Predictive chat
The app opens a conversation, asks what the person wants to do and already proposes the most likely answer. I imagined it for Tele Sena, but it fits any app.
How the idea started
When someone opens an app, the app already knows a lot: what the person did last time, whether they bought something, whether they left something unfinished. Even so, in most apps, the person is the one who has to work out what to do next.
This idea came from something I wanted to do on a previous project. Someone who buys a product and returns to the app usually wants the same thing: to know how it ended. With Tele Sena, the question is "did I win?". I wanted people to open the app and find the answer already there, before they asked. The idea is to surprise people with how efficient it is. This text is my own proposal: it never went live and has no link to the company mentioned.
The idea
I call it predictive chat: when the app opens, it starts a short conversation in a bottom sheet (a window that slides up from the bottom) and proposes the most likely next step in the journey. The sheet looks like a chat: the app speaks on the left, in bubbles, and the person's replies sit on the right, also in bubbles, ready to tap. The main reply is larger and in the highlight color; the others sit right below, more discreet. In each session, the app proposes the next step once, whatever situation they are in.
Two moments, in the Tele Sena example:
- A new draw. If the person has no ticket for the draw that just started, the next step is buying. The app says the draw has started, asks whether they want to repeat their last purchase, for example two regular tickets and one weekly, and shows the total before any tap. The reply can already sit one step ahead: "Yes, pay with Pix". The tap leads to the payment confirmation screen, with the amount in plain view, and the purchase is only completed on the next tap. Anyone who prefers another method picks "Yes, pay another way".
- After a purchase. The last thing the person did was buy. The next time they open the app, it presumes what comes next: checking whether they won. It asks whether they want to know if they won or buy more tickets for the next draw, with answers such as "Check if I won", "Buy more tickets" and "Something else".
There is always a way out for someone who wants something else: "Something else" takes them to the usual app, with nothing lost.
The prototype below shows the two moments. Use the buttons at the top to switch moments. The data is fictional.
Interactive prototype: use the buttons at the top to switch moments.
Why it works this way
In most apps, people have to look for what to do. Here, the app takes on that work, based on what they did last time. The expected result is less searching, fewer small decisions and the shortest path to finishing the task.
The bottom sheet asks without pulling the person off their screen and can be dismissed with one tap. I would limit the conversation to once per session, so it doesn't wear people out. The chat format is deliberate: it is a look everyone already knows, and a conversation invites people to try, while a row of buttons asks for a decision. That is my bet, and I would test it. The replies follow a simple rule: few (two to four), one highlighted as the main one, and always a way out.
It works in any app
Tele Sena is the example, not the limit. The mechanic has two parts: what the app knows about where the person is in the journey, and the most likely answer to the question "what now?". Just to illustrate, with made-up examples unrelated to any company:
- Delivery app: "Order the same as last Friday?", with "Yes, order" and "See other options".
- Bank: "Your bill is due tomorrow. Pay it now?", with "Pay" and "Remind me later".
- Streaming: "Pick up where you left off?", with "Continue" and "Choose something else".
Swap the app and the product, and the mechanic is the same.
The AI layer
The next step is letting the app learn from each person's behavior. The AI understands how each person uses the app and adjusts the suggestion so they get what they want in the most efficient way. A few examples of what I imagine:
- a push notification on the day the person usually buys;
- a different tone for someone who already knows the product well and for someone who is still unsure;
- the answer the person needs most on the very first screen, such as "you won" or "not this time".
The goal is the same as at the start: the person opens the app and it already answers, without being asked.
Safeguards
Suggesting a purchase calls for care, especially in a regulated capitalization product. If this idea ever left the page, I would follow these rules:
- Show the total before any tap and complete the purchase only on the next tap.
- No countdowns, no preselected option and no hidden way out.
- Never promise a win. That is why the button says "Buy more tickets", not something about increasing chances.
- Purchase suggestions only for people with an account confirmed as 18 or older.
- Let people switch the suggestions off and set a purchase limit.
- Check the wording and the offer with legal and with the sector's advertising rules before any test.
What I would like to measure
If this idea were tested, these are the numbers I would track and what each one shows:
- Tap rate on the answers: shows whether the suggestion is relevant.
- Purchase completion from the conversation: shows whether the question really leads to action.
- Time to finish the task: shows whether the path got shorter than in the usual app.
- Repurchase between draws: shows whether people come back through the shortcut.
- Conversation dismissals: if they go up, we are annoying people, and the fix is to ask less.
What I don't know
This is a hypothesis: there is no real data. I don't know, for example, how often the suggestion matches what the person wanted, or whether people who already know exactly what they want will find the question unnecessary. Those are the points I would test first.
What do you think?
Do you have an app that people keep coming back to for the same thing? Write to me, I would be glad to talk.