There isn't one correct attribution model. There's a right question to ask, and seven different ways to answer it. Here's the real math behind each one, and why half of them were deprecated by Google while staying essential in your CRM.
THE SHORT VERSION
→There isn’t one "correct" attribution model. Each of the seven answers a different question, and the right one depends on your sales cycle length and buying committee size, not which model sounds most sophisticated.
→The twist most attribution content gets wrong: Google itself deprecated four classic models, first-click, linear, time-decay, position-based, for ad-platform bidding in 2023. Content still treating these as live Google Ads options is out of date.
→Those same models are very much alive at the CRM level. HubSpot’s own attribution documentation still runs first-touch, last-touch, linear, U-shaped, W-shaped, full path, and time-decay for assigning revenue credit across deals, a different layer entirely from ad-platform bidding.
→The credit-split math is documented, not a rumor: U-shaped is 40/40/20, W-shaped is 30/30/30/10, full path is 22.5% four times plus 10%. Which one you pick moves real dollars between real channels.
→Data-driven attribution is a different thing again: an algorithmic model, not a fixed split, and the one Google Ads now defaults to for bidding. It needs real conversion volume to work, so treat it as a later upgrade, not one of today’s seven choices.
→Whichever model you pick, hold it for at least a few quarters before switching. A model change mid-year breaks your own quarter-over-quarter comparisons, which is the fastest way to lose the board’s trust in the number.
→Sales cycle length is the best predictor of the right model. Short cycle, few touchpoints: single-touch. Longer cycle, multiple stakeholders: U-shaped, W-shaped, full path, or time-decay.
→This post explains the seven core models. The next post covers how to actually match one to your specific sales cycle and run a side-by-side comparison before you commit.
A founder pulls up the attribution report before the Monday pipeline review and sees LinkedIn credited with eight deals and organic search with one. Feels decisive. Feels like an answer. Except the marketing lead switches the model from last-touch to first-touch on the same data, in the same meeting, and now organic has six deals and LinkedIn has two. Same deals, same dollars, opposite story, and the only thing that changed is which of seven possible lenses the report was looking through.
Part 1 of this series covered why attribution matters and the three definitions, touchpoint, conversion path, attribution window, that have to be solid before any model makes sense. This post is about that lens: which of the seven models you pick, and why the choice moves real budget, not just a report.
Picking a model is really picking which question about your marketing attribution you care most about right now. But before the seven models, there’s a piece of context almost every "attribution models explained" article gets wrong or leaves out, and it changes how you should read everything that follows.
The models you’ve read about are half deprecated, at the ad-platform level
If you’ve researched attribution before, you’ve probably seen a list of six models: first-click, last-click, linear, time-decay, position-based, and data-driven. Here’s what most of that content doesn’t tell you: Google itself confirms that first click, linear, time decay, and position-based attribution models are no longer supported in Google Ads. Any conversion action still using one of those four was automatically migrated to data-driven attribution. Today, Google Ads gives you exactly two choices for ad-platform bidding: last-click, or data-driven attribution, which uses machine learning to assign fractional credit based on each interaction’s actual measured contribution to conversions.
That matters, but not in the way it sounds. Google didn’t kill the concept of a U-shaped or linear model. It killed them as options for bidding on ads inside Google’s own platform, because machine-learning attribution outperforms fixed rules once you have enough conversion volume. That’s a decision about how Google optimizes ad spend, one layer of the stack, not a verdict on whether those models are useful anywhere else. They’re not. They’re the layer above ad bidding: the one that decides which channel, campaign, or piece of content gets credit for a closed deal across your entire funnel, paid and unpaid. HubSpot’s own attribution documentation still runs the full classic model set for exactly that purpose, and it’s the layer a B2B founder actually needs.
scope note
HubSpot’s current documentation also lists two additional models, J-shaped and inverse J-shaped, gated to Enterprise tier. This post covers the seven core, widely-used models that answer the questions a $1M–$20M B2B firm actually asks. The two newer variants are edge cases for specific funnel shapes, not part of the standard toolkit most teams need first.
The seven marketing attribution models, with the actual math
Here’s what each model does, straight from how HubSpot’s own attribution reporting defines them.
THE 7 MARKETING ATTRIBUTION MODELS AT A GLANCE
Model | Credit split | Best when | Blind to |
|---|---|---|---|
First-touch | 100% to first interaction | Evaluating awareness, top-of-funnel channels | Everything after the first touch, including what closed the deal |
Last-touch | 100% to final interaction | Understanding what tipped the deal over the line | Everything that built demand before the final touch |
Linear | Equal split, every touchpoint | Wanting a neutral, full-spread baseline | Which touchpoints actually mattered more |
U-shaped | 40% first, 40% lead created, 20% middle | Crediting what created awareness and what converted the lead | Everything after lead creation, a real gap on long cycles |
W-shaped | 30% first, 30% contact created, 30% deal created, 10% middle | Defined lead-to-deal process, crediting the full handoff | Late-stage activity after deal creation |
Full path | 22.5% ×4 (first, lead, deal, last), 10% middle | Large marketing and sales orgs, multiple channels | Nothing major, the most complete fixed-rule model |
Time-decay | Weighted by recency, closer touches get more | Long cycles where late activity genuinely mattered most | Early awareness activity that got the buyer in the door |
Credit-split percentages sourced directly from HubSpot’s own attribution documentation. Data-driven attribution isn’t in this table because it uses no fixed split, and lives on the ad-platform layer, not the CRM layer these seven operate on.
Each of these plays out differently on an actual deal. A quick picture of what each one looks like in practice:
01. First-touch hands 100% of a $90K deal to the blog post a buyer read eight months before signing, even if a demo, a proposal, and three follow-up calls happened after. Useful for judging whether your top-of-funnel content is doing its job. Useless for judging whether your sales team is.
02. Last-touch does the opposite: the case study downloaded the week before signature gets full credit, and the four months of newsletter reads that actually built the relationship get none. It’s the default in most tools out of the box, and the most commonly misused, because it’s the easiest number to pull and the least representative of the full journey.
03. Linear treats a single ad click and a four-month nurture sequence as equally important, because it splits credit evenly across everything. Ten touches, each gets 10%, regardless of how much work each one actually did.
04. U-shaped is where the picture starts getting more honest: 40% to the touch that first got the buyer’s attention, 40% to the touch that turned them into a known lead, and the remaining 20% spread across whatever happened in between. It ignores everything after lead creation, which is a real blind spot if your sales cycle runs long after that point.
05. W-shaped adds one more anchor: 30% each to first touch, lead creation, and deal creation, with 10% left for the middle. This is the model most B2B companies with a defined lead-to-deal process actually want, because it credits the full marketing-to-sales handoff, not just the two ends.
06. Full path stretches that same logic across four milestones instead of three, first touch, lead, deal, and the final touch before close, each at 22.5%, with 10% for everything in between. It’s the most complete fixed-rule model on this list, and the one HubSpot recommends for large, multi-channel marketing and sales orgs.
07. Time-decay assumes recent activity matters more than old activity, and weights credit accordingly, so a proposal review two days before close outweighs a blog post read five months earlier. Strong for long cycles where a late push genuinely closed the deal. Weak at valuing the early awareness work that got the buyer into the funnel in the first place.
A separate layer: data-driven attribution
Data-driven, or algorithmic, attribution doesn’t use a fixed split at all. It analyzes historical conversion data and assigns fractional credit based on which touchpoints statistically correlate with closing, comparing the paths of customers who convert against those who don’t. This is the model Google Ads now defaults to for ad-platform bidding, and it’s genuinely the most accurate approach on paper. It’s also the one that needs the most data to work: Google’s own guidance recommends a meaningful volume of monthly conversions before the model has enough signal to be trustworthy, which rules it out for most companies under $5M ARR. Treat it as a future upgrade once volume supports it, not one of the seven models you’re choosing between today.
A worked example: same three deals, four different answers
Take the scenario from the last post: a $4M ARR services firm closes three deals in a month, a $40K deal sourced from a LinkedIn ad two weeks before close, a $90K deal where the buyer read the newsletter for four months before requesting a demo, and a $50K referral intro preceded by a case study download.
Run that through four models and the "which channel worked" answer changes every time. Last-touch says LinkedIn drove $40K, the newsletter drove nothing on the $90K deal despite four months of engagement, and the referral partner gets full credit for $50K the case study may have assisted. First-touch flips it entirely: the newsletter now gets full credit for $90K, and the case study, not the referral partner, gets credit for the $50K deal. Linear spreads credit evenly across every touch on every deal, diluting the newsletter’s real four-month contribution to the same weight as a single ad click. W-shaped is the one that actually reflects what happened: it credits the newsletter meaningfully for the $90K deal because it was the first interaction, while still giving the demo request and referral intro their fair share.
The other three models tell their own versions too. U-shaped would credit the newsletter and the demo request on the $90K deal, but drop the case study on the $50K deal entirely once the referral intro became the lead-creating touch. Full path would spread the $90K deal’s credit across all four milestones, newsletter, lead creation, deal creation, demo, evening out W-shaped’s story further. Time-decay would do the opposite of first-touch: it would hand most of the $90K deal’s credit to the demo request, the most recent touch, and barely credit the four months of newsletter reading that actually built the relationship.
Same three deals, same $180K in revenue. Seven different marketing attribution stories about where to spend next quarter’s budget, and only one of them matches what actually happened.
Where sales cycle length comes in
Every model above included a "best when" and a "blind to" for a reason: the deciding factor is almost always sales cycle length and how many touchpoints a typical deal involves. A two-week cycle with three touchpoints doesn’t need W-shaped or full path, the added complexity buys you nothing a simpler model wouldn’t already show. A five-month cycle with eleven touchpoints and four stakeholders is exactly where single-touch models actively mislead you, because they’re built to ignore most of what actually happened.
That’s the real decision, and it deserves its own post rather than a rushed paragraph here. One thing worth deciding now, though: once you pick a model, hold it. Switching from last-touch to W-shaped mid-year breaks your own quarter-over-quarter comparison, and a board that watched a channel’s number change for no visible reason stops trusting the report, not just the model. This is the same discipline behind what we said in Part 1 about what to actually show a CFO: one number, one model, tracked consistently, is worth more than a technically superior model you switch into halfway through the year.
Frequently asked questions
Does it matter which of the seven models I pick if I’m not running paid ads at all?
Yes, arguably more. Without paid spend forcing a bidding decision, it’s tempting to skip attribution modeling entirely and just eyeball the CRM. But referrals, content, and organic search still need credit assigned across a real conversion path, and last-touch will still quietly overweight whatever your sales team touched last. The model choice matters exactly as much for an all-organic, all-referral pipeline as it does for a paid-heavy one.
What happens if I never touched Google Ads, does the deprecation story even apply to me?
The deprecation itself doesn’t, since it’s specific to Google’s ad-bidding product. But the confusion it causes does: a lot of "attribution models explained" content online was written assuming these six or seven models are still selectable inside Google Ads, and quietly goes stale without saying so. If your attribution work happens entirely in HubSpot, Salesforce, or a CRM, none of that changes for you. Worth knowing so you’re not second-guessing a model choice based on an ad-platform change that never touched your setup.
Can I use more than one attribution model at the same time?
Yes, and most mature B2B teams do, for analysis. HubSpot lets you run several models side by side on the same data, and comparing them is often more revealing than picking one: if first-touch and last-touch show completely different top channels, that gap tells you your buying journey has real middle-of-funnel activity worth crediting properly. The rule from the section above still holds for reporting, though: analyze with several, report to the board on one, held consistently.
My deal volume is too low for data-driven attribution. What should I use instead?
W-shaped or full path, depending on how many milestones your pipeline has. Both are fixed-rule models, so they don’t need a minimum data volume to be trustworthy, and they still credit the meaningful mid-funnel moments, lead creation, deal creation, that a data-driven model would otherwise be discovering algorithmically. Revisit data-driven once your monthly conversion volume grows into the range Google’s own guidance recommends.
How do I know which of the seven models is right for my business?
It depends on your sales cycle length, your typical touchpoint count, and how many people sit on a buying committee. We cover the full decision framework, including how to run a side-by-side comparison before committing, in the next post in this series.
Attribution models are the layer that turns "we track touchpoints" into "we know where to spend." Get the model wrong for your sales motion and you’ll optimize for the wrong signal with perfectly accurate data. Next up: how to actually choose.
Not sure which model fits your sales cycle?
One question. One personalised recommendation for your cycle length and buying committee. No sales call required.