# SaaS Marketing Attribution > First touch, last touch, linear, time decay, W and U shaped models explained with the math, plus what to do when 70 percent of the buyer journey is invisible. Source: https://saas-marketing.net/guides/saas-marketing-attribution-models/ Topic: SaaS Metrics and Analytics Type: guide Published: 2026-09-11 Last updated: 2026-09-11 Publisher: SaaS Marketing (saas-marketing.net) License: CC BY 4.0. Quote or republish with attribution and a link to https://saas-marketing.net/guides/saas-marketing-attribution-models/ ## Short answer B2B SaaS attribution assigns revenue credit to marketing touchpoints using one of six common rules: first touch, last touch, linear, time decay, W shaped and U shaped. Each rule splits the same deal differently, so the model choice changes which channel looks best without changing what actually happened. Attribution measures credit, not causality. A working setup pairs a declared model with self reported attribution on the form and periodic incrementality tests. ## Key takeaways - Six models applied to one identical five touch journey produce six different channel winners from the same underlying data. - Multi touch attribution is a reporting convention, not a causal measurement, and budget cuts made purely on its output cut the hardest to track channels. - Self reported attribution on the demo form routinely surfaces podcasts, communities and word of mouth that platform data records as direct. - AI answer engines send referrals with no referrer header, so an increasing share of pipeline arrives as direct traffic that no model can see. - Add incrementality testing once a single channel exceeds roughly 30k per month, because that is where geo holdouts become statistically readable. - Lead to account matching, not model choice, is the largest single source of attribution error in accounts with more than five contacts. --- Every attribution vendor sells the same promise: see the full journey, allocate budget with confidence. The promise is false in a specific and interesting way, and understanding exactly how it fails is more useful than picking a model. Attribution answers which touches were present. It does not answer which touches mattered. Those are different questions and they need different instruments. What follows is vendor neutral. One journey, six models, the arithmetic shown, then the parts of the problem no model solves. ## The one journey we will run through every model Take a mid market deal worth 48,000 in first year ARR. Five recorded touches, in order: 1. Organic search on a comparison page, day 0 2. Podcast mention leading to a branded search, day 22 3. Webinar registration, day 41 4. Paid retargeting ad click, day 58 5. Demo request from an email nurture, day 63, opportunity created day 65, closed day 121 That's a clean journey. Real ones have 40 touches, three contacts, and a gap where the buyer read a Reddit thread you'll never see. We'll get to that. First, the math. ## How each model splits the same 48,000 Every model is a rule for dividing one number. Here's what each one pays out on the journey above. | Model | Organic comparison page | Podcast | Webinar | Paid retargeting | Email nurture | | --- | --- | --- | --- | --- | --- | | First touch | 48,000 | 0 | 0 | 0 | 0 | | Last touch | 0 | 0 | 0 | 0 | 48,000 | | Linear | 9,600 | 9,600 | 9,600 | 9,600 | 9,600 | | Time decay (7 day half life) | 1,400 | 3,300 | 7,700 | 16,900 | 18,700 | | U shaped (40/20/40) | 19,200 | 3,200 | 3,200 | 3,200 | 19,200 | | W shaped (30/30/30, rest split) | 14,400 | 4,800 | 14,400 | 4,800 | 9,600 | Read the columns, not the rows. Paid retargeting earns 0 under first touch and 16,900 under time decay, on identical data. The organic comparison page earns 19,200 under U shaped and 1,400 under time decay. Nothing about the business changed. Only the convention did. If your CFO asks whether retargeting is working and the answer moves by 17,000 depending on a setting in a dashboard, the model is not measuring performance. It is allocating credit according to a rule someone chose. Say that out loud in the meeting rather than defending the number. W shaped assumes three moments matter most: first touch, lead creation, and opportunity creation. It's the most defensible default for B2B SaaS because those three moments map to actual stage changes in the CRM. The [six attribution models on one dataset](/examples/attribution-models-one-dataset/) worked example runs the same exercise on a larger dataset if you want to see the spread at volume. ## The identity problem is bigger than the model problem Model choice is a rounding error next to identity resolution. Four failures do most of the damage. Lead to account matching. Your buyer signs up with a Gmail address, her colleague uses the corporate domain, and procurement enters a third contact. Unless your CRM joins all three to one account, three separate "journeys" exist and the deal credits whichever one happened to convert. Gartner puts the typical B2B buying group at six to ten people, so this is the normal case rather than the edge case. Anonymous to known stitching. Someone reads eleven blog posts over four months, then a colleague requests the demo. First touch credit goes to the colleague's Google search. The eleven posts vanish. Dark social. Slack communities, private WhatsApp groups, LinkedIn DMs, a podcast listened to in a car. None of these pass a referrer. They arrive as direct traffic. AI answer engines. This is the newest hole and the fastest growing one. When ChatGPT or Perplexity recommends your product in an answer and the buyer types your name into a browser, you get a branded search with no upstream signal. The work that earned the citation is invisible to every model in the table above. **6 to 10** People in a typical B2B SaaS buying group, per Gartner, each potentially generating a separate unmatched journey ## Channel credit versus channel causality Hold these apart, because conflating them is how budgets get cut wrongly. Credit asks: of the revenue that closed, what touched it? Causality asks: if we stopped doing this, what would we lose? A brand campaign can score terribly on credit and enormously on causality, because it raises the conversion rate of every other channel without ever being the recorded touch. The practical consequence: teams that cut budget purely on multi touch output cut the channels that are hardest to track, not the channels that don't work. Podcasts, community, developer relations and organic word of mouth are all systematically under-credited by every model, because their touches happen off your property. Paid search is systematically over-credited because it sits closest to the conversion and records everything. [Multi touch attribution versus incrementality testing](/comparisons/multi-touch-attribution-vs-incrementality/) covers the trade in detail. A team kills a 15k per month podcast sponsorship because MTA shows 40k in influenced pipeline against 180k for paid search. Six months later branded search volume is down 22 percent and paid search CPCs are climbing, because the podcast was what made people search the brand. The model was never wrong. It was answering a different question than the one asked. ## The stack that actually works, by company size Three instruments, added in order. You don't need all three at once and buying them early wastes money you need elsewhere. Self reported attribution deserves its own note because it's cheap and underused. One field on the demo form: "How did you first hear about us?" Semi open, with five options and an other box. It'll surface podcasts, communities and referrals that your platform data records as direct. It won't split credit accurately, because people name the most memorable thing rather than the first thing. Use it for discovery, not allocation. Pair it with [paid media attribution for SaaS](/guides/paid-media-attribution-for-saas/) when a large share of spend sits in ad platforms whose own reporting claims every conversion twice. ## A maturity model you can actually follow **Attribution maturity, in order** ## What to tell the board Five lines, and they should be the same five every quarter so the trend is readable. Pipeline created by first touch channel. Pipeline created by last touch channel. Self reported attribution distribution. The holdout result if one ran. And a stated confidence level on the whole thing. Do not present a single blended number that implies precision you don't have. The credibility you protect by admitting uncertainty is worth more than the budget you win with a confident chart that falls apart under one good question. The wider metrics context sits in [SaaS metrics and analytics](/saas-metrics/), the tooling options in [B2B SaaS attribution tools](/tools/b2b-saas-attribution-tools/), and the model mechanics in more depth in [B2B SaaS marketing attribution](/guides/b2b-saas-attribution/) and [B2B SaaS attribution models](/guides/saas-attribution-models/). If you're defending a marketing budget line, [sales and marketing spend benchmarks](/research/sales-and-marketing-spend-benchmarks/) gives you an external comparison, and [how marketing moves NRR](/playbooks/nrr-expansion-levers/) covers the expansion side that acquisition attribution ignores entirely. ## What to do next Pull last quarter's closed won deals. Run first touch and last touch on the same list and put the two columns next to each other. Where the two disagree most is where your reporting is hiding something, and that gap is the most useful attribution finding you'll get this month without spending a cent. ## Frequently asked questions ### What attribution model should a B2B SaaS company use? Start with first touch and last touch reported side by side, because the gap between them tells you how much of the journey your model is compressing. Add a W shaped model once you have a defined opportunity stage in the CRM. Below roughly 5 million ARR, spending on a multi touch platform buys precision you cannot act on yet. ### What is the difference between marketing sourced and marketing influenced pipeline? Sourced means marketing created the first known touch on the account. Influenced means marketing touched the account at any point before close. Sourced numbers are small and defensible, influenced numbers are large and easy to dismiss. Report both with the definitions written on the slide, because the argument is almost always about definitions rather than performance. ### How do you attribute revenue in a nine month sales cycle? Attribute at the account level rather than the lead level, and stamp the touch date so credit lands in the period the touch happened, not the period the deal closed. Otherwise Q1 spend shows as Q4 return and every quarterly review misreads the trend. Keep a cohort view of accounts by first touch quarter alongside the revenue view. ### Does multi touch attribution still work without cookies? Partially. Third party cookies are gone in most contexts, so cross site journey stitching is unreliable, but first party tracking on your own domain still works. What you lose is the pre-site journey, which is exactly where dark social, podcasts and AI assistants live. That loss is why self reported attribution stopped being optional. ### What is self reported attribution and does it work? It is an open or semi open field on the demo or trial form asking how the buyer heard about you. It works better than most teams expect for discovering channels, and worse than they hope for splitting credit, since people name the last memorable thing. Use it to find channels platform data misses, not to allocate budget to the decimal. ### What is incrementality testing? You turn a channel off in some geographies or audiences and leave it on in others, then compare pipeline between the two groups. It answers whether spend caused revenue, which no attribution model can answer. It costs real pipeline during the test window and needs enough volume to read a difference, which is why it starts at scale. ### How do you report content impact to a CFO who does not believe attribution? Stop arguing about credit and show three things: the share of closed won accounts that touched a given asset before the opportunity opened, the same figure for lost deals as a control, and one holdout test. Concede the uncertainty explicitly. A number presented with its error bar survives scrutiny better than a precise one that nobody trusts.