A paper form on a phone is still a paper form

The US home care industry is worth over $100 billion, and a surprising amount of it still runs on spreadsheets, phone calls and paper. Somewhere right now a coordinator is matching a caregiver to a client by working down a list of who is free, because "free" is the only criterion there is ever time to check.

Last year I designed the platform meant to replace that: an AI-powered home care platform for US agencies that coordinate caregivers for elderly, disabled and recovering clients. A desktop hub for the agency, and mobile apps for the people requesting care and the people delivering it. Three surfaces, one design system built from scratch, and one decision I would defend harder than any of the others.

Four seats, one failure

I did not start from a feature list. I mapped the chain by role, because an agency coordinator, a client, a family member and a caregiver each experience the same broken process very differently.

Agencies match by hand, so availability becomes the only criterion that reliably gets used, and specialisation, language, personality and travel distance get considered when there is time. There usually is not. Billing crosses private pay, insurance and Medicaid or Medicare, each with its own rates and paperwork, and the industry loses an estimated 5 to 10 percent of revenue to manual invoicing. Clients cannot request a visit without a phone call, and once care starts, pricing is opaque and nobody can see who is coming or when. Families are left outside entirely, which produces a steady stream of anxious check-in calls back to the agency.

The reframe that shaped everything: these are one failure seen from four seats. The coordinator matching on availability, the family calling to ask who is coming tomorrow, and the caregiver filling in a paper form on a doorstep are all symptoms of the visit having no shared record.

So the visit became the spine of the product, from request through AI matching, scheduling, delivery, verification and payment, and each of the three interfaces became a view onto that one lifecycle.

The decision I would defend hardest

Electronic Visit Verification is required by many states for Medicaid-funded care. Every visit needs the service type, date, time, location, caregiver identity and the recipient's confirmation. Get it wrong and the consequences are audit failures, denied claims and lost funding. The people carrying that weight are caregivers, who would rather be delivering care.

The obvious move was to reproduce the existing paper EVV form as a digital form. It maps directly onto what agencies already file and what auditors already expect. Nobody would have argued with it.

I rejected it. A paper form on a phone is still a paper form. It inherits exactly the problem the paper had, which is that it gets completed at the end of a shift by someone tired, standing in someone else's living room. Compliance data is only as good as the conditions it is captured in.

So it became one continuous guided flow: service details, recipient confirmation, tasks performed, notes, and two electronic signatures, one from the caregiver and one from the recipient. Each step asks for one thing and validates as it goes. The recipient's signature sits at the natural moment the visit ends. That is what makes the record accurate rather than merely digital.

The expensive decision

Most two-sided products ship one mobile app with a role switch after login. It is cheaper to build, cheaper to maintain, and the standard choice.

I designed two. The roles share almost nothing except the visit itself. A client's job is to describe a need, choose a time and pay. A caregiver's job is to manage availability, accept work, complete regulated documentation and track earnings. One shared shell would have meant a navigation that compromised both, and every screen carrying conditional logic for an audience that would never see it. Two apps let the caregiver app carry the full weight of compliance without any of it reaching the client.

It cost more. It was worth it, and I would make the same call again.

Putting the price where the anxiety is

The industry standard is to arrange care first and bill afterwards, which avoids putting a number in front of someone at a vulnerable moment. It sounds kind. It is not.

People arranging home care are usually doing it under stress, often on behalf of a parent, and cost uncertainty is one of the reasons the decision gets delayed. Withholding the number does not remove the worry, it just moves it to an invoice that arrives later with no context. So transparent hourly pricing and a live cost estimate sit inside the booking flow, while duration and frequency can still be changed.

Families are not a distribution list

The cheap version of family involvement is a notification setting: forward visit updates to a few more email addresses. I designed it instead as Family Access, an invited, permissioned view of schedules, caregiver notes and health updates.

The anxiety behind repeated check-in calls is not fixed by pushing more messages outward. It is fixed by giving someone a place to look. And an invited view keeps the client in control of who sees their health information, which a forwarding rule does not.

What agencies reported

Agencies on the platform reported operational efficiency up 94 percent, administrative workload down 74 percent, and caregiver and client match satisfaction at 9.8 out of 10.

The honest part

Validation on this project was stakeholder review against interactive prototypes, with the product managers, engineers and AI specialists who were building it. It was not testing with caregivers or clients. The four role profiles came from client knowledge, competitive analysis and the published regulation, not from watching anyone work.

Which means the central argument of this whole piece, that compliance data depends on the conditions of capture, is one I designed for and never observed. I still believe it. It is still the first gap I would close.

What I carry into every regulated workflow now

I approached EVV at first as a question of capturing the right fields, because the rules are explicit about what has to be recorded. The fields turned out to be the easy part. Getting them right is necessary and does nothing on its own.

The other lesson is broader. Every transparency decision on this product, price during booking, caregiver ETA, payout dates, the care record opened to family, removed an ambiguity the old process relied on to avoid difficult conversations. Opacity in a legacy process is rarely protecting the user. It is usually protecting the operator from having to answer.

Design for the moment of capture. Then design for the question nobody wanted to answer.