It started with a website. Then came the packages.

The brand gives gift packages to expectant and new parents. Parents sign up online, and the packages travel through maternity hospitals, gynaecologists, paediatricians and pickup points. Behind that sits a lot of data, a lot of consents and a lot of logistics.

The website

We started by moving the website from Kentico to WordPress, one installation for both Czechia and Slovakia. Then came a banner system for partner campaigns, with click tracking.

Customer data

Sign-ups arrive from many places: the website, Facebook and Google login, lead forms on social media, partners and even paper cards. So we built a customer data platform (CDP), one place where all of it meets. Parents, their children, addresses and, above all, consents.

Data only goes to partners when the parent has agreed to it and the offer is actually relevant to them. Today login runs through the CDP as well, and a parent has one account for everything. It makes sense: roughly half of all sign-up attempts come from parents who already have an account.

In a dashboard, management can see how many parents joined at each life stage, from pregnancy to children under ten, and how complete their details are.

More sources, more parents

Because everything meets in one place, new sources can be added without rebuilding the whole system. Step by step we connected lead forms and a partner source of sign-ups. Along the way we also rebuilt sign-up and login on the website.

AI where rules aren’t enough

In lead forms on social media, parents type their due date or their child’s birth date as free text. “Around 19.5”, “end of November”, “1.5 sa narodí a bude to dievčatko” (Slovak for “born on 1 May and it’ll be a little girl”), sometimes two children in one answer. A person reads the date at once. An ordinary program doesn’t.

Today plain rules read most of the answers. A language model gets only what the rules can’t handle, and we still check its answer. A year the model fills in has to make sense, or we throw it away. A date without a year is read as the nearest one in the future; otherwise a pregnant woman would turn into a mum in the system. And that’s no edge case: pregnant women make up more than half of new sign-ups.

And where AI doesn’t belong, it isn’t. Postcodes are read from the address by rules, not by the model. A guessed postcode is worse than none: it goes to partners and decides where a package is sent. You’ll find more about how we approach AI under AI implementation.

Packages to doctors

For doctors we built delivery planning. A week ahead, the surgery gets an email with a link where it can move the delivery or change the quantities. The warehouse gets weekly lists of what to pack for where.

Packages to parents

Parents book a pickup time at a pickup point and receive a verification code. Staff can see who is coming and what they’re due. The system works out whether a parent is entitled to a package from the birth date or due date in the CDP.

Sales with partners

The brand’s own sales team sells campaigns to partner brands. For them we’re building a CRM: companies, brands, contacts, meetings, tasks and campaign orders, including media plans and multiple currencies. A salesperson can send an order to a partner as a link, so the partner can look it over without an account.

Every salesperson sees their day on the home screen: overdue tasks, upcoming meetings and their own orders. The admin is in Czech and English.

All in one piece

The parts were built one at a time, over years. But they’ve talked to each other from day one, because one team built them, not several vendors who’d have to coordinate. We’ve worked together since 2018 and we’re still adding to the systems.