B2B SaaS Marketing Attribution

A VP of Marketing shows a slide where LinkedIn drove 18 percent of pipeline. Someone asks how a buyer who spent four months lurking in a Slack community, listening to two podcast episodes and reading a competitor comparison on their phone shows up in that number. They do not. They show up as branded search, last touch, which your slide credits to SEO.

This page is the honest version. What each method can and cannot see, how to run the cheap ones properly, and which one to pick at your budget level.

Why last touch breaks on a nine month committee cycle

Three structural reasons, and they compound.

The cycle is long. On a 180 to 270 day enterprise cycle, the first meaningful touch and the closed booking sit in different fiscal quarters, often different budget years. Any model anchored to the conversion event is measuring the end of a story it never read.

This group is large. Gartner puts six or more people in a typical B2B buying group. Your tracked contact is one of them. The security reviewer who killed the deal never visited your site, and the CFO who approved it read one page your champion forwarded as a PDF.

The channels are dark. Slack communities, podcasts, LinkedIn feed posts that were never clicked, a WhatsApp message from a former colleague, an answer from ChatGPT that named you without a link. None of these carry a UTM. This is not a tooling gap you can buy your way out of, and the vendors who tell you otherwise are selling a dashboard, not a measurement.

For the model mechanics themselves, the SaaS marketing attribution models page walks first touch, last touch, linear, U shaped and W shaped in detail. This page is about what to do given that none of them are sufficient.

Self reported attribution, done properly

This is the highest return measurement change available to most B2B SaaS teams, and most implementations are done badly enough to produce noise.

Wording. Ask “How did you first hear about us?” not “How did you hear about us?” The word first changes the answer from branded search to the actual origin. Leave it as free text or free text with a few suggested options, never a closed dropdown, because a dropdown teaches people to pick the first plausible option.

Placement. Last field on the form, after email and company. Required. Making it optional cuts response to a fraction and skews it toward enthusiastic respondents. Making it first increases abandonment.

Volume. Expect 60 to 85 percent usable answers when it is required. A large chunk of the rest will say “Google” or “internet”, which is a category, not a failure. Code it as “search, origin unknown” and track its size over time. If that bucket grows while organic sessions fall, you are watching AI search strip referrers.

The honest limitation: people misremember. Someone who heard your name on a podcast in March and searched for you in July will often say Google. Self reported data is directionally right and specifically wrong, which is fine as long as you never present it to three decimal places.

Incrementality testing at B2B volumes

Incrementality is the only method that answers the question everyone actually has: would this revenue have happened anyway?

The structure is simple. Hold out a comparable group from a channel, run the channel to everyone else, compare outcomes over a period longer than your sales cycle. The problem is volume. Clean statistical reads generally need hundreds of conversions per cell, and most B2B SaaS companies produce dozens of opportunities a month in total.

Test type	Volume needed	What it proves	Realistic for
Geo holdout	High, hundreds of conversions per cell	True channel incrementality	Companies with broad geographic spread and volume
Audience holdout in ABM	Moderate, 200+ accounts per cell	Whether account targeting changes engagement	ABM programmes with defined lists
Blunt on/off test	Low	Direction only, confounded by seasonality	Most companies under $20M ARR
Brand vs non-brand search split	Low to moderate	Whether brand spend is cannibalising organic	Anyone running branded search

The brand search test in the last row is the cheapest useful experiment in B2B and almost nobody runs it. Pause branded paid search in half your geographies for four weeks and watch whether total branded traffic falls or simply shifts to organic. In many B2B accounts most of it shifts, which means a meaningful slice of paid budget is buying clicks you already had.

Lagged cohort reporting, the method to default to

If you take one thing from this page, take this. It is cheap, it needs no vendor, and it is the closest thing to honest that most teams can reach.

Build a table with one row per quarter. Columns: total marketing spend, spend by channel, opportunities created, bookings closed, and bookings closed attributable to opportunities created in that quarter. Then read spend in Q1 against bookings in Q3, because that is when the deals from Q1 spend actually close.

Read that 2026 Q1 row. Spend rose 14 percent, opportunities fell 14 percent, and cost per booked dollar got 45 percent worse. That is a finding you can act on without any model claiming to know which touch mattered. It also survives a hostile board question, because nothing in it depends on a tracking assumption.

Two requirements make it work. The cohort has to be defined on opportunity creation date, not close date. And you need at least six quarters of history before the trend means anything, which is the reason to start the table today even if it is empty.

This depends entirely on knowing your actual cycle length. If you have not measured it recently, the B2B SaaS sales cycle length page covers how to compute the median properly, which is the number that sets your lag.

What multi touch attribution can still do

Multi touch is not useless. It is misused. It can see every touch that carried a tracking parameter and reached a known contact, which makes it genuinely good at three jobs:

• Sequencing inside digital channels. Which page types appear early versus late in deals that closed, which is how you decide what to build next.
• Spotting content that appears in won deals and not lost ones, at a pattern level rather than a credit level.
• Optimising within a single paid channel, where the touches genuinely are tracked and comparable.

What it cannot do is allocate budget across channels, because the channels it cannot see get zero credit by construction. A model that assigns 0.4 percent of pipeline to podcasts is not telling you podcasts do not work. It is telling you podcasts do not have UTMs.

The paid media attribution for SaaS page goes deeper on the within-channel case, and B2B SaaS marketing attribution covers the model comparisons if you need to defend a choice internally. The multi touch attribution vs incrementality testing comparison is the shorter version of the argument in this section.

Which method at which budget

Media mix modelling needs years of spend variation across channels and enough conversion volume for the regression to separate effects. Below roughly 50 million ARR you do not have either, and a vendor selling you MMM at 15 million ARR is selling you a curve fitted to noise. That is an unpopular position and we hold it.

Sourced versus influenced, and how to stop the fight

The sourced or influenced argument is a governance problem wearing a measurement costume. Settle it in writing before the quarter starts.

Define sourced as: marketing created the first qualified contact at the account, with no prior sales activity in the previous 90 days. Define influenced as: a meaningful marketing touch, not a bare pageview, occurred within the open opportunity window. Cap the influence window at the deal’s open period rather than all time, or influenced pipeline drifts toward 100 percent and stops informing anything.

Then report both, every quarter, with the definitions restated on the slide. Put the definitions in your sales and marketing SLA so the argument is settled once rather than monthly. And expect a slice of pipeline that belongs to neither: the deals created by a buyer who was simply in market when a need arose, which is the practical consequence of the 95-5 rule and the reason brand spend never attributes well.

Where your pipeline genuinely originates, across a wider sample than one company’s data, sits in where B2B SaaS pipeline actually comes from.

The position, and what to do this week

Pick one directional method, report it consistently for four quarters, and stop pretending to precision the data cannot support. A board will forgive an imprecise number reported the same way every quarter. It will not forgive three different methods producing three different stories in three consecutive meetings.

Two actions. Add the self reported attribution field to your demo form, required, with the word first in it, and set a 15 minute weekly slot to code the answers. Then open a spreadsheet and build the lagged cohort table for the last six quarters from data you already have in your CRM. Neither costs money. Between them they will tell you more about where your pipeline comes from than any platform you could buy this year, and they connect directly to the funnel arithmetic in the B2B SaaS marketing hub.