Proactive help needs a reason

Consider what the person is doing before deciding whether to interrupt.

An aluminium gate admits one lime message into a tray while other messages wait, illustrating relevance, timing and context.

Proactive help needs to account for where the assistant is running: not just the screen, but the app. And the messages the person has already received from other services. Each team may see just one small offer, while the person sees another message from the same ecosystem.

So an assistant first needs to understand what you’re interested in right now, what you’re doing and whether it’s a good time to interrupt you. Only then should it decide what to show and where. An offer of “5% cashback on toilet paper” doesn’t become useful just because a product has launched that promotion.

The product’s principles, manifesto and local law set the boundaries for that choice. Can we even use sensitive data, such as health information? Within those boundaries, we need to consider both the person’s interests and our goal as a business: making money.

Working on assistants is a lot of fun. Highly recommended :)