Lead source attribution analysis

Lead source attribution analysis is the practice of comparing how different attribution models credit each channel, then using the differences to understand how demand is really created and captured. A single source field answers one question: where did this lead come from. The analysis answers a richer one: which channels open the journey, which assist in the middle, and which close it. By running the same set of leads through first touch, last touch, and multi-touch models, you see a channel from several angles at once.

The analysis matters because no single model is correct. First touch over-credits the channel that started the relationship and ignores everything that moved the buyer along. Last touch over-credits the final click and erases the campaigns that built awareness weeks earlier. The gap between the two is the most useful signal in the whole exercise. A channel that scores high on first touch but low on last touch is a demand creator. One that does the reverse is a demand closer. Funding them as if they did the same job wastes budget.

This is the analytical layer above lead source attribution. Attribution records the source. The analysis interrogates it, comparing models and joining the result to downstream lead conversion rate so channel decisions rest on pipeline value rather than raw lead counts.

The analysis is built from several attribution shares run over the same leads, then compared. The headline measure is model variance: the difference between a channel share under first touch and its share under last touch. The inputs below define a sound, comparable analysis.

1. 1

   A clean attributed lead set

   Start with deduplicated leads that each carry a full touch history, not just a single source. The analysis is only as good as the journey data underneath it, so identity resolution comes first.

2. 2

   Multiple attribution models

   Run the same leads through at least first touch, last touch, and one multi-touch model such as linear or position based. Each model assigns the same total leads differently across channels.

3. 3

   Model variance per channel

   For each channel, subtract its last-touch share from its first-touch share. A large positive number marks a demand creator, a large negative number marks a closer, near zero marks a self-contained channel.

4. 4

   Downstream value join

   Join each channel back to conversion and revenue, not just to lead volume. A channel can carry many leads under every model and still produce little pipeline, which only the value join reveals.

Worked example. Across 1,000 leads, organic search holds a 35 percent first-touch share but only a 20 percent last-touch share, a variance of plus 15 points. It opens many journeys but rarely closes them. Retargeting shows the opposite, 8 percent first touch and 22 percent last touch, a variance of minus 14. Read alone, last touch would cut organic and over-fund retargeting. The analysis shows retargeting only works because organic filled the top of the funnel first.

A metric tree turns the analysis into a structure rather than a set of side-by-side reports. The root is attributed pipeline. The first level splits it by the role a channel plays in the journey: demand creators, assisters, and closers. Each role then decomposes into the channels that fill it and the levers that drive each.

This framing is more useful than a flat channel list because it maps to how budget should be allocated. Demand creators are funded on reach and early engagement. Closers are funded on conversion and intent capture. When pipeline slows, the tree tells you whether the top of the journey dried up, the middle stopped assisting, or the close rate fell, three problems with three different owners.

KPI Tree builds this as a live tree with RACI ownership on each node. The variance between models becomes a tracked signal, so when a channel quietly shifts from creator to closer, the accountable owner is pushed the change. The verified impact loop then checks whether a budget shift made on the analysis actually moved pipeline, closing the gap between the decision and its result.

Benchmarks here are about the shape of the analysis, not a target channel split. What you want is enough journey data to run several models meaningfully, and a clear separation between channels that create demand and channels that close it. Flat agreement across all models usually means the journey data is too thin to see anything.

Use these to judge whether the analysis can be trusted before acting on it. If multi-touch coverage is low, the sophisticated models are guessing, and a simpler comparison of first and last touch is more honest than a multi-touch number presented with false precision.

Improving the analysis means improving the journey data it runs on, choosing models that fit the buying motion, and pushing the result through to revenue so the comparison rewards quality over volume.

The discipline that separates useful analysis from a deck of charts is acting on it and checking the result. KPI Tree carries each channel through the tree to pipeline, surfaces the model variance as a tracked node, and runs a verified impact loop so a reallocation made on the analysis is measured against what actually happened to pipeline in the next period.

1. 1

   Picking one model and defending it

   Choosing a single model and treating its output as the truth discards the comparison that makes the analysis valuable. The insight lives in the differences between models, not inside any one of them.

2. 2

   Running multi-touch on single-touch data

   A multi-touch model fed by leads that each carry only one recorded touch produces confident numbers built on nothing. Fix touch capture before trusting any multi-touch share.

3. 3

   Comparing channels on leads, not pipeline

   A channel can win every model on lead volume and still produce almost no revenue. Without weighting by pipeline value, the analysis rewards cheap volume over real outcomes.

4. 4

   Ignoring how roles drift over time

   A channel that was a demand creator can become a closer as a market matures. Running the analysis once and never again misses these shifts until they show up as a flat lead total.

5. 5

   Presenting precision the data cannot support

   Reporting a channel at 22.4 percent under a multi-touch model implies an accuracy the underlying journey data rarely has. State the confidence the data allows rather than a falsely exact figure.