Attribution vs contribution in marketing measurement: the difference, and when to use each
Attribution assigns credit for tracked conversions to the touchpoints on each customer's path, answering who made the sale. Contribution estimates how much each channel added to total revenue, including offline channels and baseline sales that arrive without marketing, answering what drove the sale. Attribution suits daily optimisation. Contribution suits budget allocation.
The two words are often used as if they meant the same thing, and the figures they produce can differ by a wide margin for the same channel. The main reason is the total each one divides. Attribution divides tracked conversions. Contribution divides all revenue, including the share that would have arrived anyway. This page explains how each is calculated, works through an example, compares them on six dimensions, sets out when each is the better choice, and shows how to report both side by side without confusion.
Attribution, defined
Attribution starts from the conversions a tracking system can observe and the touchpoints it can observe on the way to them: ad clicks, views, emails, site visits. It then assigns each conversion to those touchpoints using a rule.
Last click gives all the credit to the final touchpoint. First click gives it to the first. Linear splits it evenly, time decay favours recent touches, and data-driven models learn weights from converting and non-converting paths.
Every rule divides the same total, the tracked conversions, so attributed figures always add up to what the system could see. The method relies on cookies, pixels and user identifiers, and it updates quickly, often daily.
Ad platforms run their own attribution, each counting the conversions it can link to its own ads. When the platform figures are added together they often exceed the conversions the business recorded, because more than one platform claims the same sale.
Contribution, defined
Contribution starts from the outcome itself, usually total revenue or total orders, and estimates how much of it each driver produced. It is typically measured with a marketing mix model, a statistical model fitted to weekly spend, sales and the other factors that move sales.
Some of those drivers are marketing channels. Others are price, promotions, distribution, seasonality and the economy. A separate component, the baseline, holds the sales that arrive without marketing: returning customers, brand demand, habit and word of mouth.
Because it works from aggregate data, contribution covers offline media and needs no user-level tracking. It is usually refreshed on a planning cycle.
The model also captures effects that attribution has no way to express: the delay between spend and response, the point at which extra spend in a channel starts to return less, and the way channels support each other over weeks. Those effects are what a budget plan needs.
The differences, dimension by dimension
The question. Attribution asks which touchpoints get credit for tracked conversions. Contribution asks how much each channel added to the business.
The total being divided. Attribution divides tracked conversions. Contribution divides total revenue, including the baseline.
Coverage. Attribution covers trackable digital activity. Contribution covers every channel with a spend record, online and offline, plus price, promotions and seasonality.
The data. Attribution uses user-level events. Contribution uses aggregate weekly data.
Speed and grain. Attribution updates daily and can report down to the ad and the keyword. Contribution is re-fitted on a planning cycle and reports by channel and campaign.
Best used for. Attribution suits tactical decisions inside channels. Contribution suits budget allocation, forecasting and questions a finance team will review.
Why the same channel gets two different figures
A channel can hold a large share of attributed conversions and a smaller share of contribution, and both figures can be correct for their own scope.
Attribution's total leaves out two things. It leaves out conversions it could not track, such as those driven by television or completed in a store. And it leaves out nothing for the baseline, so conversions that would have happened anyway are still credited to whichever touchpoints appeared on the path.
Both effects make the visible channels look larger relative to the business. Channels close to the purchase, such as brand search and retargeting, are usually affected most, because customers who were already going to buy pass through them on the way.
A worked example
The figures below are illustrative. Take a business with 1,000 orders in a month. Its tracking system can see 700 of them on a digital path: the other 300 came through a store, a phone call or a device that blocks the join.
Attribution divides the 700. Suppose paid search sits on 350 of those paths as the last touch, so under last click it receives 50% of attributed conversions.
Contribution divides all 1,000. Suppose the model estimates that 550 orders would have arrived without any marketing, and that paid search added 120. Paid search then accounts for 12% of total orders and about a quarter of the orders marketing added.
Both numbers describe paid search correctly. The first answers how often it closes a tracked path. The second answers how many orders it produced. A budget decision needs the second.
How large the baseline is
The baseline is the part of revenue that attribution has no category for. Across 138 advertisers, 871 models, the median advertiser had 56% of revenue arriving as baseline, and the middle half fell between 30% and 80%.
On a typical business, then, the baseline is the largest single component of revenue. It varies widely by category, brand strength and the share of repeat customers, so the figure for your own business is worth knowing before comparing channel shares.
Where mixing the two causes problems
Budgets set on attributed share. Moving budget toward the channel with the largest attributed share can favour the channel best placed to capture demand that was already coming.
Brand investment judged on attributed conversions. Brand activity builds the baseline, which attribution does not measure, so brand spend tends to look weak on attributed figures.
Growth targets built on attributed totals. Adding up attributed revenue and calling it marketing-driven overstates what a larger budget can move, because much of that revenue was arriving regardless.
Platform totals treated as the business total. Summing each ad platform's own attributed conversions counts shared sales more than once, so the total can exceed actual orders. Contribution starts from the recorded outcome, so its channel shares cannot add up to more than the business produced.
Changing the attribution rule does not change the total
A common first response is to move to a more sophisticated attribution model, for example from last click to data-driven. That improves how credit is shared among touchpoints.
It leaves the total unchanged. Every attribution model still divides the same tracked conversions among the same tracked touchpoints, so the question of what share of the business marketing produced remains open. Answering it requires a model of total revenue.
The same applies to better tracking. Server-side tagging and improved identity matching recover some of the conversions attribution misses, which narrows one of the two gaps. The baseline gap remains, because it comes from what attribution counts, and more complete tracking does not change that.
When attribution is the better choice
Attribution is the better tool for operational questions: which creative to rotate, which keyword to cut, where a funnel loses people and how long paths take. These are questions about observed behaviour, and attribution measures behaviour in detail.
It is also more reliable where the whole path is visible, for example a logged-in digital product with a short path and no offline sales. And it is the only option when the answer is needed this week.
When contribution is the better choice
Contribution is the better tool for the budget split across channels, for offline and brand media, for forecasting, and for any figure that has to add up to the business. It is also the way to see how much revenue arrives without marketing, which attribution cannot show.
It is the stronger choice where a large share of sales happens offline or on untracked devices, where repeat customers and brand demand make up much of the revenue, and where privacy limits have reduced what attribution can see. In each of those cases the attributed total covers a smaller part of the business, and the gap between the two figures grows.
What contribution does not fix on its own
A contribution model is fitted to historical data, so on its own it can struggle to separate a driver that caused sales from one that moved at the same time. An experiment, such as a geo test, supplies that separation, and calibrating the model with experiment results makes its shares hold up.
Contribution works at channel and campaign level on a planning cycle. It does not guide creative rotation or daily bidding.
It also needs history: enough of it behind each driver, and enough movement in that driver, before the estimates are stable. A channel launched last quarter, or one whose spend has barely changed, comes back with a wide range, and that range is the accurate result for the data available.
How to choose
Choose by what the answer has to add up to.
If it has to add up to the business, such as revenue, growth or the effect of a budget change, use contribution.
If it is about a path, such as which touchpoints, in what order, over how long, use attribution.
If two reports disagree, check what each one sums to. Figures that sum to tracked conversions are attribution, and they describe the tracked subset of the business.
Cassandra is a marketing mix modeling platform that measures contribution. It decomposes the whole outcome, with the baseline as its own component, so baseline sales are reported separately from channel effects.
Its production engine is Bayesian, so each driver's share comes with its uncertainty stated. Geo experiments run in the same system, using the same method family as GeoLift, and their results calibrate the model. Reads sit at campaign level, in the planning layer.
A price is published.
Reporting both numbers
Most businesses keep both. The practical step is to label each report with its scope. An attribution dashboard headed "share of tracked conversions" and a contribution report headed "share of total revenue" no longer look contradictory once both labels are visible.
The next step is to agree in advance which number decides which decision: contribution for budget allocation and growth planning, attribution for creative, bidding and funnel work. Agreeing this before the reports disagree saves most of the argument.
Finally, share the baseline figure. Once a team sees what share of revenue arrives without marketing, attributed totals are read for what they are.
Questions
What is the difference between attribution and contribution?
Attribution assigns tracked conversions to the touchpoints on each path. Contribution estimates how much each driver added to total revenue, including a baseline of sales that arrive without marketing.
What's the difference between MTA and MMM?
Multi-touch attribution, MTA, is a form of attribution based on user-level paths. Marketing mix modeling, MMM, is the usual way to measure contribution, using aggregate weekly data across online and offline channels.
How much revenue is baseline?
Across 138 advertisers, 871 models, the median advertiser had 56%, with the middle half between 30% and 80%. It varies widely by business, so a benchmark is a starting point for your own figure.
Why does attribution give my channels more credit than contribution does?
Attribution divides only tracked conversions and has no baseline, so conversions that would have happened anyway are credited to the touchpoints it can see. Each visible channel looks larger relative to the business.
Is contribution the same as incrementality?
They are related. Contribution divides the outcome among drivers. Incrementality asks what would have happened without a specific driver. A contribution model calibrated with experiments gets close to incrementality.