Rebuilding the program from the audience up

We set out a new marketing strategy for an internet provider we have worked with for years. It replaced an annual brand campaign and quarterly acquisition pushes with an always-on model: launch creative, test it broadly, retire what underperforms, scale what works, rotate new work in. The model runs on volume – many modest pieces in market, read against the media data – rather than a few expensive ones defended for a season.

The first round had no runway. The program had a fixed launch date, and there was nothing yet in market to learn from. So round one was built the way campaigns usually are: concepts, then production, then launch.

Concepting ran wide. More than twenty-five ideas came out of the first sessions, organized against four messaging themes the program had already locked and six idea tracks meant to push past the obvious. Eleven became full concepts, each scored on six dimensions.

The plan was explicit about the order of operations: test first, then scale what wins. It also scheduled hero production ahead of launch, because a new program needs something in market on day one. Both instructions sat in the same plan, and nothing said which should give way. Five concepts went to full production.

Five hero videos: three shot live and two fully animated, each cut into the lengths and ratios the placements needed. It took about ten weeks from the first brainstorm to camera, and about five months from brainstorm to full deployment.

It had to hold up against the program's own proof. Its strongest third-party endorsement couldn't appear in an ad until the rights were cleared, so one concept was written to run without naming it until they were.

The first batch launched with the program, with about a day to spare, and rolled out over the days that followed. The second batch reached market a few weeks later.

Two months after launch we held a retrospective with our counterpart on the client side. We named the verdict and she agreed: the round had been too clunky, too slow and too expensive – in money, and in both teams' time and attention – for the model it was meant to serve. A five-month cycle can't answer what the media data is saying. The round wasn't wasted, and its hero work is carrying the next campaign. But it had run top-down, the opposite of the model, because a first round had to.

Most of what changed came from her. New work in market roughly every sixty days. A pivot inside a month when an ad stops working. One or two fully produced heroes a year, timed to when people buy. Two concepting pushes a year, which she pointed out had been the plan all along. Production costed against where the media dollars actually go. And the scorecard, she said, had validated decisions more than it drove them – so concepts will be tagged by messaging theme, and the tags will run into reporting, which answers a question the program had only been able to guess at: which message is working.

The first spending decision under the corrected model came days later. A connected-TV proposal had been built to fill the budget available for the rest of the year, and agreed in a planning call. Sales targets for the year were already met. The next day our counterpart asked in writing whether the full amount was the right number or only the available one. It was the available one, and we said so. Rebuilt from the audience up – the homes not yet on the network, at a frequency they could absorb – the buy came to about three-quarters of the budget, with the rest held as flexible funds instead of spent to fill a plan. She approved it within two weeks of asking.

It has no results yet. The first read is due once enough time has passed for a response to show. And when the retrospective summary went out, most of its changes still had no owner or date.

We'd keep scoring a round separately before discussing it. When the scores barely differ, the conversation can skip the argument and go straight to the model. And we'd keep treating a client's question about whether a number is right, or merely available, as worth more than the number. That question is what resized this buy.

We'd watch any plan that asks for two orders of operation at once. If nobody reconciles them, the launch date does. And we'd watch the changes a retrospective agrees on: without an owner and a date before the summary goes out, they wait for the next retrospective.