110K support calls cut in 4 markets with a self-service product
- 110KInbound calls↓
- 11,4MNOK saved, year one
- 24,1%Answering rate↑
We stopped asking how to answer calls faster. 110K of them never came.
01the problem/
Orders outgrew the call centres.
During the pandemic, online orders grew faster than the care centres could answer. Hiring more agents would only absorb the peaks.
02research/
Listening where the calls land.
Hotjar surveys on customer service pages in four markets, plus co-listening in the care centres. Repair status beat delivery status on impact per effort: the cause was internal, so we could fix it end to end.
An impact-effort workshop with the product owner, engineers and business analysts left three ideas: a repair tracker, a new help hub and callbacks. We parked callbacks.
03my call/
A web page, not a native app.
Customers open their order or repair from an SMS link in one tap, with no install at the counter. I dropped the login wall because it lowered first use.
04timing/
Answer before they ask.
Customers called about repairs around day 5, so the tracker sends a status update on day 4. When an order is late, the page says so first. The day-4 update drove most of the drop in calls.
−110Kinbound calls in 2022, 11,4 MNOK saved in year one
05control/
Let customers decide.
When the cost estimate is ready, customers choose on the page: repair it, scrap it, or get it back with no action. It risked some repair revenue and gave customers control.
06help hub/
Self-service first.
A help hub put self-service tools first and pushed the chatbot when phone lines were busy. We tuned it every week, and launched in Sweden first so mistakes stayed small.
+24,1%answering rate on the calls that remained
07looking back/
What worked, what didn't.
- what worked/Asking which calls should never happen. The day-4 update drove most of the drop in calls.
- what didn't/We hadn't agreed up front how to count a prevented call, so proving the drop took extra work.
- what I'd change/Agree on how to count a prevented call before launch, and design shared infrastructure first, then add each market.

