Marketing attribution models: First touch, last touch, linear, W-shaped and data-driven, compared on one B2B deal

The main marketing attribution models are first touch and last touch, which give 100% of the credit to one end of the path; linear, U-shaped and W-shaped, which split credit by fixed rules; and data-driven, which learns weights from converting and non-converting paths. W-shaped gives 30% each to first touch, lead creation and opportunity creation. On one $100,000 deal, each model credits different channels, but the total stays the same.

What are the main marketing attribution models?

Attribution is the bookkeeping of credit. Google's Analytics documentation defines it as "the act of assigning credit for important user actions to different ads, clicks, and factors along the user's path to completing the action." An attribution model is the method that does the assigning. Per the same help page, it can be a rule, a set of rules, or a data-driven algorithm.

The models fall into three groups.

• Single-touch. One touchpoint or channel gets all the credit for the conversion. First touch and last touch each give 100% of the credit to one end of the path.
• Rule-based multi-touch. Credit is shared by a fixed formula. In linear, every touchpoint gets equal credit. In position-based (also called U-shaped), Amplitude describes the usual split: the first and last touchpoints get 40% of the credit each, and the remaining 20% is distributed equally across the middle.
• Data-driven. Instead of a formula, the model learns weights from your own data. Google's version uses machine learning algorithms to evaluate both converting and non-converting paths and applies a counterfactual approach, comparing what happened with what would likely have happened without a given touch.

Amplitude groups the multi-touch side as linear models, rule-based models, and algorithmic models. We keep linear inside the rule-based group because its weights are fixed in advance.

The same deal under four models

As an example, not a benchmark, take one closed deal worth 100,000 with 4 recorded touches. Under linear, each touch gets 100,000 divided by 4, or 25,000. Under position-based, the first and last touches each get 40% of 100,000, or 40,000. The remaining 20% of 100,000, or 20,000, is split between the two middle touches at 10,000 each.

Every column adds up to the same 100,000. The model changes which touch gets the credit, not how much credit there is to give.

One practical note for anyone reporting out of Google Analytics: the first click, linear, time decay, and position-based attribution models were removed in November 2023. The tool now offers data-driven attribution and two last-click variants. If your team still wants rule-based numbers, you will need to compute them outside that tool, and the table above shows the arithmetic. For how this credit feeds pipeline reporting, see marketing-sourced vs influenced.

What is the difference between first-touch and last-touch attribution?

Both models are single-touch rules, as the previous section set out: one touchpoint takes all the credit and every other touch gets nothing. The difference is which end of the path they look at, and so which question they answer.

In our reading, first touch is a demand creation report and last touch is a demand capture report. Neither is wrong. They measure different jobs, which we cover in demand creation vs. demand capture.

A worked example on one deal

Take an example deal worth 100000 that touched 4 channels before it closed. These are example inputs, not a benchmark.

Deal value: 100000
Touches on path: 4

First touch:  touch 1 = 100000, touches 2 to 4 = 0
Last touch:   touch 4 = 100000, touches 1 to 3 = 0

The same deal produces two reports that share no credited channel. If touch 1 was a webinar and touch 4 was a branded search click, one report funds content and the other funds paid search. That is the tension a demand generation leader has to settle before choosing a model.

How Google Analytics defines "last"

Last touch in Google Analytics is not the literal last visit. Google's documentation states that "All attribution models exclude direct visits from receiving attribution credit, unless the path to key event consists entirely of direct visits."

Google's own examples show the effect:

• A path of Display > Social > Paid Search > Direct → 100% to Paid Search
• A path of only Direct → 100% Direct traffic

The documentation also notes that "Paid and organic last click and Last non-direct click are two names for the same attribution model." When a vendor report says last click, check whether direct visits were skipped. A B2B buyer who types your URL after a sales call will hand last-touch credit to whatever they clicked before that.

Matching the model to the funnel stage

HubSpot frames its three report types by stage: you can think of contact create, deal create, and revenue attribution reports as the top, middle, and bottom of your funnel. We think first touch reads most naturally on contact creation and last touch on deal creation, because those are the moments each model is built to see. Run both on the same date range before reading either one alone.

How linear, time decay, U-shaped and W-shaped models split credit

The rule-based models differ in one thing: which moments they decide matter. Linear and U-shaped were defined above, so the table below only places them next to the other two. Time decay weights recency. Per Adswerve, "channels that were visited closer to a conversion receive more credit." W-shaped adds the sales handoff as its own anchor.

The W-shaped rule comes from Prescient AI, which describes it as "30% each to first touch, lead creation, and opportunity creation, 10% split among the middle."

Why the third anchor matters

The U-shaped model, in Prescient AI's words, "can't distinguish a marketing-sourced lead from a sales-qualified one" (Prescient AI). If your team reports on sales qualified leads, that gap shows up fast.

The cost is data discipline. Heeet lists the W-shaped model's main limitation as "Requires CRM integration." We think the opportunity created date is the field most often missing or backfilled. Agree on when it gets set in your marketing and sales SLA before you trust the report.

A worked W-shaped split

This is an example, not a benchmark. Take a $100,000 closed deal with 4 nurture touches between the milestones.

On a $100,000 deal, each of the three milestones takes 30%, or $30,000. The remaining 10% of $100,000 is $10,000.

W-shaped credit per touch
milestone_credit = deal_value * 0.30        (first touch, lead creation, opportunity creation)
middle_pool      = deal_value * 0.10
middle_credit    = middle_pool / number_of_middle_touches

If one touch is both first touch and lead creation, it takes both shares.
If there are no middle touches, decide in advance where the 10% goes.

In our reading, the comparison comes down to your sales motion. Heeet places U-shaped with "PLG, no sales team or quick path from lead conversion to sale" (Heeet). A team where reps qualify every deal needs the third spike.

One B2B deal run through every model

The deal below is an example, not a benchmark. It closed at $100,000 after eight logged touches across seven channels. We read the touches from the customer relationship management (CRM) record, so the closing direct visit counts as a touch here, unlike the analytics default the earlier sections described.

Content syndication created the lead. The sales meeting created the opportunity. Email appears twice, so its credit is summed in the table. For channel context, see our content syndication guide, the B2B webinar playbook and our LinkedIn ads benchmarks.

The path and the arithmetic

Deal value: $100,000 (example input)

#  Channel               Days before close   Role
1  Paid search           56                  first touch
2  LinkedIn ads          49
3  Content syndication   42                  lead creation
4  Email                 35
5  Webinar               28
6  Email                 21
7  Sales meeting         14                  opportunity creation
8  Direct                0                   last touch

Linear:    100,000 / 8 = 12,500 per touch

U-shaped:  first  = 0.40 x 100,000 = 40,000
           last   = 0.40 x 100,000 = 40,000
           middle = 0.20 x 100,000 / 6 = 3,333.33 each

W-shaped:  first, lead, opportunity = 0.30 x 100,000 = 30,000 each
           other 5 touches = 0.10 x 100,000 / 5 = 2,000 each

Time decay, half-life 7 days:
           weight = 0.5 ^ (days before close / 7)
           56d 0.00390625 | 49d 0.0078125 | 42d 0.015625 | 35d 0.03125
           28d 0.0625     | 21d 0.125     | 14d 0.25     | 0d  1
           sum of weights = 1.49609375 (= 383/256)
           credit = 100,000 x weight / 1.49609375
           unit share = 100,000 / 383 = 261.10

Credit by channel and model

Channel               First     Last      Linear    Time decay   U-shaped   W-shaped
Paid search           100,000   0         12,500    261.10       40,000     30,000
LinkedIn ads          0         0         12,500    522.19       3,333.33   2,000
Content syndication   0         0         12,500    1,044.39     3,333.33   30,000
Email (2 touches)     0         0         25,000    10,443.86    6,666.67   4,000
Webinar               0         0         12,500    4,177.55     3,333.33   2,000
Sales meeting         0         0         12,500    16,710.18    3,333.33   30,000
Direct                0         100,000   12,500    66,840.73    40,000     2,000
Total                 100,000   100,000   100,000   100,000.00   100,000    100,000

Read the rows, not the columns. Paid search swings from full credit to almost nothing depending only on the rule. Direct follows the opposite arc. In our reading, the sales meeting and syndication rows show why W-shaped appeals to teams with a lead-to-opportunity handoff: they jump only when those milestones carry weight.

A seven-day half-life is steep. In this example it pushes most credit onto the final week, which we think suits short cycles better than an eight-week deal.

Data-driven attribution cannot be run on one deal by hand. Google's version interprets the model using an algorithm developed by Nobel Prize winner Lloyd Shapley, which needs many converting and non-converting paths. Every rule-based column above also assumes a more linear customer journey than most deals actually follow.

How does data-driven attribution work?

Google documents data-driven attribution (DDA) as a comparison. By comparing the paths of customers who convert to the paths of customers who don't, the model finds the ad interactions that tend to come before a conversion and gives them more credit. Google states that each data-driven model is specific to each advertiser. Your weights come from your own account, not from a shared benchmark.

Google's own example is a tour company. Buyers who clicked a broad Bike tour New York ad first, then a narrower Brooklyn ad, were more likely to purchase a ticket than buyers who clicked only the narrow ad. So the model moved credit toward the broad ad.

How much data the model needs

The thresholds differ by product.

In Google Ads, all conversion actions are eligible regardless of volume, but Google recommends at least 200 conversions and 2,000 ad interactions in a 30-day period. Some types of conversion actions need 3,000 interactions and 300 conversions to qualify. Search Ads 360 sets the bar at 15,000 clicks and 600 Floodlight conversions, and those clicks can come from any advertising platform added to the account. Once the sample is large enough, the initial learning period takes around 24 hours.

On thin data, Google notes that last click and DDA can have the same results. In our reading, a low-volume demo request goal can show "data-driven" in the settings and still behave like last click.

The auto-switch

Google Ads can move conversion actions to DDA for you. Admins receive an email 30 days before the change. You can let it happen, or use the "Switch to DDA" page to opt out or switch earlier. Treat that email as a reporting event. Note the switch date in your pipeline dashboard so a shift in campaign credit is not read as a change in performance.

What the model cannot see

Google Ads DDA scores interactions on Search (including Shopping), YouTube, Display, and Demand Gen ads. In our reading, that leaves out most of what moves an enterprise deal: other ad platforms, events, outbound sequences and sales calls. DDA tells you how to split credit among Google ads. It does not tell you how to split it across the whole buying group. For the bidding side of this, see our Google Ads playbook for software companies.

A practical check: Search Ads 360 lets you set up a custom conversion column and compare last click and DDA totals side by side before you trust either one.

Which models Google Analytics, Google Ads and HubSpot offer today

Each platform documents a fixed menu of models. That menu changed over the past few years, so a model named in an old dashboard or brief may no longer exist in the tool.

Google Ads

Google announced in September 2021 that it was making data-driven attribution the default for all new conversion actions. The rollout began starting in October of that year. Advertisers kept the option to manually switch to one of the five rule-based attribution models.

Google Analytics

The Attribution reports in Google Analytics now list 3 attribution models: data-driven, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position-based were no longer available as of November 2023.

Two rules apply to all three models. Direct visits get no credit unless the path to key event consists entirely of direct visits. If a path has no Google Ads click, Google paid channels last click falls back to Paid and organic last click.

HubSpot

The first four follow fixed rules. Time Decay, for example, gives more credit to interactions that occurred closer to the conversion.

Empirical works differently. It gives credit based on how often each interaction type occurs across conversion paths. Interaction types that occur less often receive more weight, and every interaction of the same type gets equal weight. If page views appear in most paths and form submissions in few, each form submission earns more credit than each page view.

Empirical replaces the U-shaped, W-shaped, J-shaped, and Inverse J-shaped models. This affects the advice in the previous section. HubSpot still records the first touch, lead create and deal create positions. It no longer calculates the W-shaped split from them, so you have to build that split outside HubSpot. Export each deal's interactions with those three position flags, then apply the W-shaped weights from the earlier section in a spreadsheet or warehouse query.

Record the model behind every reported number, using a register like this:

ATTRIBUTION MODEL REGISTER
Report name:
Platform:                 (Google Analytics / Google Ads / HubSpot / warehouse)
Conversion measured:      (key event / conversion action / contact create / deal create / deal revenue)
Model selected:
Model still offered?:     (yes / no, date checked)
Rule-based or data-based:
Built outside platform?:  (yes / no)
  If yes, inputs:         first_touch_flag, lead_create_flag, deal_create_flag
  Formula:                credit = sum(position_weight x deal_amount) per touch
Owner:
Last reviewed:

Where software attribution breaks for B2B buying groups and dark social

Every model on this page divides credit among recorded touches. The weak point in B2B is what gets recorded. Heeet notes that some journeys involve hundreds of touchpoints and multiple stakeholders, while the software keeps its ledger per person.

Contact-level tracking undercounts the buying group

HubSpot's attribution models credit the interactions that created contacts, deals, and revenue, and each interaction position is defined per contact. The deal create position is the contact's last interaction before a deal was associated to them. If a contact is never associated with the deal, that person's touches never reach the deal report.

The capture rules narrow the record further:

• A call needs to be logged to the contact's account to get credit.
• Clicks made from an email sent to the Secondary email address will not be attributed.
• For social post clicks, social media accounts must be connected.
• On very busy deals, HubSpot can process up to 100,000 associations or activities per deal through sampling.

W-shaped attribution adds a dependency on dates. Heeet lists its main limitation as requires CRM integration. The lead create position is the contact's last interaction before they were converted to a contact. A backdated deal or a contact imported late moves the peaks, so credit shifts to the wrong touches.

Take an example, not a benchmark: a $100,000 deal won from a buying group of 4 people. If each person's influence were equal, the fair share per person would be $100,000 divided by 4, or $25,000. A report that only sees the one contact tied to the deal assigns that person's touches the full amount and gives the rest of the group nothing.

Dark social: touches no pixel records

This part is our view. The buyer arrives later through a direct visit or branded search. The model then credits that arrival, and in our reading this inflates direct and search while the real source shows zero.

We think the practical fix is to collect what software cannot see, and to report it next to the model rather than inside it. Two steps:

1. Associate every known buying group member to the opportunity before it moves stages. This fits an account-based marketing motion.
2. Add a required, open-text self-reported attribution field to demo forms, then code the answers monthly.

Self-reported attribution field (demo form)
Question: How did you first hear about us?
Field type: open text, required

Monthly coding buckets:
- Podcast (note the show)
- Peer or colleague referral
- Community (note which one)
- Private message or DM
- Event or conference
- Search or ad
- Other / unclear

Report: count of opportunities per bucket,
shown beside the software attribution report

Which attribution model to use, and when to add self-reported attribution or incrementality tests

Pick the model by the question you need answered, then test the answer before money moves. The model is a lens, not a verdict.

Match the model to the question

First touch answers demand creation. It shows which programs bring new accounts into view, so use it when the question is what to fund at the top of the funnel.

W-shaped answers stage handoffs. It credits the first interaction, lead creation and deal creation, which lets marketing and sales see who moved a deal from one stage to the next. HubSpot frames its contact create, deal create and revenue reports as "the top, middle, and bottom of your funnel, respectively." In our reading, that maps cleanly onto the W-shaped positions.

Data-driven waits for volume. Use it only when the conversion action clears the threshold covered earlier. Note that in Google Analytics, conversions can be reattributed for up to 7 days after the conversion, so read credit after that window settles, not the morning after a campaign ends.

Add self-reported attribution when dark social carries weight

Free text matters: a dropdown only returns the options you thought of. Code the answers monthly and compare them against first-touch credit. The method is in our self-reported attribution playbook.

Run an incrementality test before shifting budget

Attribution assigns credit. It does not prove a channel caused the deal. Google describes its data-driven model as comparing exposed users with "similar users in a holdback group," and you can run the same logic yourself with a geo or audience holdout. See incrementality testing for paid media and the incrementality definition.

As an example, not a benchmark: a 100000 dollar test budget split evenly across 4 test regions is 25000 dollars per region, with matched holdout regions receiving no spend.

1. Write the question: demand creation, stage handoff, or channel mix.
2. Pick the model: first touch, W-shaped, or data-driven if volume qualifies.
3. Add a required free-text "How did you hear about us" field.
4. Code self-reported answers monthly; compare with first-touch credit.
5. Flag any channel where the two views disagree.
6. Run a geo or holdout test on that channel before moving budget.
7. Move budget only on the test result, then re-read attribution.

Once a test settles a channel's value, plug the result into the demand gen budget allocator.

What this page does not cover

This page does not walk through configuring attribution reports inside Google Analytics, Google Ads or HubSpot, and it gives no benchmark credit splits for any channel, since every deal figure here is an example. It also does not cover the statistical design of incrementality tests. Nothing here is financial or legal advice.