Deterministic Commission Arithmetic: Why Advisors Should Not Accept an Estimate

Software that serves an insurance advisor deals in two kinds of number, and the difference between them is not a matter of degree.

The first kind is genuinely uncertain. Whether a client renews. Whether a household lapses in March. What next year's motor premium will be after the insurer reprices. These are predictions: they depend on facts that have not happened yet, and the honest output is a range with a stated basis. A tool offering a renewal-probability estimate is doing something reasonable.

The second kind is not uncertain at all. Your remuneration on a policy you have already placed is a function of inputs that all exist: a rate written in a contract you signed, a base defined in that contract, a premium printed on a schedule that has been issued, and a set of adjustments (cancellation, short period, endorsement) that are either events that occurred or events that did not. Nothing about it is unknown. It is arithmetic, in the strict sense that the same inputs produce the same output every time, and there is exactly one correct answer.

A tool returning an estimate for the second kind has made a category error, and not a harmless one. It has taken a question with a right answer and returned something that cannot be checked, argued from, or reconciled. Approximation is a legitimate response to uncertainty and an illegitimate response to arithmetic. When a number is computable, computing it is not an optimisation. It is the job.

This matters more for a solo advisor than for anyone else in the chain, because the advisor is the only party with no finance function. An insurer has one. An intermediary has one. A POSP running 200 policies from a phone has themselves, an evening, and whatever the software chose to show them.

The reason commission can be computed rather than estimated is structural, and worth being precise about.

A POSP is remunerated by the entity that engages them, insurer or intermediary, under the contract of engagement. You are not an independent commission earner facing the insurer across a market. Where an intermediary engages you, the generally understood market structure is that the insurer pays commission to the intermediary and the intermediary then remunerates you under your contract. That is how the channel is commonly described rather than something to read out of a regulation, and the consequence is exact: the rate flowing insurer to principal and the rate flowing principal to you are two different numbers, and only the second is yours.

Which is precisely why your number is determinable. It is not set by a market, a benchmark, or what other advisors report on a WhatsApp group. It is set by a document with your signature on it. A rate that had to be estimated would be one that floats. Yours does not float. It is written down.

So the usual assumption is backwards. Advisors treat the contract as paperwork and the payout as the real thing that mysteriously arrives. The contract is the definition, and the payout is a claim about it. If the two disagree, the contract is right and the payout is wrong, and you are the only person positioned to notice.

The most common approximation in advisor tooling is also the most defensible-looking. The software does not have your contracted rate, so it applies a typical rate for the product and multiplies it by the premium the client paid. Two errors compound here.

The first is the rate. There is no fixed table to look your rate up in; since product-wise caps went in April 2023 it follows each insurer's board-approved policy, so it varies by insurer, by product and over time. A market-typical rate is therefore not an approximation of your rate. It is a different quantity that happens to share a unit. Averages describe populations. You are not a population.

The second is the base, and it is bigger. What the client transferred is almost never the figure your rate applies to. A motor policy bundles several priced elements plus tax, and they do not all carry remuneration alike. Multiplying one rate by the bundle total is not a rough version of the right calculation. It is a different calculation, wrong by a margin that moves with the composition of each policy, which means the error never averages out across a book. It tracks your product mix, so the advisor selling the most add-ons is the advisor whose estimate drifts furthest.

The tell is easy. If a tool shows you a payout without ever having asked what your contract says, it did not compute your payout. It computed somebody's.

A payout figure that ignores what is deducted from it is not answering the question the advisor asked. The advisor wants to know what reaches the bank account, and estimators routinely stop one step short.

Deductions are not a footnote to the calculation. They are part of it, and they are knowable. What an advisor needs on screen is the figure before deduction, each deduction named and shown on its own line, and the figure after. A single number, presented alone, hides which of the three it is. That ambiguity is where an advisor stops being able to check anything, because the amount they are assessed on and the amount that lands are different quantities and only one of them will appear on the paperwork their principal keeps.

The indirect tax position on advisor remuneration deserves more caution than it usually gets, from software especially. It turns on the structure of your engagement and on who is accounting for what, so it is a question for your principal and your accountant, answered in writing. A tool that silently bakes in an assumption has buried a tax position inside an arithmetic feature, which is the wrong place for one to live and the hardest place to find it.

The rule falling out of both: a deterministic calculator treats every deduction as a named, sourced input, never a constant it inherited from somewhere. Where it does not know the treatment, the defensible move is to compute what it can exactly, mark the rest unknown, and say so. That is still arithmetic. An unknown that is labelled leaves the advisor able to finish the calculation by hand. An unknown that has been quietly guessed at leaves them unable to detect that anything happened at all.

The largest gap between an estimate and the truth is not in the rate at all. It is that estimators model a policy as a single event that happened once, and policies do not behave that way.

A book of 200 retail policies generates a steady stream of events after issuance, each of which changes the payout deterministically:

• Cancellation. The client cancels, the insurer refunds premium, your payout reverses. Legitimate and expected. What must be computed is that it reverses once, against the right policy. A reversal on two consecutive statements is an ordinary systems error, and entirely your job to notice.
• Short-period cover. A policy issued for less than a full term carries a proportionate premium and therefore a proportionate payout. The proportion is defined, not estimated.
• Endorsements. A mid-term change can generate additional premium. Additional premium that carries remuneration and produces none is a gap an estimator built on the issuance figure structurally cannot see, because it never looked at the policy again.
• Rate changes. Because rates move with board decisions, terms can change mid-year, and a payout computed against the wrong version of the rate is wrong in a way that looks exactly like being correct.

An estimate takes the premium at issuance, applies a rate, and stops. Every event above happens after that moment. So it is not merely imprecise. It is answering a question about a policy that no longer exists in the state it was asked about, and it drifts further from the truth the longer the policy lives, which is the opposite of how an advisor would like information to age.

Deterministic arithmetic handles this the only way it can be. The payout is not computed once at sale. It is recomputed from the current state of the policy every time that state changes, with every adjustment traceable to the event that caused it.

Rounding gets waived through because a rupee is not worth arguing about. That is right about the rupee and wrong about the rounding.

The question is not the size of the error. It is whether the error is reproducible. A figure that rounds at a different step from the one your principal rounds at will differ from the statement by a small amount on almost every line. Small, and constant, and therefore fatal to the only thing your record was for. No threshold separates that noise from a real short-payment, because the noise is on every line and the short-payment is on one of them. The genuine gap hides inside a fog you generated yourself.

Order of operations matters more than precision. Rounding each component before summing gives a different total from summing and then rounding, and both are defensible in isolation. Only one matches what your principal did. The arithmetic is deterministic only if the sequence is fixed, and it has to be the same sequence, not merely a consistent one.

Which yields the standard worth insisting on. Your computed figure should match the statement exactly, to the rupee, on every line where nothing is wrong. That sounds like perfectionism and is the opposite: it is the only configuration in which a disagreement means something. When every line matches, a line that does not is a signal, visible instantly. When lines routinely differ by small amounts, you have no signal at all, and the two policies never credited in October sit inside the noise until the financial year closes over them.

Put the pieces together and the requirement for any tool that touches an advisor's commission is short, and it is about arithmetic rather than features.

1. It knows your rate and your base, because you gave them to it from your contract, not because it inferred them from a market average.
2. It computes the expected payout at the point of sale, before the statement arrives. A number written down after you have seen what you were paid is not an expectation. It is a rationalisation.
3. It recomputes on every event that changes the state of the policy: cancellation, short period, endorsement, reversal, rate change with an effective date.
4. It shows the working. Rate, base, each component, each adjustment, the gross, each deduction, the net. A single figure with no derivation cannot be argued with, and arguing is the entire point.
5. It rounds where your principal rounds, so a difference is information rather than noise.
6. It says "I do not know" where an input is missing rather than filling the hole with a plausible constant.

A tool meeting that standard produces a number that sits beside the statement line and either matches or does not. That binary converts a monthly evening of squinting into a short list of specific policy numbers with specific expected amounts, which is the only kind of message a principal can act on. Checking the statement you received is a separate discipline with its own method. This is the input to it: without a computed expectation there is nothing to check against, and reconciliation collapses into reading.

One caveat about the future, stated narrowly. IRDAI is preparing an overhaul of commission rules, and as of the date of this post the consultation paper has not been published; everything reported about its likely contents is a proposal and none of it is in force.

But notice what a trail structure would do to the arithmetic, if one ever arrived. A payout concentrated at sale is one calculation. A payout spread across a policy life is a schedule of calculations, each with its own trigger and conditions, computed over years. An advisor who tolerates an approximation for one number today would be tolerating a stream of them, and unverifiable errors compounding across a decade stop being a rounding problem. They become the payout.

None of which is a reason to wait. Determinism is not preparation for a rule that may never exist. It is the difference between knowing and assuming about money that moves this month.