Implementing Signals and Attribution for Power Law Businesses

The case is made, and the room agrees. A minority of accounts carries most of the revenue. Every winning ad test pulled in a little more volume and a little less value, the way a dealership’s best billboard filled the lot with the buyers worth the least. The fix is to find the signals tied to the valuable accounts and anchor the bidding to them.

Then someone asks, “So how do we build it?” and the meeting goes quiet.

Marketing owns the ads. A developer owns the forms. Sales operations owns the CRM. Underwriting owns the decisions. Finance owns the revenue. Each of them can describe their own piece, and nobody in the room can trace a single account from the ad that found it to the revenue it produced. Can you?

This guide is the build, end to end. It’s long, because the build has a lot of parts, and every part that’s skipped is a place where the thread breaks. It covers how Broad Eye Solutions recommends any business set up its systems, and in many places it’s exactly how ABPRO_, the Broad Eye Solutions platform, sets up its clients’ systems. Where ABPRO_ handles a piece its own way, the guide says so. The full blueprint is published as an open repository, with the documentation, code, tag manager setup, map files and templates (see the sources), so any team can build it. Teams that would rather not build and run it themselves can have ABPRO_ build and deploy it for them.

The guide, chapter by chapter

1. 1Where the thread breaksWhy most setups can't trace an account back to the ad that found it, and why the bidding learns from whatever you count.2 min read
2. 2One ID and the funnelOne ID that follows every lead through every system, the counts that tie the systems out, and the funnel's stages with a first and last timestamp on each.14 min read
3. 3Capture every leadName every door a lead comes through, tag every link from one vocabulary, save everything at the form, and attribute web leads without cookies.12 min read
4. 4Reports run on PRO, never on raw dataWhy reports read the PRO fields and never the raw data, the order to cut the data in, the map files that keep it consistent, and the first question to settle.12 min read
5. 5One question that sortsThe question that says what a lead is likely to be worth: where to ask it, how to store the answer, where the line goes and what's off-limits.9 min read
6. 6Tell the platformsTwo events and one bidding goal on Meta and Google, which later results go back, and the shortcuts earlier setups took that now cost more than they save.8 min read
7. 7Close the loopTier the book, check each signal before bidding on it, read the funnel by cohort and month, and catch the breaks that show up most.4 min read

Sources: every platform page and document this guide draws on, with the date each was checked.

What to change this quarter

1. Give every form an ID and every lead a hard-coded capture method, and keep a form map that says what each form is and the traffic it was built for. (Every door gets a name)
2. Give every lead one distinct ID, such as the submission’s event ID, save it on the CRM record, carry it into every system that will take it, and keep an ID map for the rest. Name the funnel’s stages, from the entry stage (prospect, conversion or lead) to revenue, keep the stage each record entered at, and give each stage a first and a last timestamp. (One ID across every system, the funnel)
3. Keep every web submission in a submission store as well as the CRM, and tie the counts out every week: clicks to visits, submissions to each platform’s conversions, the store to the CRM, longer calls to call leads. Where two systems share no ID, lay their counts side by side by date. (Every system keeps its own count)
4. Put one parameter template on every platform, from one vocabulary, with the IDs as well as the names. Let the platforms fill in the medium. (Tag every link)
5. Write the capture URL on every web and chat lead, without exception, so the leads without one are known to be off the website. (Capture everything at the form)
6. Run every report on the PRO fields, never on the raw data. Keep the raw data untouched, resolve each PRO field from all of the lead’s evidence through map files, with the evidence behind each value, and use zNAA only where the evidence runs out. Fix the maps, never the CRM. (Reports run on PRO)
7. Add one question about the business, stored as an ordered code, and check it against your accounts before you bid on it. (One question that sorts)
8. Make the qualified conversion the one the platforms bid on. Save first, then send it from the browser and the server with one ID. Keep the answers and codes in your own systems, and send nothing about visitors who opted out. (Two events and one bidding goal)
9. Read it from the top down. Settle whether each platform works, by above-average accounts and revenue per lead, before arguing about which ad does. (Is the platform working?)
10. Read every source, door and cohort by month, at one, three and twelve months. (Close the loop)

Cost per lead will rise once the bidding stops buying the cheapest leads. Put revenue per lead beside it on the same report, so the rise is read against what it bought.

Back in the meeting

The same question comes up a quarter later. Which ad brought in the best account? This time someone looks up the account’s lead ID, and the answer comes back in one line: the ad, the form, the answer it gave, the rep who called, the underwriting decision and the revenue since. Cost per lead is higher than it’s been all year. So is revenue per lead.