The platform sits inside 130+ banking apps, reaching around 150 million customers with over a billion insights a month. The brief: pick a financial behavior worth improving, and design a proactive in-app experience to move it — idea to launch.
I chose building a cushion — Step 2 on the seven-step financial-stability ladder. Almost every savings feature already knows who should save and how much. The part nobody gets right is when to ask. Ask on the wrong day and a “you could save ₪300” notification lands right before rent clears, feels tone-deaf, and quietly trains the customer to ignore the bank. So I built the feature around timing.
Generic savings notifications fail because they ignore timing.
A cushion isn't built from willpower — it's built from surplus, and surplus only exists on certain days. Most engagement tools treat the month as flat and fire prompts on a schedule. The result is a stream of well-meaning notifications that arrive when the account is at its tightest.
- Wrong moment — asking to save before the mortgage and credit-card bills clear
- No account of anomalies — a one-off tight month gets the same nag as a healthy one
- No account of real distress — someone in structural deficit is told to “save more”
- Trust erosion — every mistimed prompt makes the next insight easier to dismiss
Every customer has a monthly rhythm — and a quiet window inside it.
After three or more months of transaction history, the system maps a customer's recurring obligations and finds the day the last big one clears. In the Israeli cadence, standing orders fire on days 1–3 and credit cards debit on the 10th — so the Quiet Window opens on day 10+, within 24–48 hours of the card bill clearing, once a 14-day look-ahead confirms nothing large is coming. It's a detected moment, not a fixed end-of-month date.
The feature then reads one of three states and responds differently to each — because the same account behaves like three different people across a year.
The window is open
More than ₪300 is genuinely free after all bills clear. The system fires a single, well-timed notification: here's what you could move toward a goal, right now.
A one-off anomaly
This month is unusually thin but not a pattern. Instead of nagging, the system reassures — no ask, just acknowledgement — and waits for the next window.
Recurring negative
The numbers don't close month after month. Saving is the wrong advice, so the system routes the customer back to Step 1, Know Your Money, instead of asking for more.
Guidance that adapts
Deficit users coached back to Step 1 are moved forward again the moment their window reopens — a feedback loop between the two ladder steps rather than a dead end.
One notification, then a five-step plan the customer controls.
When the window opens in a surplus month, the customer gets one in-app notification. Tapping it starts a short wizard: what we found this month → pick a goal → set an amount → choose a pace (Conservative, Balanced, or Ambitious) → fine-tune with a live slider that always keeps a ₪300 comfort buffer reserved. The alternate path is a one-tap “Not now” that dismisses without penalty and simply waits for the next window.
The whole flow is built in a phone prototype — goal grid, plan tiers, recurring-save toggle, and the safety disclaimer that saving only happens after the balance is real. A bank can configure the goal types, plan rates, and buffer floor; the timing logic and safety checks stay locked.
Guidance before advice — GenAI explains, it doesn't decide.
This is people's money, so the decision logic stays rule-governed and explainable: whether the window is open, which state you're in, and how much is safe to move are all deterministic. AI does the part it's actually good at — detecting the rhythm from messy transaction data, and putting the reason for each notification into plain language the customer trusts. No black box sits between a customer and their balance.
Everyone knows how much to save. The hard part is knowing when to ask.