Marketing Measurement

In one week the ads ran, a press story went out, and the subscriber email went out. Signups rose by 300 that week. At Monday's review the ads owner reports 300, the content owner reports 300, and the email owner reports 300. None of them is lying; all three are looking at the same 300. Faced with a report that adds up to 900, the founder can only ask one thing: where does next month's money go?

Without an answer, budget follows whoever sounds most certain. The channel that felt good last month, the one a confident owner argues for, the one a competitor started doing — the quarter ends somewhere in that drift. The old line about half of advertising spend being wasted with no way to tell which half has survived for nearly a century because this is where it comes from.

The reason it's hard is that people don't arrive through one door. Someone reads a post on a professional network, lands on the blog three days later through a search, subscribes to the newsletter that weekend, and two weeks after that searches the company name directly and pays. The last step in the record is a search, but what moved that person was the post two weeks earlier. Give all the credit to the last step and the search budget keeps growing while nobody is assigned to write the posts that started the whole thing. Give it all to the first step instead and everything that pushed the decision over the line this week disappears. Whichever end you load, one side inflates and the other reads as zero.

So measurement is really two jobs bundled under one name: recording what happened without gaps, and agreeing in advance on how credit gets divided across those records. Without the first there is no material to divide. Without the second, three people look at the same material and produce three different answers. The rule doesn't have to be the fairest one imaginable. It has to be written down, applied to every channel alike, and left alone long enough that this month can be compared with last month.

Everything under marketing measurement splits three ways: tracking, which leaves the record of what happened; attribution, which divides credit across that record; and analysis, which decides what shape to read the accumulated numbers in. Tracking comes first in sequence — with no record, the other two can't start.

Tracking is where the raw material is made, and four tools stack up.

1. UTM parameters — tags appended to a link. The same blog post address goes out with utm_source=newsletter on it in the email and utm_source=ads on it in the campaign. In the analytics tool, visits to that one page then split by route. Traffic sent without tags usually collapses into an undifferentiated direct bucket, and once the send has gone out there is no way to recover the origin after the fact.
2. Tracking pixel — a few lines of code placed on your own pages. The name comes from the original method, which loaded an invisible one-pixel image. Its job is to tell an ad platform who viewed which page, which is how a platform can know whether the person who clicked an ad eventually paid. That feedback is also what lets the platform's own targeting improve, so a broken pixel quietly degrades campaign performance long before anyone notices a reporting gap.
3. Conversion tracking — deciding what counts as success and recording it when it happens. Whether a free signup counts, or inviting a teammate counts, or only a first payment counts changes every number downstream. Changing the definition breaks comparison with last month, so teams touch it rarely and deliberately.
4. Postback — a signal sent back to the ad platform's server saying a conversion occurred. As browsers and phones have restricted tracking records, sending the news server to server has replaced announcing it from the page. That method introduces a gap between when a conversion happens and when the platform learns about it, so yesterday's ad performance often can't be settled this morning.

Attribution is where that record gets divided into credit, and it's the part of measurement people argue about most.

1. Attribution — the rule for which touchpoint gets the credit for one conversion. Giving 100 percent to the last ad clicked is the most common choice and the easiest to compute, but everything that brought the person in earlier scores zero. The alternatives give it all to the first touch, split it evenly across four touches at 25 percent each, or weight touches closer to the conversion more heavily. None of them is correct, so the working practice is to pick one, write it down, and compare only within it. Switch rules and the channel ranking in the same month's report flips wholesale, so a change with no date recorded beside it makes last quarter incomparable.
2. Marketing mix modeling (MMM) — estimating each channel's contribution statistically from years of spend and total revenue, without person-level records. Because nothing has to follow an individual, channels with no click at all — billboards, television — go into the same table as the rest. The trade is that it needs years of accumulated data and its answers arrive by the quarter rather than by the week.
3. Incrementality — the lift that remains after subtracting the results that would have happened anyway. Someone who already decided to buy searches the company name, clicks the ad sitting above the results, and the record shows an ad-driven conversion when all the ad did was spend money. Separating the two takes a group deliberately excluded from the campaign. If that comparison group — the holdout — produces the same signups, the campaign was paying for traffic it already had.

Analysis is where the numbers get read. Cohort analysis groups people by when they arrived and compares those groups over time. If 40 percent of the people who joined in March are still active three months later and only 18 percent of the June group is, something changed in between: the mix of routes, or the quality of the people a campaign brought in. A single blended average hides that, because both groups sit inside it, and the March group is usually large enough to drag the blended line toward the older, better-behaved numbers for months. A marketing dashboard is where those numbers get pinned to one screen. Seeing the same metrics in the same order every week is the entire point, so deciding what goes on it and what comes off matters more than how it looks. Put twenty metrics up and nobody reads it; cut it to five and the row that moved is obvious thirty seconds into the meeting.

The three branches run in one direction. Tracking creates the record, attribution assigns credit across it, and analysis splits the credited numbers by time and by group. However refined the back end is, an empty front end voids it. A campaign sent without tags collects no credit under any attribution rule, and a conversion defined differently by two teams puts two different numbers for the same week on two dashboards.

"This link went out without UTM tags. The email traffic is going to land in the direct bucket."

The record is missing, so this send's performance can't be reconstructed later.

"Under last click, branded search takes everything."

Credit piles onto people who arrived by searching the company name, which makes whatever first moved them read as zero.

"Let's run this one with a holdout."

Hold a group out of the campaign and compare, so we can tell whether the extra signups really came from the ads.

At Pagedeck, Ravi splits the previous month's traffic by channel in the first week of every month. Last quarter Grace asked what would happen if paid spend dropped by 30 percent. Instead of answering, Ravi ran a month with ads switched off in one region and left on everywhere else. Signups in the switched-off region fell 12 percent, which put the real lift from advertising at roughly half of what the dashboard had been showing. Part of that budget moved to writing for search, and the region test became something Ravi repeats once a quarter rather than a one-time answer to one question.

On the business side he keeps definitions aligned with Sofia. Ravi's records hold the route a company came in through at the moment it submitted an inquiry; Sofia's hold whether that inquiry became a contract. Joining the two by company name is the only way to see which article produced a contract, and since a small difference in spelling makes one company look like two, the first step is agreeing on the names. A different thread goes to Mia. When people from one particular campaign create no documents in their first week, that's a sign the ad is promising something the product doesn't, so the two of them look at it together.

The most common mistake is treating measurement as tool installation. Attach an analytics tool, place the pixel, and the numbers should appear on their own — that's how it reads from outside. What actually consumes the time is agreeing on rules. Without a decision about what counts as a conversion, how credit gets divided, and which tool's figure is the official one, three installed tools produce three signup counts for the same week. The conversion figure an ad platform shows on its own screen usually runs higher than your analytics tool, because the platform counts anything that happened within a few days of a click as its own. Bring both numbers into a meeting without knowing why they differ and the hour goes to explaining the gap rather than to the budget. So the rules get written on one page before any tool is installed, and edits get dated as they're made. Six months later, when a number jumps, that's what separates a change in the product from a change in how it was counted.

The second mistake is believing that careful setup produces exact values. Dashboards print decimals, which looks like precision. What comes out is an estimate. Browsers block tracking records, one person moves between a work laptop and a personal phone, and a colleague copies a link into a chat window, which strips the origin off everything that follows. Someone whose card was collected at an offline event isn't in the record at all, and word of mouth — the thing that moves more purchases than most line items on the plan — leaves no trace anywhere by design. Treat these values as exact and move budget on a 3 percent difference, and next month the difference points the other way and the budget moves back. Measurement reads direction and order of magnitude. It can answer whether channel A is roughly twice as efficient as channel B; it can't settle a gap of 3 percent. The bigger the decision, the less weight one tool's number carries on its own — what matters is whether the divided-credit view and the switched-off-ads view point the same way before anyone acts.