Budget allocator template for Fintech

The budget allocator with three columns a regulated financial product needs: which product a share is for, which market it may be spent in, and the approval rate the share's economics assume.

Everything else is the general allocator, unchanged. One row per destination, a basis for every share, a floor and a ceiling, and a reallocation rule written before it is needed.

Use it when the cost of an acquisition depends on a rate you do not control. A split argued on cost per application will fund whichever row attracts the most people, including the ones underwriting will decline, and it will look efficient the whole time.

Use it again whenever the approval rate moves, because every row's economics were written against the old one.

In most categories an allocator is argued from advertising evidence: what each destination produced, at what cost, over what window. For a regulated financial product there is a second number sitting underneath all of it, and it belongs to somebody else.

The approval rate is the proportion of applications that become customers. It is set by underwriting, it moves for reasons that have nothing to do with marketing, and it changes the economics of every row at once. A share that looks generous at one rate is tight at another, and nothing visible in an ad account tells you which world you are in.

So the assumed rate goes in the row, with the window it was read over. It does two things. It makes the row's economics explicit, so a reader can see what the share depends on. And when the rate moves, it turns the re-planning job into a filter rather than a rebuild.

The most common expensive mistake in this category is allocating on cost per application.

Applications are cheap to generate and easy to generate badly. A creative that promises more than the product delivers, or that reaches people the product is not for, will produce a very low cost per application and a very low approval rate, and the two effects cancel out in a way that is invisible unless somebody joins the numbers together.

So the basis for a performance row should be a funded account or an approved application, and the evidence column should name the window over which that was read. It is a slower measurement — funded accounts lag applications — and that is precisely why the cadence column matters: a weekly reallocation rule on a metric that takes three weeks to settle is a rule that will chase incomplete data.

The product column sets the economics. Approval rates, funding rates and margins differ enough between an account, a card and a loan that a blended share describes none of them.

The jurisdiction column sets the permission, and it is not a formality in an allocator. A share assigned to a market where the product is not authorised is money that will either go unspent or be spent somewhere it should not have been, and the second is a problem that no performance number makes smaller. One market per row, and where the same product runs in two markets with different wording, two rows.

Floors matter here for an unusual reason: the metrics that matter are ratios of relatively rare events.

A row funded below its floor produces a handful of applications and one or two funded accounts, and the approval rate calculated from that sample is noise. It will swing wildly month to month, and whichever direction it swings will be used as an argument. Setting a floor from the volume needed to read the rate — rather than from what is left over after the big rows are funded — is the only way an allocator in this category produces information about itself.

Ceilings are worth filling in wherever they are known, because demand for a financial product is often bounded in ways that are not obvious.

Search demand for a category is finite in a given market and a given week. A comparison surface has a fixed amount of inventory. An audience defined by eligibility is smaller than the platform's estimate suggests. Money poured past those points buys frequency and duplicate applications rather than new customers.

Where a ceiling is unknown, write unknown rather than leaving the cell empty. In a sheet, an empty cell reads as unlimited, and that is the interpretation that gets acted on at the exact moment somebody is moving a large amount quickly.

The reason the assumed rate is a column rather than a footnote is the day it changes.

When underwriting tightens or loosens, every row whose economics assumed the old rate needs re-arguing, and the allocator is the list. Filter on the assumed rate, re-read the shares against the new one, and record the change with its date in the notes. Rows that were marginal at the old rate are often the first to fall below their floor at the new one, and that is a decision worth making deliberately rather than discovering through a month of worse results.

Almost every account in this category has a retargeting row, and it is almost always the cheapest row in the sheet. That makes it the row most likely to be given more money and least likely to be examined.

Two things belong in its cells. The first is a ceiling, because the audience is finite by construction: it is made of people who already started something, and it cannot absorb an arbitrary amount. The second is a note about what may be said to it, since an audience defined by an unfinished application is defined by a financial circumstance, and the wording that reaches it is narrower than the wording that reaches a cold audience.

Neither of those is an advertising judgement, which is exactly why they belong in a document somebody re-reads before moving money.

Every row names one product and one jurisdiction, and no share is assigned to a market the product is not authorised in. Every row records the approval rate its economics assume, with the window it was read over. Shares sum to one whole per currency. Every row has a floor set from the volume needed to read a rate, not from what was left over. Every row carries a reallocation rule written while nothing was going wrong. And every assumption is labelled as one, in words, in the notes.