The Tsunami Is Not Coming from Inside the Bank

A couple in Winterthur wants to buy a house. They earn well, have saved diligently, and qualify comfortably. Ten years ago, they would have walked into their Hausbank and started the conversation there. Five years ago, they might have compared three offers. Today, they ask their AI agent to find the best mortgage in Switzerland.

The agent does not walk into a bank. It contacts thirty of them. Simultaneously. Each request is perfectly formatted. Each carries the same financial profile. The couple's effort is unchanged. They are still buying one house.

But before we get to the agent, it is worth understanding the funnel that already exists. Because the waste in the current model is larger than most banks acknowledge.

The funnel that already leaks

Switzerland originates roughly 85'000 new mortgages per year, based on SNB survey data (H1 2025: 44'607 new mortgage loans, annualised). Of those, approximately half are for owner-occupied residential property. The rest are refinancing, rental properties, and commercial lending.

Meanwhile, roughly 50'000 to 60'000 residential properties change hands per year (BFS and Wüest Partner estimates, with recent volumes at the lower end due to interest rate effects).

Here is the part that matters: each of those 50'000 transactions generates multiple mortgage applications. The average buyer contacts three to five banks. Many people who request mortgage quotes never buy at all. They are searching, comparing, testing affordability, or losing out in competitive bidding processes. Online platforms report quote-to-close ratios of ten to one or higher.

The illustrative funnel looks like this:

Read that funnel from bottom to top. For every property that sells, banks collectively process four to six formal applications. Most of that work produces no revenue. The CHF 1'500 processing cost is spent on applications from buyers who chose a different bank, buyers who never purchased, and seekers who were never serious.

This is the baseline. The system already runs at a conversion rate well below one in three. And it works only because the volume is manageable at human speed.

What the agent does to this funnel

Now give every property seeker an AI agent.

The agent does not wait until the buyer has found a specific house. It queries banks as part of the search process, for every property the buyer is considering. Where a human would request one mortgage quote per serious candidate property, an agent requests quotes for ten. And where a human would contact three banks per property, the agent contacts thirty.

The middle of the funnel explodes. The bottom stays the same.

The number of houses sold has not changed. The number of applications has multiplied. Banks process ten times the volume for the same revenue pool.

The arithmetic

Follow the numbers for a single Swiss bank.

Today: Based on industry estimates, processing one mortgage application costs roughly CHF 1'500. With a conversion rate of roughly one in four (accounting for the existing funnel leakage), the expected value per application is around CHF 5'000. The unit economics are manageable.

In the agentic world: The conversion rate drops to one in forty or worse. The cost per application has not changed. Expected value per application: CHF 500. Every application you process but do not win costs you CHF 1'000.

To break even at the new conversion rate, processing cost must fall from CHF 1'500 to under CHF 500. A reduction of more than 65 percent. Through a structurally different process, not through incremental efficiency.

Most banks cannot achieve this by optimising what they already do. The economics demand automation.

The three-day bank and the three-hour bank

There is a second dynamic.

Picture the couple in Winterthur again. Their agent has sent thirty requests. Within two hours, eight banks have returned indicative offers. By evening, three have provided binding terms. The couple's agent ranks them, highlights the trade-offs, and recommends a shortlist.

Bank number twenty-three responds on Wednesday. Its offer is competitive. It is also irrelevant. The couple signed on Monday.

The slow bank did the work. It ran the credit check. It prepared the offer. It spent the CHF 1'500. And it lost, because it arrived after the decision was made.

Today, response time is a service metric. In an agentic market, it is the conversion driver. The bank that responds while the customer is still engaged wins disproportionately. The bank that responds after has performed free labour.

Free price discovery

In the old model, every bank that submitted an offer had a reasonable chance of winning. The customer compared three, maybe four. The effort of applying created a natural floor on conversion rates.

In the agentic model, slow banks become free price discovery tools. The customer's agent collects their offers to benchmark against the offers it has already received. The bank provides data. Someone else provides the mortgage.

The work is identical. The cost is identical. The revenue is zero.

Beyond mortgages

The mortgage example is vivid, but the pattern applies wherever a customer's agent can query multiple banks simultaneously.

Corporate lending. Foreign exchange. Trade finance. Insurance. Any product where comparison was previously limited by the customer's time and effort will see the same dynamics: volume explosion, conversion rate compression, speed premium, and the inversion of who captures value.

The agentic target operating model is a structural response to a structural change in how demand reaches the bank. Customers are not changing what they want. They are changing how they ask for it.

The moat and the deadline

The institutions that automate first will not just survive the volume. They will use it.

Ten times more inbound means ten times more data. Ten times more pricing signal. Ten times more opportunity to learn what customers actually want and what they will actually accept. For the banks that can process at low marginal cost and respond in hours, with escalation designed in: the tsunami is a moat.

The couple in Winterthur does not care about any of this. They want a good rate, a fast answer, and a bank they trust. Their agent will find all three.

Your bank is either one of them, or the one that spent CHF 1'500 preparing an offer nobody read.