The pioneer of a brand-new niche.
Our client, Duty Refunds, is the pioneer of a new industry in the UK. They started their business after Brexit, when people could get a refund of customs charges for goods they returned to the EU. Duty Refunds has one prime goal: to make the tax-refund process easy, online and secure, so people do not have to worry about a tiresome, complex process when they want to reclaim import charges.
But not enough people knew about this business or its niche, which made generating leads a slow process. Daniil approached us at the end of 2021, seeing potential for growth with paid advertising on Google and Facebook.
Even though it was new, Duty Refunds had already worked with some more prominent brands and carriers. The work began with a landing page that served as the only URL for all the campaigns we launched. But that was not enough for our client to reach the goal it strived for.
One landing page, a brand-new idea to explain.
As we mentioned, only one landing page was available as the final URL while the website was still in progress. This made the situation a tad more challenging for advertising, which is why this page served as a short-term solution.
After some brainstorming, we started working on a new landing page to increase the conversion rate and refine our first steps in building it. And there we stumbled upon a question: what can we learn from the current landing page to find solutions for the future?
Since this was a new way to reclaim money lost on import tax and VAT charges, the first challenge came down to helping people understand what they could return to the EU and reclaim, and to increasing trustworthiness.
Build small, optimize daily, then scale.
a Single Keyword Ad Group
As this was a new account, in a brand-new niche, with the goal of validating an MVP, a limited ad budget and one landing page, we knew we only had room in the budget for Exact match keywords.
But it was not only ever about the budget. We did not know which keywords would work, we had no negatives, and we had no data yet to start with a Target CPA smart bidding strategy.
The service we provide is relevant only to one type of tax. We had to pay extra attention not to go too broad with our keywords, or to show our ads for any other type of tax refund.
In our research, we focused mostly on bottom-of-the-funnel, long-tail keywords, but at the same time we were building out a list of short-tail relevant keywords.
Since we were going to start with manual bidding, we decided to group our initial keywords in Single Keyword Ad Groups (SKAGs). Search volume per ad group was not the factor at that moment, since we were not running smart bidding strategies.
With SKAGs, we had a high level of control over the search terms triggering our ads, and full alignment between the search query and the ad that shows. It is a great way to structure an account in the starting phase, while you sort out negatives, learn bids, and collect enough data to switch to tCPA bidding.
As our budget was limited at the start, it was crucial to reach our desired CPL as quickly as possible, since that was the first milestone towards scaling. We ended month one with a 13% lower CPL than expected.
Everyone was happy with what we accomplished, but we knew it was just the start, and that we needed to do much more to make this account grow consistently.
Month one closed 13% under target CPL.
b Tracking and analytics, from scratch
Before we trusted a single number, we built the measurement. Through Google Tag Manager we set up GA4 and the full conversion funnel, and as the website came together we tracked every step: Check Eligibility, Initiate Refund, Confirm Email, all the way to an HMRC claim submission and approval, the closest signal to real revenue this business produces.
We did not stop at Google Analytics. We piped the events into BigQuery and built Data Studio dashboards the client could actually read, so Daniil and his team could see the whole funnel in one place. For the qualitative side we added Hotjar and LogRocket: heatmaps, scroll maps and session recordings that showed us how real people used the page.
That measurement layer is what made everything after it work. Once the funnel was honest, every bid, keyword and landing-page decision ran on real signal instead of a guess.
The whole funnel measured, from the first click to an HMRC claim, on GA4, BigQuery and Data Studio.
c Daily, Bi-Weekly & Weekly Account Optimization
When you start a new search campaign, the best thing you can do is focus on search terms and eliminate irrelevant search queries. It becomes even more important on a new account with newly set up conversion tracking and smart bidding strategies that are not helping you yet.
As we started with manual, data-light bidding, we had to make sure our bids were neither too high nor too low. The goal was to gradually increase the aggressiveness of our bidding as we strengthened our negative keywords.
We analysed search terms daily, looking for relevant long-tail keywords and adding them to our exact-match campaigns. At the same time, we were always on the lookout for anything irrelevant or too broad for us at this stage. This kind of daily rigor is what disciplined Google Ads management looks like.
We had no funnel to work with yet, nor the website and social media to build one on. We had to be very careful to target only bottom-of-the-funnel keywords and maximize the number of leads from the given ad spend, since the business was just starting out.
d Landing Page Optimization
Quickly after the first campaign started running, we implemented Hotjar, a user-behavior analysis tool. Watching closely, we analysed heat and scroll maps and screen recordings. The goal was to understand how good a job our landing page actually did of explaining what we do.
We spotted two main problems: a lack of social proof and a lack of informative content on the page. The landing page was fairly short, and since the niche was new, we suspected this was the issue.
The recordings showed users browsing back and forth across the page, reading things over and over, trying to understand what we are all about.
We asked our client to reach out to a few people who had used the service and ask for testimonials. We wanted those on the landing page, and they proved to be the crucial element that significantly increased the trustworthiness of our service.
We also wrote additional sections that explained how the process works, its benefits, and which items people can reclaim import and VAT charges for.
As soon as the landing page was deployed and conversion tracking was done, we created an experiment in Google Ads, splitting our search traffic 50-50 and testing the original landing page against the one with additional content. The first iteration of the original landing page was live two weeks after the account started.
Over the next four months, we had three more iterations of the same page, improving it based on findings from Google Analytics and Hotjar. The most significant improvement between two iterations was when we tested v3 against v4.
We made more significant element changes (layout, copy and step changes), which resulted in 25% higher conversion rate and 27% lower CPL.
The v3 vs v4 test gave +25% conversion rate at -27% CPL.
e From SKAGs to STAGs
The first big structural change came at the beginning of our third month in the account, when we switched from Exact match keywords to Phrase match keywords. Our exact-match keywords had surpassed 90% Search Impression Share, and we could not grow much more relying only on what was live in the account.
Our SKAG structure ran on maximize clicks, and the goal with it was to reach the point where we had built up enough conversions over the trailing two weeks, so we could start transitioning to smart bidding strategies. But SKAGs, and the way the account was structured, were not optimal for machine learning.
We had to change our approach if we wanted the account to grow. We grouped some of our top-performing exact-match keywords into themed groups, so phrase-match keywords could trigger them.
Unlike SKAGs, STAGs (Single Themed Ad Groups) support automated bidding strategies much better, because the search volume is higher at the ad-group level and the machine has more room to learn.
The STAG structure let us optimize the budget a bit more easily, and a higher number of impressions per ad group created the perfect conditions for machine learning. By then we had a strong list of negative keywords and a nice number of conversions recorded in the account.
The switch was not instant magic. The first two weeks on phrase and Target CPA actually ran below our two-month manual-bidding average while the machine learned. We kept our heads down, fed it negatives, and it recovered in the second half of that month back to our previous performance, then pushed past it: a phrase campaign with 42% lower CPL by the end of month two.
f Broad Match Keywords Transition
The second big structural change happened at the beginning of our sixth month in the account, three months after we switched from Exact to Phrase match keywords. By then we were spending three times our starting budget, and we quickly got to the point where we could not grow much more relying only on what we already had.
We already had every relevant short and long-tail keyword in our account. Since we needed to scale more, conversion volume became the priority over CPL, and the next step for us was broad keywords.
As with everything, we always start small, then scale. We picked a few best-performing long-tail keywords and created a broad-match campaign. We knew that, with our conversion tracking and the number of leads we were getting monthly, tCPA would push this much further, the question was just how long it would take.
Broad keywords can be dangerous, and we had a few occasions where it did not go as planned. But we knew we had super-strong negative keyword lists, and our pixel had been fed with the correct data for months, so we were sure tCPA would bring quality leads.
We started seeing incredible performance even before the 7-day learning phase was completed. June vs May 2022: we had 137% more leads with 19% lower CPL at the account level, for 91% more spend. And it only got better in July.
July vs June 2022: we had 63% more leads with 12% lower CPL, for 43% more spend. The key to scaling in June and July while maintaining and improving profitability was avoiding resetting the learning phase as we optimized. We changed tCPA and budget once a week, with changes no higher than 10-15%.
June +137% leads at -19% CPL (+91% spend); July +63% leads at -12% CPL (+43% spend).
g Migrating the conversion, and widening the channels
We started with one Unbounce landing page that only collected leads. Once the website was ready, it carried a multi-step form that estimated a customer's refund, and our landing-page form began handing users straight into that website funnel.
We set up tracking for the entire website, every step of the new form, but our campaigns had been optimized toward the original lead for months. We could not just switch it out, or we would have thrown away all of that learning. So we created the new conversion alongside the old, let it gather a healthy volume of data, then transitioned the bidding across through experiments rather than overnight.
With the funnel deeper and the tracking solid, we widened the account: a Discovery campaign, Display, audience layers, and a Facebook campaign going into the Q4 season, alongside the brand and tone-of-voice work the client was building out. Pushing into colder traffic made volume choppier week to week, but the efficiency held, the cost of an initiated claim falling about 25% month on month through the early autumn.
The conversion migrated cleanly to the website funnel, and the cost per initiated claim fell about 25% as we expanded into Display and Facebook.
Three records, three structural moves, on numbers we trusted.
March vs February 2022: the first structural change, Exact to Phrase. The new phrase campaign was even slightly more expensive in its first weeks, because the smart bidding needed time to learn and we needed time to negative out new, irrelevant search terms. By the end of its second month it had turned: a 42% lower CPL.
June vs May 2022: the second structural change, the move to Broad match on a matured tCPA, with three full months of clean conversion data behind it. It broke out: 137% more leads at a 19% lower Cost per Lead, for 91% more spend.
July vs June 2022: a month for the machine to settle, and with us assisting on negatives, another record. We got 63% more leads at a 12% lower Cost per Lead, for 43% more spend.
The growth did not cost us quality, because we could see the whole funnel. We tracked all the way to an HMRC claim, so we optimized toward real claims rather than form-fills, and into the autumn the cost of an initiated claim kept falling, about 25% month on month, even as we pushed into colder Display and Facebook traffic.
We ran the account through 2022 and kept it efficient into early 2023, when the client moved into a cost-cutting phase ahead of a fundraise. We left it at its most measured and scalable: a textbook build, handed over in good health.
Proactive beats rushed, every time.
Paid advertising is the way to go for generating leads, but scaling fast for quick results, just to save time, was not the answer.
The most crucial thing in the process is gradually testing and changing the account. Being proactive was the key to success, because it gave us time to choose and take every step carefully and test different strategies.
It helped us keep performance running smoothly despite all the account restructuring and testing, ultimately leaving enough time to scale the account every month.
With Duty Refunds, this process ran smoothly from week to week. Here at Diligent, we know how essential teamwork is, and how closely we should work with clients to grow their business online.
We are a team that always listens, with young professionals who devote time to carefully planning and executing PPC, Web Analytics, and CRO strategies to increase our clients' online profits.
- SKAG launch, then SKAG to STAG to Broad match
- Full-funnel tracking (GA4, BigQuery, Data Studio)
- Hotjar and LogRocket behaviour analysis
- Landing-page testing, four iterations
- Negative keywords and search-term discipline
- Target CPA scaling without resetting learning
- Conversion migrated to the website funnel
- Display and Facebook channel expansion
