Correction, September 2, 2026. This post argued that balance-sheet movement comes before a software purchase closely enough to rank buying intent. In September 2026 we tested that claim for the first time, against vendor adoptions that credit unions announced publicly. It did not hold. Take one credit union that announced an adoption and one that did not. In two yearly test groups, the scorecard built on the reasoning below ranked the adopter higher only 51 and 54 times out of 100 pairs. That is a coin flip. Ranking the same credit unions by asset size alone got 68 and 67. Not one of the 25 highest-ranked credit unions announced an adoption in the year that followed. The technical figures are on the disclosures page. The test sees announcements, not purchases. So a miss may be a buyer that stayed quiet, and the true hit rate can only be higher than measured. That qualifies the result. It does not rescue it. The post is left standing rather than removed. The reasoning is worth reading, and a claim published in public should be corrected in public. What replaced it does not infer intent from financials at all. It records observed relationships: which credit union runs which platform, since when, and when a contract can be shown to come up.
By the time a credit union publishes an RFP, the useful part of the sales cycle is over. A committee has already agreed there is a problem. It has already talked to two or three vendors while framing it, and already written requirements that reflect those conversations. The vendors who arrive at the RFP are competing to be the cheapest version of a solution someone else scoped.
The interesting question is what the institution looked like six to twelve months earlier. By then the problem was undeniable inside the building, but nothing had been published yet. That period leaves marks on the quarterly financials.
The general shape
Software gets bought when a capability stops keeping up with the volume flowing through it. On a balance sheet, that shows up as a divergence: one quantity growing much faster than the thing that has to service it.
That is the whole idea. It is not "this institution is big, so it buys things." That is a list sorted by assets, and every competitor already has one. The signal is a rate of change out of line with its own history and its own peers.
Five categories, five distinct divergences.
Lending and loan origination
A loan book growing considerably faster than the institution's overall balance sheet means the volume of new loans is climbing. The capacity to underwrite them is fixed. Manual processes that were adequate at the old pace stop being adequate. Inside the building, the symptom is turnaround time, not cost. Growth concentrated in a single loan category makes it worse. A specialized workflow that used to be the exception becomes the main line of business.
Collections, fraud and risk
Delinquency rising faster than the loan book is the classic tell. Collections teams are staffed for a historical delinquency rate. When the rate moves and the staffing does not, something has to give. If charge-offs (loans written off as losses) climb alongside it, the timeline shortens, because the cost of not acting becomes a number the board sees.
The important nuance: this signal fires before the losses are large. An institution already in serious credit trouble has bigger problems than software, and is a worse prospect, not a better one.
Digital banking and engagement
Here the signal is member growth pulling away from asset growth, in either direction. Members growing faster than assets means a widening base of thinner relationships. That is the digital-onboarding problem. Assets growing while membership shrinks means a more concentrated base of aging relationships. That is the retention problem. The same category of product solves both, and both are invisible if you look at either number alone.
Treasury and liquidity
The loan-to-share piece covers this in detail. Loan-to-share compares a credit union's loans with its members' deposits. Take a balance sheet crossing into the loaned-up band, with cash thinning or deposits shrinking. It has just acquired a funding problem it did not previously have to manage actively. That is when treasury workstations, liquidity forecasting and participation platforms (which let lenders sell parts of loans to each other) stop being nice-to-have.
Core and reg-tech
This one is driven by thresholds, not trends. Certain asset levels change what an institution must report and how it is supervised. Approaching one of those lines from below, or having just crossed it, creates a compliance and reporting workload that arrives on a schedule. Growth rate matters here mainly because it tells you when the line gets crossed.
Why peer-relative is not optional
The single most common failure in scoring buying intent is comparing institutions with each other instead of with their peers.
A $90M credit union and a $4B credit union are not in the same business. Their expense ratios, growth rates, delinquency levels and member behavior differ structurally, not because one is performing better. Rank them on the same absolute scale and the score becomes a stand-in for size. That is a list you could have produced with a spreadsheet sort and no model.
Scoring against peer-group and asset-tier medians fixes this. The question stops being "is this ratio high?" and becomes "is this ratio high for an institution like this one?" That is the question a salesperson is actually asking. It also surfaces the mid-size institutions that absolute ranking buries. A $300M credit union whose loan growth is in the 95th percentile of its tier is a far better call than a $5B institution sitting at its own tier's median.
The "why now" is the deliverable
A propensity score is a model's estimate of how likely an institution is to buy. A bare score of 84 is not usable on a call. What is usable is the sentence underneath it: loans up 14% year over year against 4% asset growth, delinquency up 40 basis points, both in the top tenth of the peer group.
That sentence does three things a score cannot. It tells the rep what to open with. It tells them what to ask about, so the first call is a diagnosis rather than a pitch. And it lets them drop the account quickly if the reason does not hold up. That is worth as much as a good lead. In enterprise sales, the expensive failure is not the call you skip. It is the quarter you spend on an account that was never going to move.
Any intent product that gives you a number without the reason is asking you to trust it on faith. It will be wrong often enough that faith is the wrong posture.
What this cannot see
Quarterly financials do not record when contracts expire. A core conversion (swapping out the main account system) depends more on when the current agreement ends than on anything on the balance sheet. Nor do the financials record a new CIO's mandate or a board's strategic plan. They miss a peer institution's CEO recommending a vendor at a conference. They do not see budget.
What they do see is pressure: the operational strain that builds until someone raises the issue. Pressure is what a scoring model reads well, and it is a genuinely leading indicator. It is not a purchase order, and treating a score as one is how a sales team learns to distrust the whole tool.
Used as a ranked ordering of where to spend the quarter's attention, with the reason attached so a rep can qualify out fast, it is the difference between working a territory and working a list.
Scores described here are model estimates computed from the quarterly panel. They are not investment, credit or merger advice and not a recommendation about any institution. See the disclaimer and disclosures.