Sometime in the next six to eight weeks, someone at your institution will cite a vendor’s conversion figure in a budget conversation, and someone else will answer with the sentence that kills vendor figures everywhere: of course, the vendor says that. The objection is fair. Vendors select their best results. No cited study, no case example, no benchmark survives it, because none of it came from your institution.
One kind of number is immune. The number your own institution produced, from its own traffic, last quarter. Finding it takes weeks, which is why the search starts now, while the 2027 conversations are still forming.
The Five Questions
Ask five questions about last quarter, all of them about activity on your website. How many visits touched our rates. How many visitors used a calculator. How many started an application. How many finished one. Where did the rest go.
These arrive as questions instead of metrics because the answers live in different systems that different people own. The first two sit in your website analytics, if anyone set them up to be counted. The application questions sit somewhere else entirely, in the origination platforms, and at most institutions that word is plural. One platform opens deposit accounts. Another handles consumer lending. Mortgages run on a third. Business lending and credit cards often have their own. Each platform produces its own reporting, each report has its own owner, and your website analytics cannot see into any of them.
Notice what that structure means before a single number comes back. “How many applications started last quarter” reads as one question and behaves as four or five, asked separately to four or five systems. Nobody at the institution owns the assembled answer. Running the exercise produces something that rarely exists in one place: a single view of last quarter’s intent, from the rates page through a completed application, across every product line.
Most institutions run their web analytics in the default configuration, where calculator use, application funnels, and visitor paths were never instrumented. Coming up empty on several of the five is the expected result. So is partial coverage, the institution that can answer for consumer lending but has nothing on mortgage or can answer for deposits but has nothing on cards. Across the institutions where we have watched this exercise run, most can answer two of the five. The point of asking all five is finding out which two are yours.
“We had forty thousand visits to our loan pages last quarter, and we cannot tell you how many of those people were shopping.”
Both Results Win
Suppose the answers come back. You walk into budget season holding the one number no committee can wave off, because your institution produced it from its own traffic. When the conversation turns to what lives on the website, you are the only person in the room quoting your own quarter rather than someone else’s case study.
Suppose they mostly do not come back. The gaps compress into one sentence that carries the entire case: we had forty thousand visits to our loan pages last quarter, and we cannot tell you how many of those people were shopping. Swap in your own traffic count, which is the one answer nearly every institution can produce. One sentence, built from your own numbers, demonstrating the exact condition a behavioral intelligence request exists to close.
Even a complete set of answers stops at the same place, and if you ran the exercise, you have already felt it. Counting is the ceiling. Your analytics can report how many visits touched the rates page last quarter. It cannot say which of those visitors was shopping a loan and which was passing time before a meeting, and it cannot respond to either one while they are still on the site. The distance between what your systems count and what they cannot read is the exact thing a 2027 behavioral intelligence request funds. You arrive at that conclusion holding your own evidence for it, which no vendor deck can supply.
What the Gap Was Worth Somewhere Real
One client of ours, a large Southern California credit union, lives with the traffic profile this exercise usually reveals. Most visits come from members logging in, handling their business, and leaving. The loan shoppers were inside that stream the whole time, invisible because they looked like everyone else. When the site began reading behavior during the visit, the behaviors the five questions ask about turned out to be the strongest signals of readiness: visitors who used a financial calculator completed applications at five to nearly ten times the site baseline, and rate shoppers at 3.6 times. Visits where the experience responded to that intent converted at 0.37 percent against 0.17 percent for the standard experience and outperformed in every measured month across nine months.
The caveat belongs next to the numbers. Results vary, personalization is never the only variable in play, and across our client base the magnitude moves while the direction holds. Read these figures as an illustration from one real institution. Your equivalent starts with your own five answers.
This Week
Write the five questions down. Route each one to whoever owns the system that holds its answer, however your institution is organized: the website questions to whoever runs your analytics, the application questions to whoever owns reporting on each origination platform and expect that to be more than one person. The weeks this takes are mostly the weeks of finding those owners and waiting on their answers, which is the practical reason it starts now. Budget conversations peak in the coming month. Either result arms you, which makes this the rare exercise that cannot fail.