Least-Intrusive Prompting, and Designing Technology That Waits
Every support plan contains a hierarchy nobody wrote down. It runs from the natural cue at one end to hand-over-hand assistance at the other, and where you enter it decides whether a person keeps a skill or loses it. Most technology enters at completely the wrong level — and it does so because the wrong level is easier to build.
Prompting works from least intrusive to most, and you only move up when the previous level fails. Most reminder technology skips straight to a direct verbal instruction, every time, regardless of whether the person needed it. That is not support. That is interruption on a schedule.
The hierarchy, and why the order matters
Direct support professionals are usually trained in some version of a prompting hierarchy. The specific wording varies between agencies, but the structure is consistent, running from the least intrusive intervention to the most:
- Natural cue — the environment itself signals what happens next. The kettle is on the counter. The coat is by the door.
- Indirect verbal — a question rather than an instruction. “What comes next?”
- Direct verbal — naming the step. “Time for your jacket.”
- Gestural — pointing, or looking towards the object.
- Model — demonstrating the step so it can be copied.
- Partial physical — a light touch to initiate movement.
- Full physical — hand-over-hand.
The principle is simple to state and difficult to hold: start at the least intrusive level that could plausibly work, and only escalate when it does not. Every level you skip is a level of independence you have quietly taken over.
Where technology usually enters — and why
Almost every reminder system on the market operates at level three. A notification fires at a set time and states the task. It does this whether or not the person was already reaching for their jacket, whether or not the natural cue was sufficient, and whether or not they have done this unprompted for six months.
This happens for an unglamorous reason: level three is the easiest level to guarantee. A system that reliably says the thing at the right time is straightforward to build and simple to demonstrate. A system that waits, observes whether help is actually needed, and only then offers the lightest possible nudge is harder to build and much harder to show off in a demonstration.
So the industry optimises for reliability, and reliability quietly means intrusion.
Prompt dependency: the cost nobody records
When a prompt consistently arrives before someone has the chance to initiate a step themselves, a predictable thing happens. The prompt becomes part of the task. The person is no longer making coffee; they are waiting to be told to make coffee, and then making it.
This is usually described as prompt dependency, and its most difficult feature is that it looks like success in the record. Task completion goes up. Incidents go down. Everyone involved is doing their job well by every measure being collected. Meanwhile, the amount of the day the person originates themselves is shrinking, and nothing in the documentation captures it.
Support systems almost universally measure whether the task was completed. Very few measure who initiated it. Until initiation is recorded, prompt dependency is invisible — and what is invisible does not get corrected.
Designing technology that waits
The design instinct that serves independence runs opposite to the design instinct that serves reliability. In practice it comes down to four decisions:
Delay before prompting
Build a deliberate gap between the scheduled time and the prompt. If the person starts the task in that window, the prompt never fires. The pause is not an inefficiency; it is the whole point.
Enter low, escalate slowly
Open with the least intrusive form that could work — an ambient cue, an indirect question — and only escalate to a direct instruction if nothing follows.
Fade on purpose
A prompt schedule that never changes is a plan with no ambition. Review it on a set cadence and reduce it deliberately, the same way a staffing plan would be reviewed.
Make it feel like a person, not an alarm
A familiar voice offering a question lands differently from a device announcing a task. Tone changes whether a prompt is experienced as support or as being managed.
What this looks like in a real morning
Consider a morning routine with medication at eight, breakfast, and a bus at nine.
The level-three version: at 8:00 a device announces “time to take your medication.” At 8:20 it announces breakfast. At 8:45 it announces the bus. It does this identically on day one and on day four hundred. The routine is completed. The person has been narrated through their own morning.
The waiting version: at 8:00 nothing happens. The pill organiser is where it always is — the natural cue is doing its work. At 8:06, if nothing has moved, a familiar voice asks a question rather than giving an instruction: “anything before breakfast?” If the answer is action, it stops there. If not, at 8:12 the prompt becomes specific. The escalation is real but it is earned, and most days it never gets past the question.
Over months, the second version generates something the first cannot: evidence about which parts of the routine the person actually initiates. That is the data a support plan should be built from, and almost nobody collects it.
How this shapes what we build
Luka is designed around the assumption that a prompt is a cost as well as a help. Reminders and routines can be scheduled, but the goal of a good configuration is for many of them never to fire. Because prompts are delivered as speech in a familiar voice — including, with consent, a cloned family voice — an indirect question is available as a genuine option rather than a novelty, which makes entering the hierarchy low actually practical.
The caregiver app lets an approved person adjust that configuration remotely as someone becomes more independent, which matters because fading only happens if changing the plan is easy. A fading schedule that requires a site visit to adjust will not be adjusted.
“Which prompts have we removed this quarter?” If the answer is none, the plan is not supporting independence. It is maintaining a level of assistance that was set once and never revisited.
The uncomfortable conclusion
A support system doing its job well should be doing less over time. That is difficult in practice, because reducing support looks like reducing service, and every incentive in the field — contractual, regulatory and emotional — pushes towards more.
The organisations that manage it tend to share one habit: they treat the fading schedule as a deliverable with a date on it, not as an aspiration in a document. The pause before the prompt is where independence lives, and it only exists if somebody deliberately designed it in.
Luka offers routine, reminders, and a familiar presence in the room, and lets an approved person check in and be reached remotely — supporting independent living as assistive / enabling technology.
Luka Bear is an assistive, non-medical companion device. It is not a medical device and is not an emergency response system. In an emergency, always call 911.