A local landmark in a video thumbnail helped solar ads work in markets where our usual winners struggled. We tried recognizable local scenes across a portfolio of 50+ accounts, so we could compare the idea across markets instead of leaving each buyer to discover it separately. We still couldn't make some markets profitable. Keeping those differences visible was part of the job.
I was consulting for an agency serving separate solar operators across different regions, offers and sales teams. Its customers bought leads. The agency managed the advertising behind those leads, with independent accounts for the different customers.
I helped put together the data infrastructure and built a dashboard that let the agency see which clients were doing well and which weren't. We needed to turn observations from those accounts into actions another media buyer could use.
A win in one account rarely reached the others
Each account was run on its own, and a successful offer or creative in one market rarely made it into the others. Analysts kept working out things a colleague had already found. More accounts meant more opportunities to learn, but that knowledge wasn't reliably available to the next person making a campaign decision.
The accounts were similar enough to compare and different enough to make blind copying risky. They served different regions and used different offers, audiences and funnel structures. A strong result in one place gave us something to investigate elsewhere. It couldn't tell us in advance that another market would respond the same way.
We needed a shared record of what ran, where it ran and how it performed. That context would let buyers reuse a finding without mistaking a local success for a rule about every solar customer.
Dozens of accounts, each managed on its own, each buyer rediscovering what somebody two desks away already knew. When one market found something that worked, nothing carried it to the others.
Shared results became tasks for available buyers
We brought performance data together and tracked offers, creatives, audiences and funnel structures across their different contexts. The findings became playbooks for launching clients and responding to accounts that slipped. The local-landmark test was one example. We could try a recognizable scene in solar-video thumbnails across the accounts and see where the treatment worked.
The playbooks also gave buyers specific actions. We used about eight to twelve creatives and combined them with copy and headline variations. Buyers compared the current day's performance with the previous three days and the previous seven days. The working rule was to stop what was getting worse and keep what was improving.
The agency had about six media buyers. Work reached them through an automated task list rather than a fixed assignment of each buyer to particular accounts. Make.com automations watched for deviations from averages or anomalous results, then generated tasks from a set of around 80 task types. Examples included increasing an ad set's budget by 10% or making changes at the ad or campaign level.
The tasks were randomized and distributed among the buyers. Someone taking time off didn't have to leave an account dependent on their return. The shared dashboard helped the team identify the work. The queue gave available buyers work to do.
Deviation on an account fires a trigger, the trigger writes a ticket from a list of around 80 task types, and a shuffled queue deals it to whichever buyer is on shift. Step through all four.
One test, a local landmark in the video thumbnail, ran in every account at once, won some markets the usual winning ads had never moved, and showed the sales team which markets to work.
Supermetrics collected data across the accounts
I used Supermetrics to aggregate the account data, BigQuery to store it and Google Data Studio to present it. The owner didn't want a large infrastructure budget. Supermetrics made bringing the data together affordable for that engagement.
The agency needed account data collected in one place before it could compare performance. BigQuery held the data, and Data Studio gave the team a report they could read. Shared documentation and testing protocols carried the findings into repeatable work, while Make.com handled the task automations.
That was the stack I used then. The useful distinction for an agency owner is between collecting data, comparing it and assigning the resulting work. A report can expose an account that needs attention. Someone still needs a defined action and responsibility for doing it.
Some markets never became profitable
The written case record reports that new solar clients ramped faster and struggling accounts recovered using patterns found elsewhere in the portfolio. It also reports improvements in agency margin and perceived expertise. The record doesn't quantify those changes. I can't give a defensible percentage lift or a number of days saved from the material available.
My recollection of the individual rescues is less specific than my recollection of the aggregate work. Some markets were difficult to make profitable, and there were markets we never cracked. The data helped us distinguish places where we had a working approach from places where we still had a problem.
The local-scene treatment worked well in our tests. I described the idea as speculation that we tested. The observation doesn't establish what people thought when they saw a thumbnail, or promise the same response in another place.
The agency could inspect the portfolio together. The team could compare accounts, identify patterns and use those patterns in its buying decisions. The unsuccessful markets remained part of that picture.
Two cards, presented identically, one variable between them. The generic product creative stalls; the same offer wearing a local landmark moves. One account running this test learns something about one market. Running it across the whole portfolio at once turned it into a playbook.
Sales needed to know which markets remained difficult
The market differences mattered beyond the media-buying team. Salespeople needed to know where the agency had evidence of a workable approach and where results remained difficult. A successful campaign elsewhere couldn't settle that question for them. The portfolio comparisons gave them information about the limits of the work they were selling.
Account managers had playbooks to start from and a clearer escalation path when an account slipped. Buyers had shared performance information and defined tasks they could pick up across accounts.
I met with the agency weekly, shared insights during the week when I saw trends and helped advise the media-buying team. The dashboard supported those conversations. It gave the team a common set of account results to discuss, including the results that still needed work.
A winning creative could still fail in another market
The landmark experiment gave us a treatment to test elsewhere. It also showed why a portfolio playbook needs to keep the conditions around a result. An ad that worked across many places could still fail in a particular market. A useful record has to preserve that exception alongside the wins.
I remember a similar difference in how often we could show ads to the same audience. Higher frequency worked less well in the California markets we discussed than it did in the South. That's an observation from this engagement. It doesn't establish a geographic rule for another agency's campaigns, and I don't have a supported frequency threshold to recommend from it.
Carry a finding into the next account as a testable starting point. Keep the market and the treatment attached to the result so the next buyer can see what they're trying to repeat.
New accounts can start with tests run elsewhere
This approach fits an agency or network repeating similar campaigns across many customers. The opportunity is to give a new account the benefit of tests already run elsewhere, then compare its results with the contexts those tests came from.
That requires work beyond buying reporting software. Someone has to record the offer and creative treatment, keep results interpretable by market and turn findings into actions buyers can follow. The agency also has to accept evidence that a market remains difficult. A shared dashboard won't make those judgments on its behalf.
The next buyer should be able to find the local-thumbnail test, understand where it helped and see where the agency was still struggling. That's enough to begin a better-informed test in the next account without promising its outcome.