Case study

Oskoole: building an AI helper for the people tech usually forgets

How we designed, built and shipped an accessible AI assistant for the 50+ community, and what it taught us about designing AI for social impact.

Sector: Aged care, retirement living, digital inclusion
What we built: A multilingual AI assistant, white-label ready
Languages: 76, text and voice, generated on demand
Deployment: Seven days to a branded, working platform

The problem

Digital exclusion is a design problem, not a skills problem

66.5% of people over 65 experience digital exclusion, and being digitally excluded can cost people up to 25% more for essential goods and services. So this isn't about convenience. It's about money, independence and safety.

The standard response is training. Run a class, hand out a booklet, put on a workshop at the library. We think that misreads the problem. Older adults aren't failing to learn. They're handed tools designed for someone else, then blamed when those tools don't fit.

We didn't find this in a market report. It started with one of our founders being the family IT department, and the Loom videos and YouTube channel that came out of it. That story is here.

Two things stood out once we started listening properly.

Nobody wants to ask twice.The barrier isn't understanding. It's the fear of looking foolish and the sense that you're taking up someone's time. People stop asking long before they stop wondering.
Help disappears.Someone shows you how to do it, you nod, you get home, and it's gone. There's no replay button on a grandchild.

The brief wasn't teach older people technology. It was: let someone ask a question they've been putting off, get an answer they can follow, and come back to it as often as they like without anyone knowing.

Our approach

Start at the moment of need, not the curriculum

A question, typed in someone's own words, in the middle of the afternoon. Is this text message a scam? How do I video call my granddaughter? From there we worked backwards to the smallest thing that could answer it well.

Start from real questions

We already had years of them, from family and then from YouTube comments. Every early feature answered something a real person had actually asked.

Ship narrow, then widen

Text tutorials first. Then voice, images, and 76 languages. Each layer only got added once the one underneath worked for people.

Test with the hardest users

If it works for someone who has never installed an app, it works for everyone. The reverse is never true.

Fit the programs already running

It adds another layer to the digital literacy programs an aged care residence or retirement village already runs, rather than asking them to start something new.

The role of AI

One job: turn a messy question into clear, ordered steps

That sounds small. It's the difference between a search result that assumes you already know the vocabulary, and an answer built for the exact thing you asked.

Tutorials generated on demandFrom a question written in plain, messy, human language. No prompt-crafting, no special phrasing, no having to ask it the right way.
Voice narration in 76 languagesGenerated automatically, so a tutorial arrives in someone's first language rather than their second. For culturally and linguistically diverse communities this is often the whole thing.
One step at a timeThe interface never drops a wall of instructions on someone.
Every answer savedAI makes the answer, but the value compounds because the answer stays. Come back at 11pm three weeks later and it's still there.

Oskoole isn't a chatbot you have a relationship with. It's a helper that answers and gets out of the way. People asked us for help, not a friend.

Brain Skoole

A product that only answers questions has a problem. You open it when something's wrong, then don't think about it again for three weeks. That's a tool, not an assistant.

So every subscription includes Brain Skoole, a cognitive wellness suite of daily exercises for memory, speed and attention, with streaks and session tracking so people can see themselves improving. It does three jobs.

It gives people a reason to come backIf the app is already familiar, asking it something feels like nothing. If you haven't opened it since March, asking feels like an event.
It changes what the product isNot a help desk you visit in a crisis, but something you spend time with.
It serves a need the sector already hasCognitive engagement is core business for aged care and retirement living. Lifestyle teams are already running puzzle mornings and trivia.

Accessibility and human-centred design

Accessibility set the constraints, it wasn't a pass at the end

One step per screenNo scrolling to find where you were.
Voice on everythingFor low vision, for low literacy, and for anyone who just prefers to be told.
Steps you can stop and startDownload them for print, or share them with someone else.
High contrast, large targets, generous spacingBuilt for an unsteady hand on a tablet, not a mouse on a desktop.
Three free questions, no account, no credit cardSign-up walls are where digitally excluded people fall off, so we moved the wall.

Oskoole exists so nobody has to ask a busy relative for the fourth time. Preserving dignity was a functional requirement.

The white-label approach

Every residence gets its own assistant, under its own name

We were never going to reach older Australians one at a time. Aged care residences and retirement villages already have them, already have the relationship, and already have a brand residents trust.

Look at any residence's website and you'll see the same things: the rooms, the grounds, the community, and a lifestyle program of exercise and creative classes. That program is what operators compete on, and it's where an AI assistant belongs.

Org-level branding at the authentication layerA resident signs in to their village's product, not ours.
Tenant isolation availableFor operators who need their user pool fully separated.
Seven-day deploymentAn operator can have a branded, working platform in a week.
Two offerings under one brandTech support and a cognitive engagement program in the same app. That's far easier for a lifestyle manager to justify than a help desk alone.

Key decisions

What we chose, and why

Step-by-step tutorials instead of a chat windowA chat interface asks the user to be good at asking. Steps ask nothing.

Generate on demand instead of a fixed course libraryPeople's questions are specific and endless. A library is always missing the one they have.

Save every answerRepetition without embarrassment is the feature.

76 languages from day oneCulturally and linguistically diverse older Australians are the most excluded group and the least served.

Voice by default, not as an accessibility toggleOpt-in accessibility features mostly don't get opted into.

Free before signupSign-up walls are where the least confident users quit.

Bundle Brain Skoole rather than sell it separatelyA question-only product gets opened rarely. Games build the habit that makes the help useful, and give partners a cognitive engagement program in the same licence.

White-label rather than direct-to-facility brandingOperators want a resident experience that's theirs. Fighting that would have cost us the channel.

What we learned

Five things we'd tell anyone building this

Reading level does more work than feature count.

The single biggest lift in comprehension came from rewriting copy, not from adding capability.

The person paying often isn't the person using.

For consumers the buyer is frequently an adult child buying for a parent. For facilities it's a lifestyle or operations manager. The product has to make sense to both, and they want completely different things from the first screen.

Simple is expensive.

Stripping an interface back to one action per screen takes far longer than building a dense one. Most of the build time went into removal.

An assistant needs a reason to exist on the quiet days.

Support-only products get opened in emergencies, which is exactly when a nervous user is least willing to learn something new. Adding Brain Skoole was as much an accessibility decision as an engagement one.

Don't let AI be the story.

Nobody using Oskoole cares that it's AI-powered. They care that the answer arrives and makes sense. We pulled AI language out of the user-facing copy almost entirely, and engagement improved.

White Label

These screens are from My Care Residence, a demo operator we built end to end so partners can see exactly what their own version looks like before they commit.

Ask a question
The entry screen under a partner's brand. One field, plain language, no account required.

A tutorial, step one
One instruction per screen, large type, voice playing.

The same tutorial in Spanish
Any question, answered in any of 76 languages.

Saved tutorials
Every answer stays available, without asking again.

Brain Skoole
Memory, speed and attention exercises, included with every subscription.

Could we build something like this for your community?

Oskoole is one project. The approach behind it travels. We've applied the same method to Ngarrgu, an AI assistant built with and for First Nations communities, and to the Dissemination Digest for researchers and students.

We're looking for pilot partners in aged care and retirement living for Oskoole, and First Nations organisations for Ngarrgu.

Start a pilot conversation See Oskoole in action