Guide

First touch vs last touch attribution: a data-backed guide for small businesses

Vincent RuanFounder, Attrifast · February 28, 2026

Multi-touch attribution sounds more sophisticated than single-touch models^[1]^[2] — but for businesses under 500 conversions a month, that sophistication is statistical noise. Here is the evidence for why first-touch attribution is the right default^[5], and exactly when it is worth switching to something more complex.

We default to first-touch on Attrifast and only nudge customers toward multi-touch when they pass 500 conversions/month and 5+ active channels. The 500-conversion threshold below is from observing where multi-touch starts producing meaningfully different rankings, not a textbook number.

Updated August 2026 · 14 min read

TL;DR

Under 500 conversions/month, first-touch attribution gives the clearest, most actionable signal.
Multi-touch models (linear, time-decay, data-driven) require large datasets to be statistically meaningful.
First-touch answers "where do paying customers discover me?" — the most important question for growth.
Switch to multi-touch only when you exceed 500 monthly conversions and run 5+ marketing channels.
AI discovery strengthens the case for first-touch: when ChatGPT sends the buyer and they return via Direct days later, only first-touch names the channel you can actually invest in.

Attribution models explained

Four canonical models exist^[3]^[4] — first-touch, last-touch, linear, and time-decay. GA4 ships several^[5] but defaults to data-driven, which silently muddies channel-level decisions for small datasets. This page compares the two single-touch models in depth; if you want the full taxonomy including the U-shaped and W-shaped variants, our guide to all seven marketing attribution models walks through each one with its own budget-impact example.

Before comparing models, it helps to understand what each one actually measures. Every model asks the same question — which marketing touchpoint gets credit for a conversion? — but answers it differently.

First-touch attribution

100% of the credit goes to the very first interaction. If a visitor discovered you through Google Organic, that channel receives the full revenue credit regardless of what happened before they bought.

Best for answering: Where do paying customers first discover me?

Last-touch attribution

100% of the credit goes to the final interaction before purchase. The visitor clicked your newsletter and bought? Newsletter gets all the revenue credit.

Best for answering: Which channel closes the deal?

Linear attribution

Equal credit to every touchpoint in the journey. A 4-touch path gives each channel 25%. Fair in theory, but dilutes signal quickly — and requires enough conversions that the averages mean something.

Best for answering: Which channels participate most in the path to purchase?

Time-decay attribution

Touchpoints closer to conversion receive more credit. The final interaction gets the most; early touchpoints get a fraction. A compromise between first-touch and last-touch — but it demands large data volumes to produce stable numbers.

Best for answering: Which touchpoints matter most as purchase intent rises?

Data-driven / algorithmic attribution

Machine learning assigns credit based on statistical patterns across thousands of conversion paths. Google uses this in GA4. It is the most "accurate" model in theory — but it is a black box, and it requires massive data volumes before the model produces anything meaningful.

Best for answering: What does the true marginal contribution of each channel look like at scale?

Same sale, four different answers

A customer takes 11 days and four touchpoints to make a $120 purchase. Watch how each attribution model distributes that revenue differently — and consider which answer would actually change how you spend your marketing budget.

Touchpoint	First	Last	Linear	Time-decay
Day 1Google OrganicDiscovers brand via blog post	$120	—	$30	$12
Day 5Twitter / XEngages with a thread	—	—	$30	$20
Day 9Retargeting adClicks display ad	—	—	$30	$36
Day 11NewsletterClicks email, purchases $120	—	$120	$30	$52

First-touch and last-touch both give you a single, unambiguous signal. Linear and time-decay spread credit across four channels — which sounds fairer, but only produces reliable channel rankings if you have enough conversions for the averages to stabilise. At low volumes, distributed credit is just distributed noise.

Worked example: one $30 ad click, one $49 payment

The four-touch table above shows how credit moves. This one shows what it costs you to get it wrong. Every number here is small enough to check by hand, and the two models end up recommending opposite actions on the same campaign.

The journey

Touchpoint	First-touch	Last-touch
Day 0Google AdsClicks a paid search ad on a high-intent keyword. Cost of that click: $30.	$49	$0
Day 4Google Organic (branded)Searches your brand name, clicks the organic result, subscribes to a $49/mo plan.	$0	$49

The arithmetic on one sale

First-touch: Google Ads is credited $49 against a $30 click. Return on ad spend is $49 ÷ $30 = 1.63x, and the first payment alone nets $49 − $30 = $19.
Last-touch: Google Ads is credited $0 against the same $30 click. Return on ad spend is $0 ÷ $30 = 0.00x, and the campaign books a $30 loss. The $49 lands on branded organic search, which spent nothing.
Same customer, same $49, same $30. The two reports differ by $49 on one line and $30 on another — and nothing about the underlying business changed.

The monthly roll-up: 20 clicks, 8 customers

One sale is a story; a month of them is a budget decision. Run 20 identical $30 clicks for $600 of spend, and suppose 8 of those visitors come back within the week and subscribe at $49. Here is what each model puts in front of you.

Metric	First-touch	Last-touch
Google Ads spend20 clicks × $30	$600	$600
Revenue credited to Google Ads8 of the 20 clickers paid $49	$392	$0
Revenue credited to Google OrganicSame 8 payments, different owner	$0	$392
First-month ROAS on Google Ads$392 ÷ $600 vs $0 ÷ $600	0.65x	0.00x
Cost per acquired customer$600 ÷ 8 customers	$75	Undefined
LTV:CAC at a $441 lifetime value$441 ÷ $75 vs $441 ÷ $0 credited	5.9:1	0:1
Decision this report producesSame month, same money	Keep spending	Pause the campaign

The lifetime value in that table is deliberately conservative: a $49/mo plan held for a median 9 months is 9 × $49 = $441 per customer. Against a $75 acquisition cost that is an LTV:CAC ratio of $441 ÷ $75 = 5.9:1 — above the 3.4 cross-industry median and even the 5.6 top-quartile mark in published 2026 benchmarks^[9]. Pausing the campaign on the last-touch read removes 8 customers a month — $392 of new monthly revenue and $3,528 of eventual lifetime revenue (8 × $441).

And the branded organic search that looked free would shrink too, because those brand queries only happened because someone saw the ad first. Last-touch does not just credit the wrong channel here; it credits a channel whose volume is a downstream effect of the channel it zeroed out.

The honest caveat

First-touch is wrong in the opposite direction: it gives the closing channel nothing, and some of those 8 buyers would have found you anyway. Neither single-touch model is accurate. First-touch is the safer default at low volume because of which way it errs — over-crediting discovery makes you keep an acquisition channel, while last-touch's error makes you cut one. The first mistake costs you some ad spend; the second costs you the pipeline.

When the first touch is an AI engine

The worked example above has a 2026 variant that breaks last-touch even harder. Someone asks ChatGPT for the best tool in your category. Your product is named in the answer. They read about you, close the tab, and three days later type your URL straight into the address bar and pay. There is no ad click and no branded search — just an AI recommendation followed by a Direct session.

First-touch credits the engine

The opening session is the ChatGPT referral, so the revenue lands on ChatGPT. That is a channel you can act on: you can write the comparison page it cited, fix the claim it got wrong, or measure whether Perplexity converts better.

Last-touch credits Direct

The closing session had no referrer, so the revenue lands in Direct. Direct is not a channel — it is the bucket for every visit the analytics tool could not identify. There is no budget line for "more Direct."

That bucket is also dirtier than most dashboards admit. AI engines strip the referrer when a user clicks a link inside an answer, so those sessions arrive looking like someone typed your URL. In our 200-site AI attribution benchmark, a median 34% of GA4 "Direct" traffic was actually AI-referred. So last-touch on an AI-influenced journey does not merely credit the wrong channel — it credits a bucket that is roughly a third mislabelled before you start reading it^[10].

First-touch has the opposite property. It needs exactly one session identified correctly — the first one — to route credit to the right place. Every ambiguous Direct session afterwards is irrelevant to it. That is why AI discovery makes the first-touch default stronger rather than weaker for small businesses.

The caveat is that first-touch only helps if the AI session is detected in the first place, which takes server-side, multi-signal detection rather than referrer parsing alone. And once you are past the 500-conversion threshold, the model that handles these journeys best is a hybrid rather than either single-touch model — we work through that architecture in our breakdown of attribution models for AI search traffic.

Statistical significance thresholds by attribution model

This is the evidence most attribution guides skip. Every model has a minimum data requirement — a conversion volume below which the model's output is statistically unreliable. The more touchpoints a model tries to weight, the more conversions it needs to produce stable, actionable numbers.

First-touch

Reliable for most small businesses

50conv/mo

Last-touch

Reliable for most small businesses

50conv/mo

Linear

Needs more data to distribute credit meaningfully

200conv/mo

Time-decay

Needs even more data for weighted distribution

500conv/mo

Data-driven / Algorithmic

Requires massive datasets — enterprise only

5,000conv/mo

Key insight

Under 500 conversions a month, multi-touch attribution is statistically noise. First-touch gives you the cleanest, most actionable signal at any volume — because it makes no attempt to distribute credit across touchpoints that you may not have enough data to rank reliably.

Attrifast dashboard — first-touch model applied across visitors, orders, revenue, RPV — The full picture under first-touch: every metric (RPV, conversion rate, Revenue) is tied back to the channel that first surfaced the visitor. CAC^[6] and LTV:CAC^[9] calculations downstream remain consistent.

The recommendation: match model to volume

SMB SaaS volume sits well below the threshold where multi-touch models stop being statistical noise^[7]^[8]. First-touch wins by default for the sub-500-conversion segment because it gives you a stable answer at low N.

There is no single "best" attribution model in the abstract. There is only the model that produces reliable output at your current conversion volume. Here is a practical framework.

Under 500 conv/mo

Use first-touch attribution

First-touch tells you the most actionable thing at low volume: where paying customers first encountered your brand. That single data point is enough to make real budget decisions — double down on channels that introduce buyers, cut channels that only attract browsers.

500–2,000 conv/mo

Consider linear attribution

At this volume, distributing credit equally across touchpoints starts to produce stable averages. Linear attribution can help you understand which channels appear across many conversion paths — useful if you are running multiple channels simultaneously and want a fuller picture.

5,000+ conv/mo

Multi-touch and data-driven models

Above this threshold, time-decay and data-driven models begin producing statistically meaningful weightings. This is the territory of established DTC brands and enterprise marketing teams with dedicated analysts.

Which model to use, by business stage and conversion volume

Volume alone is not quite enough. Two businesses at 150 conversions a month behave very differently if one is a three-channel indie SaaS and the other is a nine-channel DTC brand. Find the row where your stage and your monthly paid conversions both fit — and read the last column, which is the condition that should make you move.

Stage	Conversions	Use this model	Why	What would change it
Pre-PMF / first customers	Under 10 / mo	First-touch only	At this volume every model is anecdote. First-touch at least names the channel that produced your first buyers, which is the only attribution question worth answering yet.	Nothing. Do not spend a day on attribution modelling at this stage.
Early traction (indie SaaS, new store)	10–50 / mo	First-touch	One channel almost always dominates early. Distributing credit across four touchpoints hides the very concentration you need to see.	If a single paid campaign is more than half your spend, add last-touch as a secondary read to check it is not just closing demand you already created.
Growing SMB	50–200 / mo	First-touch primary, last-touch secondary	Two single-touch reads side by side cost nothing and expose every journey where discovery and close are different channels — the exact case that breaks budget decisions.	Linear becomes readable at the top of this band, around 200 conversions.
Scaling SMB	200–500 / mo	First-touch primary, linear secondary	Linear averages start to stabilise around 200 conversions per month, so a distributed view finally carries information rather than noise.	Running five or more active channels simultaneously — that is when linear earns its place as the primary.
Established SMB / mid-market	500–2,000 / mo	Linear or U-shaped, first-touch as the sanity check	Enough conversions per channel per month for distributed credit to mean something. Keep first-touch visible so you can spot when the multi-touch model quietly stops crediting acquisition.	Sales cycles longer than 60 days push you toward time-decay instead of U-shaped.
High-volume DTC / multi-channel B2B	2,000–5,000 / mo	Time-decay	Weighted distribution needs roughly 500+ conversions to stabilise and considerably more to stay stable across channels. This band is where the weighting is genuinely informative.	If most journeys are single-touch, time-decay collapses back into last-touch and is not worth the complexity.
Enterprise / large DTC	5,000+ / mo	Data-driven, audited against first-touch	Only above this threshold does an algorithmic model see enough distinct conversion paths to learn from rather than overfit.	If the data-driven model disagrees with first-touch by more than 20% on your top channel, trust neither until you have found out why.

Pre-PMF / first customers

Under 10 / mo

First-touch only

At this volume every model is anecdote. First-touch at least names the channel that produced your first buyers, which is the only attribution question worth answering yet.

What would change it: Nothing. Do not spend a day on attribution modelling at this stage.

Early traction (indie SaaS, new store)

10–50 / mo

First-touch

One channel almost always dominates early. Distributing credit across four touchpoints hides the very concentration you need to see.

What would change it: If a single paid campaign is more than half your spend, add last-touch as a secondary read to check it is not just closing demand you already created.

Growing SMB

50–200 / mo

First-touch primary, last-touch secondary

Two single-touch reads side by side cost nothing and expose every journey where discovery and close are different channels — the exact case that breaks budget decisions.

What would change it: Linear becomes readable at the top of this band, around 200 conversions.

Scaling SMB

200–500 / mo

First-touch primary, linear secondary

Linear averages start to stabilise around 200 conversions per month, so a distributed view finally carries information rather than noise.

What would change it: Running five or more active channels simultaneously — that is when linear earns its place as the primary.

Established SMB / mid-market

500–2,000 / mo

Linear or U-shaped, first-touch as the sanity check

Enough conversions per channel per month for distributed credit to mean something. Keep first-touch visible so you can spot when the multi-touch model quietly stops crediting acquisition.

What would change it: Sales cycles longer than 60 days push you toward time-decay instead of U-shaped.

High-volume DTC / multi-channel B2B

2,000–5,000 / mo

Time-decay

Weighted distribution needs roughly 500+ conversions to stabilise and considerably more to stay stable across channels. This band is where the weighting is genuinely informative.

What would change it: If most journeys are single-touch, time-decay collapses back into last-touch and is not worth the complexity.

Enterprise / large DTC

5,000+ / mo

Data-driven, audited against first-touch

Only above this threshold does an algorithmic model see enough distinct conversion paths to learn from rather than overfit.

What would change it: If the data-driven model disagrees with first-touch by more than 20% on your top channel, trust neither until you have found out why.

Conversion bands follow the statistical thresholds above: linear needs 200+/month, time-decay 500+, data-driven 5,000+. Stage is the second axis because channel count and sales-cycle length move the answer independently of raw volume.

Why Attrifast uses first-touch by default

When we built Attrifast, we made a deliberate choice: first-touch attribution as the default. Not because multi-touch models are wrong, but because the businesses using Attrifast — bootstrapped founders, indie developers, small ecommerce stores — are almost never above the volume thresholds where multi-touch models add clarity.

First-touch also has a practical advantage: it answers the question that matters most for growth-stage businesses. You can optimise a channel that closes deals once you already have buyers in the funnel. But you cannot build a funnel if you do not know where buyers come from in the first place.

Clean signal at low volume

First-touch requires the fewest conversions to produce reliable rankings. A business with 80 sales a month can confidently identify its top acquisition channel.

No touchpoint tracking complexity

Multi-touch models require you to stitch together every session a visitor has ever had. First-touch only needs to capture the original source — simpler to implement, fewer failure points.

Directly actionable

Knowing Google Organic brought 70% of your paying customers is a budget decision waiting to happen. Knowing it contributed 23% of "weighted credit" in a linear model is a data exercise.

Works without cookies

Attrifast is cookie-free and privacy-friendly. First-touch attribution is robust in a cookieless environment — you capture source on the first visit and that is all you need.

Revenue Attribution Track Channel Revenue Multi-Touch for Small Ecommerce Attrifast for Bootstrapped SaaS Traffic Attribution Complete Guide

Attribution model usage among B2B SaaS teams

Sources

Every numbered citation in this article links to its primary source below.

[1]Marketing Attribution Models Compared — first-touch vs last-touch vs multi-touch decision matrix — MCP Analytics (2025).
[2]Mastering multi-touch attribution models — first-touch, last-touch, linear, U-shape, time-decay, data-driven — Matomo Analytics (2025).
[3]Marketing attribution — models and best practices — Adobe (2024).
[4]How Attribution Models Have Changed: Last-Touch vs Multi-Touch — Ruler Analytics (2024).
[5][GA4] Attribution models — data-driven, last-click, first-click, linear, position-based, time-decay — Google Analytics Help (2025).
[6]B2B SaaS Customer Acquisition Cost: 2025 Report — median CAC $702 self-serve, $11,400 sales-led — First Page Sage (2025).
[7]CAC Benchmarks by Channel for 2025 — Inbound $200, Paid Ads $350, Events $500, Outbound $400 — Phoenix Strategy Group (2025).
[8]2024 B2B SaaS Performance Metrics Benchmarks Report — New CAC Ratio jumped 14% to $2.00 — Pavilion (2024).
[9]The LTV to CAC Ratio Benchmark — 2026 cross-industry median 3.4, top quartile 5.6 — First Page Sage (2025).
[10]Why "Direct Traffic" in GA4 Is More Confusing Than You Think — only 10–15% is genuinely direct — AdFixus (2024).

Frequently asked questions

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

First-touch gives 100% credit to the channel where a customer first discovered you; last-touch gives 100% credit to the final touchpoint before purchase.

Which attribution model is best for small businesses?

First-touch attribution is best for businesses under 500 conversions per month — it produces reliable channel rankings with the least data required.

How many conversions does multi-touch attribution need to be statistically reliable?

Linear attribution needs 200+ conversions/month, time-decay needs 500+, and data-driven models require 5,000+ monthly conversions to produce stable results.

When should I switch from first-touch to multi-touch attribution?

Switch to multi-touch when you exceed 500 monthly conversions and are running 5 or more active marketing channels simultaneously.

If someone clicks a $30 ad and buys days later through a different channel, which model credits the ad?

First-touch credits the ad the full $49 payment; last-touch credits it $0 and hands the revenue to whichever channel the buyer returned through. Across 20 such clicks — $600 of spend producing 8 payments — first-touch reports a $75 cost per customer and a 5.9:1 LTV:CAC ratio, while last-touch reports 0.00x ROAS and tells you to pause a campaign that is actually profitable.

Does first-touch attribution work for AI traffic from ChatGPT?

Yes, provided the AI-referred session is detected server-side. AI engines frequently strip the referrer, so those visits fall into Direct by default — in the Attrifast 200-site benchmark a median 34% of GA4 "Direct" traffic was actually AI-referred. First-touch only needs that one opening session identified correctly to route credit to the right engine, whereas last-touch credits a Direct bucket you cannot buy more of.

Can I run first-touch and last-touch attribution at the same time?

Yes, and at 50 or more conversions a month you should. Both are single-touch models, so neither needs extra data to be reliable. The gap between the two reports is itself the signal: channels that rank high on first-touch and low on last-touch are your discovery engines, and channels with the reverse pattern are closers.

What attribution model should a business with 100 conversions a month use?

First-touch as the primary model, with last-touch as a secondary read. At 100 monthly conversions you are well below the 200 needed for linear attribution to produce stable channel rankings, so distributed credit would add noise rather than accuracy.

50% OFF · launch promotion

First-touch attribution that works

See which channel first introduced your paying customers. No complex multi-touch models needed.

One script tag and a Stripe key — live in minutes
Cookieless, so no consent banner for analytics
Every AI referral matched to the payment it produced

See your first-touch data →

7-day free trial · $0 due today · then $9.99/mo · cancel anytime

Attrifast dashboard: prompt-level AI visibility with estimated value, revenue split by channel across ChatGPT, Google, Perplexity, Claude and Direct, competitor position tracking, and per-engine scan settings.