This week's highlights: an owner upgrade programme for a premium LED skincare device brand

TLDR: We built an owner upgrade programme for a premium LED skincare device brand, an early warning system that reads support conversations for a premium fashion basics brand, and a family of paid search landing pages for a family farm attraction.

The thread running through this fortnight was knowing exactly who you are talking to. A returning device owner should not be treated like a stranger. A shopper who searched for one specific thing should not land on a generic page. A customer who has emailed twice about the same problem should show up somewhere before the sales chart does.

So most of the work was about recognition: recognising loyal customers, recognising early signals, and recognising intent. There is a rules engine for checkout in here too, along with a new language across a storefront. Here is the rundown.

This week's highlights: an owner upgrade programme for a premium LED skincare device brand

The standout build is for a premium LED skincare device brand with a happy problem. Thousands of people already own one of their masks, and a newer professional model has since arrived. Rewarding those existing owners with a credit towards the upgrade is an obviously good idea. Doing it by hand, across a spreadsheet of orders and a pile of one-off discount codes, is obviously not.

So we built the whole programme properly. Owners land on a dedicated page and sign in with the account they bought through. The page checks their purchase history, greets them by their own first name, and tells them within a second or two whether they qualify. There are no codes to hunt for and no forms to fill in. If they are eligible, one button claims the credit and holds it for them for a set window, so nobody feels rushed into a decision they have not made yet.

From there the credit follows them around the site. The professional product page shows the reduced price rather than the standard one, so the offer stays visible while they read the detail and compare models. A notice sits at the top of every page while the credit is live, and the floating navigation politely steps aside so the two never collide. At checkout, the credit applies itself. This is the same thinking behind the Shopify loyalty program work we wrote about earlier this month: a reward is only a reward if claiming it is effortless.

The brand's team drives all of it from a single admin screen. They can see who is eligible, invite the list by email, and send a test version to several colleagues before anything goes out more widely. The invitation email carries their real branding rather than a plain-text apology for itself. Small thing, big difference to how an upgrade offer reads.

The nicest detail sits underneath. Every possible outcome raises an event into their email marketing platform: claimed, not eligible, already claimed, credit expired. Each of those groups can now be followed up differently, because "you claimed but have not ordered" and "you are not eligible yet" want very different messages. The programme therefore keeps working long after the first invitation lands.

Under the hood: an early warning system that reads the inbox

The engineering story of the fortnight comes from a premium fashion basics brand and their reporting platform. They now have an early warning dashboard, plus a digest that arrives once a week. Both answer one question: what are customers unhappy about right now, before it shows up in the numbers?

Every night the platform sweeps the brand's customer service conversations. A language model reads each one and sorts it by theme and by tone, turning thousands of individual messages into a handful of countable categories. Those counts then sit right next to orders, returns and delivery data in the same reporting database. A rise in delivery complaints can be read against actual delivery times, in one view, on the same day it starts.

Two more views joined the platform alongside it. A global opportunity map reads across all time rather than the selected dates, showing where in the world demand exists that nobody is actively serving. And their direct mail reporting gained postal-area coverage, mapping which neighbourhoods received a mailing against where the resulting orders actually came from. The earlier ecommerce dashboard work gave us the foundation to add these quickly.

Worth noting for anyone sitting on a support inbox: your earliest signal is almost always unstructured. Sales charts confirm, two weeks late, what customers were already telling your team in plain English. Classify those conversations into a small set of themes, count them daily, and put the counts beside your trading numbers. It is the cheapest trend detection you will ever build.

Other work this fortnight

A family farm attraction had a big fortnight on the paid search side. Their landing pages used to be near-identical, which meant a family searching for a rainy day activity in one town saw much the same page as somebody comparing days out in another. We rebuilt twelve of those templates so each one speaks to its own intent and its own town, added a page aimed squarely at people looking for an alternative to the zoo, and built a browsable grid of things to do that any of those pages can drop in. A long-form editorial page also went live, built around forty-seven questions about what a proper childhood should include, which is a lovely piece of writing to have a page worthy of.

Their booking operation moved forward too. When a family amends a booking, staff can now see the price difference clearly and raise an invoice for it without leaving the order screen, so date changes and extra tickets stop being a phone call. Their summer pass has a proper product page with a configurable layout and a sticky booking button that follows you down the page on a phone. And a weekly performance report now writes itself into a shared spreadsheet at one o'clock every Monday morning, with a log tab recording every run, so anyone can see at a glance that the numbers are current.

A hair wellness brand gained a rules engine at checkout. Delivery options and payment methods can now be hidden based on conditions the team sets, so a free international delivery option that was never meant for Canadian orders simply does not appear there. Crucially, those rules are managed from their own admin screens, which means a policy change is now a two-minute edit rather than a development ticket. Their consultation quiz also opened with fresh copy, and their reporting now splits new customers from returning ones.

A luxury fragrance house went live with Spanish across their storefront, and we filled the remaining gaps in their German. Every button, label, message and error state has to exist in each language, or shoppers fall through into English at exactly the wrong moment. It builds on the Shopify multi language work we did for the same market earlier in the summer.

Our subscription tooling learned to look after itself. Subscribers can now manage everything from inside their account area on the store: change what is in an order, update payment and delivery details, pause, skip or cancel, all without emailing anyone. The admin side was rebuilt to match, giving full control over any contract in a few clicks. A proper customer self service portal is worth building because it removes the two things subscribers dislike most: waiting, and having to ask.

The quick wins

A few smaller pieces made it out of the door as well. Our review platform now publishes product reviews to Google Shopping with the right product identifiers attached, so star ratings appear against the correct item rather than quietly failing validation. A backfill tool went with it, pushing years of existing reviews into that feed in one run rather than waiting for new ones to trickle in. Feeds also now identify each brand by its own name, which matters when one platform serves several businesses at once.

That is the fortnight. The upgrade programme opens to its full list shortly, and the early warning dashboard gets its first proper test the moment something goes wrong. Thanks for reading. More next Tuesday.

Photo by Ashley Piszek on Unsplash