The short version
There is no best attribution model in general. There is a best one for your sales cycle, your deal volume and how well your CRM records each stage of a deal.
Start with the decision the model has to support. Splitting budget between channels needs a different model from checking whether your awareness work is pulling its weight.
For most B2B firms with a few dozen trackable conversions a month, a fixed-rule model such as W-shaped beats data-driven attribution, and it only works if your CRM records when leads and deals were created.
Run two or three models side by side on the last 12 months of closed deals before you choose. Then pick one and hold it for at least four quarters.
Our guide to the seven marketing attribution models explains what each one does and how it splits credit. This post answers the next question: which one should you actually run?
It matters more than it sounds. The attribution model you pick decides which channels look like they're working, and so which ones get next quarter's budget. Two firms with identical results will make opposite budget calls if they run different models.
Start with the decision, not the model
Every model answers a slightly different question. Decide which question you need answered first.
"Which channels should get more budget?" You need a multi-touch model that credits the start, the middle and the end of the journey.
"Is our awareness work bringing in new buyers?" First-touch answers this well, as a diagnostic view alongside your main model.
"What tips buyers into booking a call?" Last-touch answers this, again as a diagnostic, not as the basis for budget.
Most founders are asking the first question, so most of this post is about choosing a multi-touch model for budget decisions.
The four things that decide it
1. How long your sales cycle is. The longer the journey, the more of it a simple model ignores. Dreamdata's 2026 benchmarks, built on 3.5 million customer journeys across its customer base, found that 81% of the B2B customer journey now happens outside the sales pipeline, with an average of 88 touchpoints across four channels. A model that only looks at what happened near the deal misses most of it.
2. How many people buy. A deal signed by one person can be read from one contact's journey. A deal approved by a buying committee needs every contact linked to the same company record, or most of the journey disappears.
3. How many trackable conversions you get a month. Form fills, demo requests and other online actions you track. Fixed-rule models work at any volume. Data-driven attribution needs real volume before its output means anything, as we explain in data-driven attribution explained.
4. Whether your CRM records each stage. W-shaped and full path give fixed credit to the touch that created the lead and the touch that created the deal. If your CRM doesn't record those moments reliably, those models are guessing.
Which model fits
Using HubSpot's attribution model definitions, here is a practical starting point.
Your situation | Start with | Why |
|---|---|---|
Fewer than 20 trackable conversions a month | Linear | With so few deals, a weighted model swings on a handful of journeys. Equal credit is stable and honest about what you don't know. Move to W-shaped once your CRM records lead and deal stages reliably. |
Cycle under 3 months, one or two decision-makers | U-shaped | Credits the touch that found the buyer and the touch that turned them into a lead, which covers most of a short journey. |
Cycle of 3 to 6 months, a defined lead-to-deal handoff | W-shaped | Adds credit for the touch that created the deal, so the handoff from marketing to sales counts. |
Cycle over 6 months, several channels, a buying committee | Full path | Credits first touch, lead creation, deal creation and the last touch before close, with the rest spread across the middle. |
A late push genuinely closes most deals | Time decay | Weights recent activity more heavily, at the cost of undervaluing early awareness work. |
Hundreds of conversions a month | Keep a fixed-rule model, test data-driven | Run data-driven as a second view and switch only once its numbers stay stable for several months. |
First-touch and last-touch stay useful as diagnostic views. Just don't set budget on them.
The CRM setup your model depends on
A model is only as good as the records it reads. Before you switch anything on, check five things:
Every paid and social link is tagged. Without consistent UTM tags, paid and social visits show up as direct traffic and the model credits the wrong channel. Our UTM parameters guide covers a naming system that holds up.
Each channel has one name. "Google Ads", "google/cpc" and "AdWords" recorded as separate sources split one channel into three small ones. Lock the source field to a fixed list.
Every contact is linked to a company. Without it, the CFO and the operations lead on the same deal look like strangers, and the committee's journey falls apart.
Lead and deal creation are dated. The CRM needs a reliable record of when someone became a lead and when the deal was opened. W-shaped and full path depend on it.
Closed-won revenue flows back. If attribution stops at the form fill, high-volume, low-value channels look better than they are. Connect closed deals back to the channels that produced them. Our guide to attribution in GA4 and Google Ads covers the setup.
A quick test: if direct traffic is rising while branded searches are flat, or your source report has several names for the same channel, fix the records before you change the model.
Test before you commit
Don't choose a model from a table alone. Run two or three candidates side by side on the last 12 months of closed-won deals, then compare:
Which channels move the most between models. A channel that looks weak under last-touch and strong under W-shaped is usually starting deals that other channels finish.
Whether the result matches what buyers tell you. If the model says a channel barely matters but buyers keep naming it on first calls, trust the buyers and look harder at the tracking.
What you were about to cut. The most expensive attribution mistake is cutting a channel that was quietly helping the others. Check its assists first, as we cover in multi-touch attribution explained.
Pick one and hold it
Once you've chosen, keep the same model for at least four quarters. Switching mid-year breaks every comparison with the previous quarter, and a board that watches a channel's number jump for no visible reason stops trusting the report. Analyse with as many models as you like. Report on one.
Ask how they heard about you
No model sees everything. Podcasts, LinkedIn posts read without clicking, and a recommendation from a former colleague leave no trackable touch. That's the dark social gap.
Close it with self-reported attribution: ask every new prospect how they first heard about you and what made them get in touch now, and record the answer in a fixed CRM field. After two quarters, the channels buyers name most often but your model credits least are the ones you're most likely to be underfunding.
Worked example: choosing a model
// illustrative example based on patterns we see in B2B services firms
Archfield Consulting is an advisory firm at $6M ARR. Its sales cycle averages 110 days, four to seven people sign off on each deal, and it records about 25 trackable conversions a month across Google Ads, organic search, LinkedIn and a monthly newsletter.
It had been running last-touch. That report gave Google Ads 52% of closed-won revenue and LinkedIn 9%, and the founder was about to halve LinkedIn and content spend to put more into Google Ads.
On the grid, a 110-day cycle with 25 conversions a month points to W-shaped. Before switching, the team fixed duplicate source names, linked contacts to their companies and backfilled lead and deal creation dates. Then they ran last-touch and W-shaped on the same 12 months of closed deals.
Under W-shaped, LinkedIn rose to 22% and organic search and content to 27%. Both were starting deals that Google Ads and branded searches finished. Self-reported answers backed it up: nine of the last 20 new clients said a LinkedIn post or an article was how they first came across the firm.
The cut was cancelled. Google Ads stayed at its current budget instead of growing, and W-shaped became the reporting model for the next four quarters.
Not sure your attribution report is telling the truth? In a 30-minute call we'll look at your sales cycle, your CRM records and the model you run, and tell you which one fits.
Frequently asked questions
What is the best attribution model for B2B?
There is no single best model. For most B2B firms with a sales cycle of three to six months and a defined handoff from marketing to sales, W-shaped attribution is a strong starting point. Firms with very few conversions a month often do better with linear, and firms with long cycles and buying committees with full path.
How does sales cycle length affect the choice of attribution model?
Longer cycles involve more touchpoints spread over more months, so single-touch models ignore more of what happened. As cycles lengthen, models that credit several milestones, such as W-shaped and full path, give a fairer picture than first-touch or last-touch.
Is data-driven attribution better than a rule-based model?
Only with enough data. Data-driven attribution learns from conversion paths, so with low monthly volume its output can swing from week to week. Most B2B firms with high-value deals get more stable results from a fixed-rule model, with data-driven run as a second view once volume grows.
Can a company use more than one attribution model?
Yes. Comparing models on the same data is one of the best ways to see which channels start deals and which finish them. For reporting to leadership, it is better to pick one model and keep it for at least four quarters so results stay comparable.
What CRM setup does attribution need?
At a minimum: consistent UTM tags on paid and social links, a fixed list of source names, every contact linked to a company record, dated lead and deal stages, and closed-won revenue connected back to its channels. Without these, any model produces precise-looking numbers built on incomplete records.