Four agents kept notes. All four chose a local bank.

The choices

In the persistent-notes group, two agents recommended Frost Bank and two recommended Austin Telco Federal Credit Union.

In the twelve comparison records, nine primary recommendations were Capital One 360 Checking. The remaining three were Ally, Charles Schwab and SoFi. Two comparison records also named local institutions among their alternatives.

This is not a story of “local banks versus banks without branches.” Several comparison recommendations explicitly discussed branch access. The distinction here is the actual institutions selected, not an invented division between human service and digital-only service.

Why it is interesting

The accompanying study account describes agents carrying their own notes across four earlier consumer decisions before the final banking task. The later recommendation was therefore not necessarily based only on the current request and the pages read during that task.

For an enterprise, that raises a different question from whether its website is clear. What preferences, concerns or earlier experiences might the customer’s agent bring into the choice?

For a CX operator, the customer relationship may no longer begin with an empty conversation history. Some of the context may have been assembled elsewhere, by software working for the customer.

That does not mean the enterprise should have access to a customer’s private notes. It means the resulting requirements and choices may carry a history the enterprise did not create.

What this does—and does not—show

The choices differ in this small comparison. They are consistent with persistent context influencing a later recommendation.

They do not establish how often that happens, which particular note changed a choice, or a population-wide effect of memory. The released records do not contain all original prompts and configuration details needed to independently reconstruct a matched causal comparison.

This is also a different study from our earlier six-brand memory pilot, where selection counts did not differ between conditions. Both results remain visible because they asked different questions.

Open the decisions

The evidence view lists all sixteen final banking recommendations, with primary choices and any local alternatives distinguished. Four final decisions came from a twenty-record persistent-notes dataset spanning earlier tasks; the twelve comparison banking decisions are part of the separate forty-three-run dataset. These numbers must not be added together as sixteen new journeys on top of both datasets.