Tending.app: building digital sales for a service people didn’t know how to search for
Product design
UX research & checkout
1.
Trust starts before checkout
Redesign of the website homepage. At the start, it was just a landing page. I turned it into a full website that captured the first page of search results.
In this category, you cannot assume that a person will see an ad and calmly fill out a form. They may be grieving, feeling guilty, or simply thinking, “I need to do something, but I don’t know what.” And then an unfamiliar company offers to help with the care of a loved one’s grave.
The website had to answer not only “how much does it cost,” but also the questions people usually do not write in a brief:
is this a real and legitimate service?
who will come to the cemetery and how will they find the grave?
what will I receive after the work is done?
what happens if something is done poorly?
does this feel like a business built on someone else’s grief?
That is why trust work was not limited to one page. We strengthened search visibility, explanatory site content, proof points, reviews, and external mentions. Not for decoration. If a brand leaves no trace online, no set of icons in checkout can save it.
The inner pages answered all key user questions, and by spring 2026 Google results for grave-care-related searches were largely made up of pages from our site.
And yes, in normal ecommerce you can play with urgency and discounts. Here, that becomes ugly very quickly. We had to speak to a person like a person, not like a segment from an ad account.
2.
The need existed. Demand did not yet
About 47% of customers lived more than 500 miles from the cemetery. Some were older people who could no longer travel easily. Others were adult children aged 40–55 living in another state. The situation was real, and the pain was real.
But need is not the same as demand.
For demand to exist, a person first has to understand what their problem is called, that it can be solved, and that hiring help is normal. You cannot google a service you do not know exists.
Category language
There was no single ready-made term. Cemetery care, grave care, memorial care, cleaning, restoration, and maintenance were read differently across states and audience groups. In some contexts, restoration sounded almost like repair. Subscription sounded too commercial. Maintenance felt cold, like servicing an air conditioner.
Together with marketing and sales, we tested messages, collected reactions, and checked wording with native speakers. AI helped us quickly generate options, map them to pains, and test nuance. But it was a drafting tool, not the author of the solution: the final validation came from real people, the sales team, and common sense.
Gradually, we moved away from subscription and maintenance toward one-time care, ongoing care, and year-round care. Not because these were perfect words, but because they required less explanation and did not turn the service into a SaaS subscription for graves. Which, frankly, is a rather questionable genre.
What worked in communication
Emotional messages brought people in. Rational messages helped them buy.
For example, one of the strongest lines, You don’t need to do it yourself, achieved around 9% CTR compared with a typical 2–4%. It revealed the barrier well: the question was often not “can I afford this,” but “do I have the moral right to hand this over to someone else?”
After that, people needed proof: who performs the work, how the grave is found, what is included in the plan, what the photo report looks like, what happens if something goes wrong, and how much it costs. The communication became multi-step: gently name the pain first, then explain the solution, then provide proof.
Content and advertising production
The blog was not just an SEO layer. Articles for Mother’s Day, Memorial Day, and other remembrance dates showed the scenario: “I am far away, I cannot come, but I can care in another way.” We were not only capturing existing demand; we were helping people see that this form of care existed at all.
Real shoots were difficult. People rarely want to tell such stories publicly, and running around cemeteries with a film crew is a bad idea for several reasons at once. So we used AI production: based on real pains, we assembled scripts, hooks, illustrations, and videos for typical situations.
I assembled the first creative versions in Google Flow and then edited them in Apple Motion. Over time, the team built an AI pipeline with several tools, covering everything from scriptwriting to final editing in an almost automated mode.
AI significantly reduced the cost of a first version. But then the usual design work began: checking the rhythm of the video, the copy, the visual emphasis, and the respectfulness of the frame. The goal was not to create fake drama where there was already enough real drama.
3. The main checkout problem was not UI
The old flow asked a person to choose the type and size of a memorial, and only then showed a clear price. Inside the business, this looked logical: memorials are different, the work is different, so they need to be classified.
First version of the configurator. Пользователюприходилось решать какой у него тип памятника чтобы начать оформление заказа
At first, I tried to solve the problem as a designer: I worked with imagery, step structure, explanations, icons, helper text, and compared single-page and multi-step variants. These were reasonable hypotheses. They just did not solve the main problem.
For the user, it felt strange. They might remember the cemetery, the name, or have an old photo. But they should not have to understand how a plaque differs from an upright monument, what size it is, or which internal category it belongs to.
I took real completed orders and tried to classify the memorials myself. Many cases did not fit neatly into the categories. It turned out that managers who had worked in the industry for years were also confused. Even the founder struggled.
Manual memorial-type sorting caused problems even for the sales team that was supposed to sell it.
That was when it became clear: if people inside the business could not classify a memorial consistently, a user in an emotionally difficult situation certainly should not have to do it.
The problem was not UI. The pricing model required an expert decision from a non-expert.
Packages instead of classification
The founder decided to replace technical classification with three service packages. It is important to attribute this honestly: the package structure and pricing were the company’s decision. My work was around the new model — the flow, checkout logic, copy, package explanation, and moving unnecessary complexity after payment.
We changed the question from “what type of memorial do you have?” to “what level of care do you need?” This is closer to how a person thinks in this situation.
The updated configurator, which asked users to choose a level of care instead of a memorial type, reached the target conversion metrics.
Later, we saw a need for care for several memorials at once. We added a more convenient quantity selection and simplified name entry, but did not turn the first screen into a form for a cemetery dispatcher.
Only here did the UI work truly start working for the product: visual hierarchy, package icons, clear step names, hints, and human payment errors. Details matter, but they cannot mask broken logic.
Wizards, prototypes, and the mobile surface
In parallel, we designed a wizard / questionnaire that guided the person through several clear questions and helped with the choice. We tested where it should start: with the person’s situation or the service, with distance to the cemetery or the task, with price or with explanation. We tested different copy and argument sequences.
AI was especially useful for these early prototypes: quickly assembling several flows, a draft quiz, or a landing page, and checking the logic. It does not replace manual work, but it makes the first version cheap. That means the team can argue about the substance faster, not about who has time to draw the mockup.
Stepping back from the mobile app
The product had both a web and mobile app. Initially, the app was made to give the service more trust, but in practice users did not need it. There were few purchases in the app, and people mostly used it to understand the service or talk to support. But every checkout iteration had to be mirrored in the app, built in Flutter. That slowed down our experiments: two checkouts, two logins, two configurators handled by two teams. So we gradually stopped developing the app as an independent product center.
It is not the most spectacular part of the story, but it is useful. Sometimes a design leader is needed not to add another surface, but to stop multiplying surfaces in time.
4. Analytics became a way to argue with evidence, not taste
By spring, we had accumulated a substantial body of material: hundreds of recorded support and customer success calls, Stripe payments, GA4/Appsflyer, a buyer database enriched with demographics, surveys, and ad data.
AI helped with the first pass: classifying call topics, extracting objections, and gathering raw material for conclusions. But a summary by itself does not solve anything. We connected calls with Stripe orders and looked at who actually bought and how they moved toward payment.
One of reports made by Claude
This removed several convenient but false explanations:
the payment screen was not the main bottleneck: 46–50% of users who reached it paid; the main drop-off was at the beginning of the flow;
around 78% of customers completed checkout independently, although phone remained important for recurring and more complex cases;
purchases often matured 4–8 weeks after the first advertising touch.
The last finding changed the strategy significantly. Instead of trying to sell to a cold person from the first landing page, we built a longer communication path: articles, quizzes, advertorial pages, retargeting, and then rational proof closer to purchase.
It also confirmed an unpleasant but useful lesson about survivorship bias. Surveys, interviews, and calls made the 55+ audience very visible. But purchases revealed a large quiet cohort aged 40–55. These people did not write emotional comments and did not always agree to interviews. They simply bought.
Result and what I took away
The old flow with memorial selection topped out at around 10 orders per month. After the move to packages, a lighter checkout, and broader work on language, trust, and the funnel, online sales reached 100–150 orders per month.
This is not a story where “I redesigned checkout and caused all the growth.” During this period, traffic, advertising, and the overall work of the team improved as well. But the new flow removed a fundamental barrier: people stopped being given a task that even an experienced company employee could not solve.
My mistake at the beginning was treating the pricing model as a business constant. I tried to make it more understandable, when I should have questioned the model itself earlier. The founder proposed the decision, and I do not want to retroactively claim it as mine. But after this story, the rule became simple for me: if the data points to a problem in the business logic, discussing it is part of the design job.
Another takeaway: need and demand are different things. People wanted to care for loved ones, but they did not immediately understand that it could be done this way. So the solution was not to pour traffic into checkout, but to guide the person through clear language, content, trust, and several touchpoints: emotional hook first, rational arguments later.
And finally, AI. It sharply accelerated scripts, prototypes, landing pages, material analysis, and ad production. But as of summer 2026, quality still depends on experience, taste, and manual refinement. AI is good at helping produce. It is worse at understanding why people do not buy.
The company is now restructuring toward AI-driven development, so the role of traditional design and development is shrinking. I understand the logic and use these tools a lot myself. But Tending taught me one very grounded thing: you can quickly build ten more quizzes. If you do not understand that the barrier is guilt, language, or a broken choice model, they will simply become ten fast ways to repeat the same mistake.