Selling what Google Shopping bans

A product feed bolted to a search campaign

ServicesGoogle Ads · Microsoft Ads · Feed Management

The first pass was housekeeping. We cleaned up what was already there, and that alone carried the account to the brief we had been given, a ROAS of 3, inside the projections from the audit and sooner than anyone had planned for.

Turning the feed into search ads

That still left a catalog of several hundred products with no Shopping to put it in. So we took the product feed and hooked it to a Dynamic Search Ads campaign instead. Someone searches for a mortgage search on a legal entity, and the campaign builds a headline out of that product and drops them on its exact page. It is the match Shopping would have made, arriving without the product card.

A plain Dynamic Search Ads campaign would have wandered off across the whole site, spending money on pages nobody buys from and reporting back one blurry number for the lot. Driving it from the feed keeps it on the pages that sell and reports product by product, so VisureItalia can see which documents earn their keep on Google instead of inferring it from site totals.

At the time, that was unusual enough to win an award. Everyone else launched Dynamic Search Ads by pointing Google Ads at the site and letting it crawl, and running one off a feed we managed in-house was new. We took the VisureItalia account to AdWorld Experience as a case study and came away with first prize.

We retired the legacy keyword campaigns and rebuilt on that structure, segmented by price range. DataFeedWatch handles the feed behind it, and the feed work covered 3 things.

• Ad text that fills itself in from the feed, so delivery time, price and the benefit reach the ad without anyone writing a variant per product
• Clean product data, with custom labels for availability, margin and performance
• A live sync, so a change in the catalog reaches the campaigns the same day

Performance Max went in early too, in the first wave after Google launched it.

Deciding what to advertise meant knowing what each document actually earns, and no dashboard could tell us. The true margin on a single record extract takes in the cost of producing it and the effort of prying it out of a public administration, turnaround time included, and none of that reaches a platform report. So we kept a standing conversation running with the client and the people who held those numbers, pushed the products that paid, and set aside the ones that did not.

Then we split the account where the numbers actually differ, so the bidding had something real to work with instead of one blended average.

• Product pages and category pages, which convert at different rates
• Brand traffic by device, since mobile is the weaker of the two and most product-page traffic arrives on desktop

The suspension

Then Google suspended the account outright, on a policy reading that treated VisureItalia’s documents as services, and as government acts in the wording of the ruling. Getting it back took over a month, and nothing we changed inside Google Ads shifted the ruling, because it was about the products rather than the ads. The old account could not be saved, so we rebuilt it from scratch.

The fix turned out to be on the site. We went through the landing pages with VisureItalia and rewrote the wording that had triggered the classification, so the catalog stopped reading to Google as something it would not carry. The rebuilt account went live and has not been disapproved since.

The same structure went onto Microsoft Ads, and the account opened up to Italians living abroad.

One campaign per buyer persona

ServicesMeta Ads · Creative · Market Research

Meta Ads arrived in 2022 in 2 layers. Acquisition runs on lookalike audiences, and remarketing works windows from 7 to 180 days, with the message written for the stage the person has reached.

Meta’s classifier kept reading the property-record ads as housing listings, a special category that forbids audience targeting, and stopped them week after week. We eventually worked out which campaign structures kept those products clear of the classifier and moved them into campaigns of their own. Meta locked the whole account once as well, and getting it back took persistence and the right contacts.

A line per persona

The current phase splits B2B from B2C. Instead of one campaign talking to everybody, the structure runs vertically, with one line for each persona.

• Lawyers
• Accountants
• Companies
• Real estate agencies
• Condominium administrators
• B2C

Each one gets sold on the afternoon it saves them, not on the document.

We built a market research file covering what each persona expects, what they are afraid of, and which angle lands for each product we wanted to advertise. Alongside it went a competitor comparison and a trawl through 1, 2, 3 and 5-star reviews across the category, to find out what buyers say when nobody is selling to them. We drafted the personas in-house, then sat down with VisureItalia and went through them properly before anyone wrote a word.

An AI workflow built on the account’s brand, persona and sales material produced the first pass of the research and later of the ad copy, roughly 80% of the groundwork.

A legal document is a sheet of paper, and it photographs exactly like one. So we gathered reference creative that works when the product itself gives you nothing much to show. The VisureItalia team produced a batch from it, and we went through the results together on hooks, angles and copy. The tone stays institutional, which is how this brand has to sound, and every ad leads with the problem the reader wants gone.

Removing the hesitation in front of checkout

ServicesConversion Audit · UX & Design · Testing & Implementation

For a year we ran the conversion work on the site itself, a customized PrestaShop build looked after by the client’s own developer. The diagnosis came from watching rather than guessing, through heatmaps, session recordings and on-site surveys, which is how the hesitation on the certificate pages surfaced. The site runs to hundreds of pages across products, categories and the blog. We ranked them on what a fix would return against what it would take, then started where that gap was widest, on the product pages and the homepage.

What changed on the pages

• Answers on the page. The questions customer care kept fielding went on the product page itself, high up, in a modular layout.
• Sample images and explainer video, so people can see the document before they buy it.
• A reorganized catalog that follows the routes people actually take, so the main categories surface the documents they are most likely to want.
• A better site search, with metadata added to the products so a search returns what was asked for.

That took volume off customer care and cut down on people ordering the wrong certificate, which on a catalog this technical is an expensive error to unwind.

Testing ran in VWO, on the 3 places where most people dropped off, the checkout, the certificate landing pages and the affiliation pages. The homepage took a design pass and a reworked menu, tested the same way. Not every product page carries enough traffic to call a test, so where the numbers were too thin we asked people directly instead, with Hotjar surveys on the pages in question.

The affiliation pages, which bring in businesses that resell VisureItalia’s services, got design work of their own, drawn for lead generation rather than for ecommerce.

Everything went over as Figma mockups for the homepage, the checkout and the affiliation pages, built with the client’s developer and handed across ready to implement. Over the year, conversion rate climbed 33.48%, bounce rate dropped 48%, and average order value rose 5%.

Finding out which pages were competing for the same searches

ServicesGoogle Search Console · BigQuery · Looker Studio

Search Console is where most people go looking for this, and it runs out of room quickly, at 1,000 rows at a time and 16 months of history. VisureItalia generates millions of rows. So we piped Search Console into BigQuery, where nothing is capped and every query arrives with the page it landed on, the date and the device attached.

Then we ran the queries through a semantic clustering algorithm, grouping them by what people were actually asking for, and matched each cluster against the pages it was feeding. That is where the conflicts showed up. Over the years, pages built around very similar queries had come to share single clusters across several URLs, each one taking a share of the same traffic. Pulling a cluster back onto one page hands that page all of it. VisureItalia has its own SEO agency, so we passed the raw output from the clustering across to them, and they reworked it into the changes they wanted to make.

We built the Looker Studio reporting to match how the business actually thinks, with performance per product and per category, at the granularity it buys and sells at, rather than one number standing in for the whole site.