Why are the leads down? The causal growth marketing problem

Part 1 of the Marketer Graph series. Start with the overview.

Every week, thousands of insurance marketers across the world get a Monday morning message asking "Why are the leads down?". It is a simple question to ask, a lot of work to answer directionally and a genuinely hard one to answer backed by verifiable data, because the answer is never only in one place and is always a messy, complex interaction of effects.

To answer it properly, marketers write off the rest of their day. They open the dashboard to check what it says (if the pipelines didn't break), they open Meta, Google and TikTok, each of which counts a conversion slightly differently for conversion actions, they either log a ticket that will take 4+ hours to resolve by engineering or open Amplitude or Mixpanel to check event capture and whether the lead events that fired are actually down, they check budgets to see whether spend moved or perhaps shifted between channels. Did competitors start outbidding them on paid channels? They try to pull the latest reach and frequency data on digital out of home, radio or connected TV, which gives no clear indication. They check whether a top creative was paused partway through the month, which means going into Asana or Monday or Jira to find the decision and the date. They check their notes and ping all other channel owners to understand what happened in the other channels. They work out whether the creative that replaced it was a genuinely new concept or a recut of something the audience had already seen. Is it creative fatigue? Did someone push an audience change? What about seasonality, is shopping seasonality dragging lead counts down? They check the CRM to understand whether selective deduplication affected the numbers. Was it the discount offers that were recently paused? Then they tie all of that back to what the business actually counts as a lead, which lives in the CRM and doesn't reconcile with what they're seeing in the ad platforms, event logs or the weekly automated reports. By the time they have a directional answer it is Tuesday or Wednesday, the evidence is scattered across 13 Excel tabs, 10 Slack threads, a text to the person on leave, and the answer is a series of "we think it's mainly".

Now say the issue is identified. That is only half the job, because marketers are already behind goal for the month and the quarter, and now they actually have to fix it. So they start working through the scenarios: how much more to spend and on which channels, what that does to CPL and CAC as the algorithms go hunting for new audiences, whether a sharp increase claws the gap back this month or just spikes cost and bleeds into the next, and what any of it does to the premium actually retained at the end of it. They build another model, pull the team into another call, land on a best guess, and then go and implement it by hand across Meta, Google, TikTok and everything else, one platform and one budget field at a time, hoping the numbers they started from were the right ones. By the time the changes are live, half the month is gone and they are optimizing against a picture that has already moved. Both halves of this, working out why it happened and working out what to do about it, are really the same exercise: establishing cause and effect by hand, from systems that were never wired together to show it.

We have been working on this problem since we started Fount. The rest of this series is how we moved it forward, beginning with how our agents got here.

---

← Overview · Next: The evolution of Fount's agent harness →