshold outweighed the cuts below it, or because a retroactive tier raised the rate on an agent's whole production rather than the amount above the line. A backtest on one real month shows this before the announcement.

Should comp plan changes apply to past months? No. Apply the new rules from a future effective date, the first of a month, and leave paid months as they were. Restating a paid month is the fastest way to lose agents' trust in every statement that follows.

How far back should a comp plan backtest go? Three months at minimum, six if you have them, and always including at least one strong and one weak month. A plan tested only on an average month hides both the cost in good months and the cliff effects in bad ones.

What is comp plan liability? The payout a plan commits you to as production grows. Flat splits grow in proportion to the book. Tiered plans grow faster, because more agents cross thresholds as they grow. Modeling the plan at 110% and 120% of current production shows the difference.

Can software backtest a comp plan? SplitRun builds a plan from a plain-language description, then runs it against past periods of real data: total payout, agent count, per-agent results, a projected liability, and flags for the problems above, before the plan is activated with an effective date. See Commission agreements and effective dates for how dated rates work once a plan is live.