How banks and credit unions test a new deposit product before launch

Testing a deposit product before launch means comparing several candidate designs against customer fit, your existing portfolio and competitor offers, modelling the economics of each one, and finding the input values where your preferred design stops being the best one. The output is a specification your teams can build from, with the evidence behind it kept attached.

Last updated: October 20, 2026

Why test a deposit product before you build it?

Building the product is the expensive step. Configuration work, origination changes and core work all follow from the product decision, so a weak design costs real money to reverse once work has started. Deposits are a practical place to run the full testing process. Deposit economics are legible, competitor offers are observable in the open market, pricing has a material effect on the result, and funding matters to every bank and credit union.

There is a second reason, and it is recent. Curinos reports that a TD Bank study found 55% of Americans consult AI about their financial well-being, up from about 10% the year before. Consumers are asking AI assistants where to put their money. How a product reads to those assistants is now part of whether the product is considered at all. Curinos also warns that AI answers tend to flatten value propositions into lists ranked by rate, which makes a product that competes on more than its headline rate harder to surface. Source: Curinos.

A team that tests the design before pricing it can see which structure serves the objective. Pricing work then goes into the design that earned it.

What are the steps to test a new deposit product?

1. Define the commercial objective. Write the outcome, the customer problem, the constraints and the current product position in plain terms. An objective that can be tested sounds like this: increase stable retail deposits from existing households, with no reliance on a short-term headline-rate promotion, and with no material rise in expected annual funding cost. "Grow deposits" on its own is too loose to test.

2. Generate credible alternatives. Create designs that differ in the proposition, the price structure, the benefits, the eligibility rules and the target segment. Rate levels alone do not make designs different from each other. A test with one idea and no alternatives can only confirm the idea the team started with.

3. Test each proposition. Compare every design against customer fit, so you can see who would respond and who would stay away. Compare it against your existing portfolio, so you can see where the new product draws balances from. Compare it against observable competitor offers, with the source and the date recorded beside each one.

4. Model the economics. Carry margin and funding value through each design, and include acquisition cost, expected attrition after any promotional period ends, cannibalisation of existing products, and relevant capital use. A design that leads on rate alone often changes rank once moved money and runoff are counted.

5. Find what breaks the case. Move the uncertain inputs and identify the values where the recommendation changes. Relationship adoption, retention, cannibalisation and the value of alternative funding are usually the inputs worth moving first. A recommendation that holds across most of the plausible range deserves more confidence than one that flips on a small change.

6. Select and optimise. Choose the strongest configuration and write the trade-offs down beside the choice. The losing designs stay in the record with the reasons they lost, so the decision can be revisited when conditions change.

7. Produce the specification. Turn the selected design into product fields, rules, economics and evidence, in the form your teams use for their delivery work. The specification carries the reasoning with it, so a reader can follow the decision back to its inputs.

What does a worked example look like?

Northline Bank is fictional. This example is illustrative. It shows how four deposit designs are compared against one objective.

Objective: increase stable retail deposits from existing households, with no reliance on a short-term headline-rate promotion, and with no material increase in expected annual funding cost.

Why B is selected: at the modelled target volume, B provides the strongest balance of stable funding, expected cost and retention. The recommendation changes if relationship adoption, cannibalisation, retention or the value of alternative funding move beyond their tested breakpoints.

The selected configuration becomes the specification:

These values are illustrative. Real values come from an institution's own objective, inputs, constraints and evidence.

How do you know the recommendation will hold?

Three checks keep a result defensible. Each one is cheap to run while the designs are still on paper, and each one is expensive to skip.

Breakpoints. Identify the input values where another design becomes stronger. A team that knows its breakpoints knows in advance which number to watch after launch.

Sensitivity. Move the uncertain inputs together and report how often the same answer survives inside their defined ranges. The output states the range it was tested across, so a reader can judge how much weight the answer carries.

Evidence labels. Classify each figure as a fact, an assumption, an estimate or a calculation. Keep competitor evidence with its source, its date and a confidence note. Where an answer cannot be computed from the information provided, the missing input is named in the output and the gap stays visible, rather than being filled with a quiet guess.

Keep each run's inputs and outputs on file, so versions can be compared and the decision can be reconstructed later by someone who was not in the room.

Who needs to be involved?

Division heads with profit responsibility own the decision, and their finance leads own the numbers behind it. Product, strategy and analytics teams run the work day to day: they build the candidate designs, assemble the evidence and maintain the record. Retail leaders who will carry the product to market read the same record as everyone else, which cuts the rework that comes from each team rebuilding the analysis in its own format.

Does testing replace the product decision?

Testing produces the analysis and the specification. People at the institution decide what proceeds, set the final pricing and own the result. Nothing is launched, priced or changed on an institution's behalf by the testing process, and the record stays available for the people who revisit the decision later.

How does Deepify approach this?

Deepify is software that runs this process. It generates alternatives, tests them against market evidence and competitors, models the economics, challenges the assumptions and produces the build-ready specification. Teams use it themselves, with always-on access to the platform, and they receive the verdict in writing with the math behind it.

For evidence, Deepify asks live, web-connected assistants the questions customers actually ask, phrased the way a person types them. Every answer is kept. Each projection carries what it rests on, whether that is a live demand signal, a web-verified comparable or reasoning, and it carries a falsifier: the specific thing that would prove it wrong. Confidence on a projection is capped in code at what the evidence gathered in the run supports, and the cap can only lower a number. The figures clients act on are produced by deterministic code. Language models assemble evidence, and no language model sits inside the arithmetic. The full method is set out in How Deepify works.

On data: each institution works in its own workspace, kept apart from every other firm. Material is stored in Canada and is visible only to your organisation. Deepify receives the decision under test and any documents you choose to attach to a run. It does not receive or hold your transaction data, your customer data or your loyalty data.

How can your team run a first test?

Two set offers are available, priced in Canadian dollars.

Decision Test: $7,500 CAD. Two or three related investments are tested together. Half the fee is due at signing and half on delivery. The full $7,500 is credited toward a Move Set Test if your firm expands within 30 days.

Move Set Test: $40,000 CAD. Up to eight related moves in one division are tested together. The full $40,000 is credited toward year one. A written result is agreed at signing, before the test starts.

Both offers begin with a 90-minute shaping session, where each move is put on a one-page form. The run itself takes hours. A results meeting is held within five working days of the inputs being complete and checked, and any move that lacks the facts for a sound number is reported as not priced, with the missing fact named.

Book your walkthrough: awatson@deepify.cloud

Frequently asked questions