Slice it right
Slice a real feature into releases and see when value becomes available. A good plan gets usable value out early and avoids long waits with nothing new.
- Write your slices
- Put them in order
- See when value arrives
“As a customer, I want to cancel an order by myself”
An online retailer sells through its website. Today, customers who want to cancel an order have to phone the contact centre, where staff handle it. The business wants self-service cancellation on the website, to cut cancellation calls and the cost of handling them.
How would you slice it?
One sticky per slice, in the order you’d release them. Say what customers can do once it’s live. Drag the ⋮⋮ handle to reorder (press and hold on a phone).
When does value arrive?
The staircase shows potential value made available: what customers could get once each slice is live, on a fixed 0–100 scale. It’s not a forecast of how many will use it.
The value scale behind the chart
Every plan is scored on the same fixed scale, so different slicing strategies can be compared. Points are an agreed assumption about increasing capability, not a measure of business impact.
| 5 | Show whether an order is cancellable | |
| 25 | Request cancellation online, processed manually | |
| 40 | Automatically cancel single-item orders | |
| 55 | Automatically cancel entire multi-item orders | |
| 80 | Cancel selected items within an order | |
| 100 | Handle all eligible orders, including complex and marketplace orders |
- A plan’s value is the highest step its released slices have reached. Skipped steps count once a higher one is live.
- Technical work, supporting pieces (confirmation emails, refunds, validation) and experiments like a fake door add no points on their own.
- The longest wait counts weeks with no new value, including at the end of the plan.
Here is the reference solution
David’s plan does the highest-value, smallest slices first, so value arrives fast and then tails off. Compare the two value paths over time.
Prototype: slices are read by a simple keyword model for now, not AI.