Attribution model: how to measure which channel actually drives sales

What an attribution model is and how to calculate it

An attribution model is a rule that divides the value of a sale (or lead) among the touchpoints a customer has with your company. Without it, you cannot answer whether a campaign in Google Ads is earning revenue or just collecting traffic that would have come from a referral anyway.

The base formula is simple. For each conversion with a value of V, you break it down into a touchpoint path of length n and assign each touchpoint a weight w_i, where the sum of weights equals 1:

Attributed revenue for a channel = Σ (V × w_i) for all conversions in which the channel appeared

The difference between models lies solely in how you set the weights:

• Last click: the last touchpoint before conversion gets w = 1, the rest get 0.
• First click: the first touchpoint gets 1.
• Linear: each touchpoint gets 1/n.
• Time decay: weights increase exponentially toward the conversion, for example with a 7-day half-life.
• Position-based (U-shaped), 40% to the first touch, 40% to the last touch, and 20% split across the middle.
• Data-driven, where a regression model or a Markov chain calculates the weights based on your actual paths.

In B2B, the problem is that the path rarely ends with a click. It often ends with a phone call, a meeting, and a signed contract 90 days later. That is why what matters in practice is whether the path data comes back to you at all, more than the formula itself. If your form only records the last visit, no model will fix that.

What really affects your marketing attribution results

Five things affect whether your attribution result is good. The order is not random, because the first two factors distort the result more often than the rest combined.

1. Completeness of path data

If 30% of leads in your CRM have an empty source field, you do not have an attribution model. You have an attribution model on 70% of the data and guesswork for the rest. Before you choose any model, check what share of conversions has a complete path: UTM, click, form, phone call, CRM record.

2. Attribution window

In B2B, the sales cycle often lasts from 30 to 180 days. If you set the attribution window to 30 days, campaigns that build awareness at the top of the funnel will disappear from the report. The effect is predictable: the report will show that branded search performs best, and that usually closes traffic someone else generated earlier.

3. Definition of conversion

An attribution model assigns value to an event. If the conversion is a submitted contact form rather than a qualified lead, then attribution measures popularity, not sales. It is worth separating two layers: lead attribution (who brought the contact) and revenue attribution (who brought the customer who paid).

4. Offline channels and referrals

In B2B companies, a large share of sales comes from referrals, trade shows, or conversations at industry events. These touchpoints rarely enter the system automatically. If you do not add them manually, the attribution model will favor digital channels simply because they are measurable.

5. How leads are assigned to sales reps

Picture a sales rep coming back from a trade show, typing "referral" into the CRM, and never linking that lead to the campaign the client saw three weeks earlier. That is how you end up with data quality roughly on par with a phone survey. It is worth enforcing a dropdown list of sources and a first-contact field in your CRM, so you are not entering data that nobody can defend later in a board meeting.

What realistic ranges look like for B2B companies

There is no single number that tells you whether your attribution model is working well. Ranges vary by industry, sales cycle length, and channels. Instead of inventing a benchmark, here are four metrics genuinely worth tracking, and how to interpret them.

Based on projects our agency has delivered, we most often see a buying path with 4 to 9 touchpoints and a cycle of 45 to 150 days. This is our own observation from implementations, not an industry benchmark or market data. If your team falls within these ranges, you likely need a model that does not rely on a single click.

It is also worth remembering that choosing a model is not the goal in itself. The goal is a budget decision. An attribution model should answer the question: where should we add money, and where should we take it away. If your report does not settle that, the model is too weak.

How to improve your attribution model in practice

Here are five actions we implement with clients. The order matters, because each one builds on the previous.

Step 1: Clean up your data collection

Before you change the model, make sure every lead has a recorded source, campaign, and first-contact date. In practice this means: consistent UTMs, a source field in the CRM as a dropdown list, a separate field for the sales rep's first contact, and a requirement to fill in the source for referral leads. If you want to check where the gaps are in your data, a marketing audit.

Step 2: Separate lead attribution from revenue attribution

Lead attribution answers where a contact came from. Revenue attribution answers where the customer who actually paid came from. These are two different metrics, and mixing them in a single report leads to conclusions like "the campaign generates plenty of leads, but nothing comes of it".

Step 3: Set the attribution window to match your sales cycle

If your average cycle is 90 days, a 30-day window will cut off half the journey. Set the window to 1.5 times the average cycle and check how the conclusions change. It often turns out that a channel considered weak suddenly looks very different.

Step 4: Compare two models side by side

You do not have to implement a data-driven model right away. Start by comparing last click with a linear or positional model. If the differences in attributed revenue are large, you have proof that the simplified model distorts the picture. That is an argument that works better in board meetings than theory.

Step 5: Tie budget decisions to the model

An attribution model that does not change how budget is allocated is an expensive report. Set a simple rule: once a quarter, move 10-20% of the budget from the lowest-return channel to the highest-return one. This enforces discipline and shows whether the model actually works.

If you want to turn these conclusions into an action plan for your company, it is worth looking at the bigger picture, in the context of your entire B2B marketing strategy. An attribution model is a tool, not a goal in itself.

Which attribution model to choose: a decision in three steps

The question "which attribution model to choose" comes down to three criteria: the length of the sales cycle, the number of channels, and the maturity of your data.

1. Short cycle (up to 30 days), a single channel - last click is enough. Don't overcomplicate it.
2. Medium cycle (30-90 days), several channels - linear or positional model. Simple to communicate, shows supporting channels.
3. Long cycle (90+ days), many touchpoints - a data-driven or time decay model. However, it requires complete data, otherwise it will give false precision.

One caveat is worth adding. A data-driven model can be worse than a simple one if the input data is incomplete. A statistical model will learn from the errors and confidently point to wrong conclusions. In that case, a simple linear model with manually added offline channels works better.

At Growth Group we work with B2B companies across the region and beyond. Over 16+ years we have managed a large volume of B2B marketing budgets (agency internal data), and the most common mistake we see is missing data on half of the journey, not a bad model. That is why we start with a data audit, not with choosing a tool.

If you want to check how this looks in your case, you can start with a simple exercise: take your last 20 closed clients and try to reconstruct their contact journey. If for more than five of them you cannot identify the first point of contact, you have your answer on where to start.