Turning traffic into booked demos

Aeva’s receptionists answer the phone, book appointments and write straight into practice software like Cliniko, so a physio with no front desk stops losing patients to voicemail. It worked. The website didn’t.

That is the founder below, reviewing the site I built him. It is the only thing on this page in someone else’s words, and it runs forty-seven seconds.

The founders were about to push hard on Facebook, with a real budget ramped over weeks. They wanted the site rethought before the spend went in, in months rather than quarters.

Nobody could tell me which part was broken. The copy is too technical. The pricing is missing. It looks dated. People aren’t ready for AI. Every explanation on offer was plausible and none of them was evidence, and I had three months to spend on whichever turned out to be true.

So I spent the first fortnight not designing. In a freelance engagement where the client is paying for screens that is an uncomfortable invoice to send. It was also the only way to avoid rebuilding a site around whichever theory was loudest in the room.

The homepage opened with what the technology is rather than what it does for you. A practice manager landing from an ad had to translate before they could evaluate. The illustrations were stock: on a page asking a healthcare business to trust an automated voice with its patients, generic reads worse than plain. And there was no obvious next step, so the interest the ad bought got spent deciding what to do.

The proof was the part that annoyed me. Aeva had video testimonials from practitioners, filmed on their own phones in their own clinics. It is the most persuasive asset the company owns. It sat eight screens down.

And it had not been built for a phone.

Four strands, run in parallel. A fortnight does not survive being run in sequence.

The churned prospects were the useful ones. Current users tell you why they stayed, which is a story about the product. People who left tell you what the site made them believe, which is a story about my work.

P5 is the one I kept coming back to. The answer was already on the page. It reached her through a phone call with a founder, which does not scale and does not run at two in the morning when someone clicks an ad.

Nearly three quarters of the audience was on a phone. Most of the budget went to Facebook, a phone-first channel, and it delivered people to a page laid out for a desktop. That is not a conversion problem you can write your way out of.

Coding the interviews and the recordings together gave three clusters. Two of them were the site’s to fix. The third was the product’s to keep earning.

That was the brief I actually designed against.

The old homepage was arranged around the product: what it is, what it has, who made it. The new one is arranged around five questions. Do you understand what this does for me. Can I hear it. Has someone like me done this. Is it a big job. Am I the kind of business this is for. That order is the design.

aevaai.com

The section I argued hardest for is the one that turns people away. “Want to see if Aeva’s a good fit for your clinic?” has two tabs, and the second is Not ideal for. In a market where every competitor claims everything, saying who a product is wrong for is the cheapest trust you can buy. It also does the sales team’s filtering: the demos that come through are from clinics that have already qualified themselves.

Aeva Outbound shipped after the site did. I designed the first version of that block; what is on the page now is that design after the business changed what it needed to say. It went in without a redesign. That is a softer test than a stranger dropping a section in — I drew the thing I am praising — but the part I would still point at is that the rhythm held while the content underneath it moved.

One content model, three layouts. Three quarters of the audience arrives on a phone, so the phone was the layout I designed first and the desktop is the one that got the extra column.

The trio above is there to show that one layout survives three widths. It also gives the laptop three and a half times the phone’s width, which is the wrong emphasis for a site where 73% of the audience arrives on a phone. So here is the one that actually mattered, at a size where you can read it.

Everything that matters is in the first screen. The headline, the proposition and a demo you can press, with no scroll required — on the layout 73% of the audience actually gets. The old site put its testimonials seven screens below this point and its setup time four. This is the only mobile capture I have that is worth showing: the other three in the repo are the same screenful as each other, and the full-page one is blank below the fold. Worth re-shooting.

You can call an AI receptionist straight from the page, no signup, and choose which of four voices answers. Thirty seconds, no commitment. The biggest source of friction in the decision stopped being copy I had to get right.

Where the old site described a capability, the new one stages it. “What Aeva does for your clinic” is a stepped section with a live call transcript running beside it, so the claim and its evidence are the same object.

And the proof moved. The video testimonials came up directly under the hero.

The barrier was never the product. It was the word AI. Clinic owners heard it and pictured months of setup and an IT department they do not have.

The obvious fix is to show the thing working. I was not allowed to. Explain how Aeva works without ever showing what it looks like. That was the brief.

I built micro-animations instead. One behaviour each, abstracted into a loop: a missed call being rescued, a slot filling, the queue updating live. Prototyped in Figma to get the timing right, then rebuilt as lazy-loaded React components so they cost nothing until they are on screen. Enough for a clinic owner to understand the product. Not enough for a competitor to copy the flow. And the effect I actually wanted: the whole thing looked like something you could set up in an afternoon.

The site went live alongside the ramped Facebook campaign, which are the worst possible conditions for reading a result. Anyone reading a case study should discount the headline number on sight. The one I lean on is underneath it: bookings from organic and direct traffic climbed too. That is a better signal, though not a clean one — a paid ramp lifts branded search and direct as a side effect, so some of that movement is the campaign arriving by another door. What it does rule out is the page having got worse, which is the thing a redesign is most likely to do.

The 78% is the one that tells me the research was right rather than lucky. Setup anxiety was the second largest hesitation in the interviews and the biggest single drop in the funnel. The redesign attacked it harder than anything else.

What I am most pleased with is duller than any of that. The site scales. A whole new product section went in months later without a redesign, and the pieces are components rather than a layout, so the next person to add a section is not negotiating with my decisions. I designed that section too, so it is not the independent test it would be if someone else had.

Good products lose to bad websites more often than people admit.

None of it was clever. The setup time was already on the site, the testimonials already existed, the demo already worked. All I did was work out what order they went in, and the only reason I knew was that I spent two weeks asking before I spent three months building. The other half of my work is a design system →