Data-driven attribution is the most accurate attribution model available, when there's enough data behind it. Most B2B companies can turn it on. Far fewer have the volume to trust what it tells them. Here is what it actually requires and what to run instead.
// THE SHORT VERSION
→ Data-driven attribution (DDA) is not gated behind a hard volume minimum the way most content claims. Google’s own guidance states most new conversion actions are eligible regardless of volume. The real issue is reliability, not eligibility.
→ Google recommends at least 200 conversions and 2,000 ad interactions in 30 days for DDA to perform well, though it will technically run below that. A separate, stricter threshold, 300 conversions and 3,000 interactions, applies only when manually switching an existing conversion action off a legacy rule-based model.
→ GA4 has no publicly documented conversion-volume minimum for its own data-driven model. The commonly cited "~400 conversions a month" figure is widely repeated across the industry, but it traces back to old Universal Analytics guidance, not a current Google-stated GA4 rule. Treat it as a practitioner rule of thumb, not an official threshold.
→ The practical problem for high-ACV B2B firms is volume, not a locked door. A company closing $100,000 deals rarely produces enough monthly conversion events for the model to find real signal instead of noise, even though nothing stops it from turning DDA on.
→ If your volume is thin, position-based attribution combined with self-reported attribution is the more trustworthy approach, not because DDA is unavailable, but because DDA on insufficient data produces confident-looking output that is actually less reliable than a transparent, rule-based assumption.
Data-driven attribution gets mentioned in every discussion about sophisticated marketing measurement. It sounds like the obvious upgrade from rule-based models: instead of assuming how credit should be distributed, let the data decide. The catch: letting the data decide only works when there is enough data to decide from. For most B2B firms, there is not. Running DDA without qualifying data volume does not produce a better answer. It produces a confident-sounding wrong answer, which is worse than acknowledging you need a simpler model.
What data-driven attribution actually does
Every rule-based attribution model applies a fixed formula to credit distribution. Position-based, for instance, gives fixed weight to a small number of milestone touchpoints regardless of what the actual conversion data shows about which touchpoints in your specific business correlate with conversion. Data-driven attribution replaces the fixed formula with a machine learning model trained on your actual conversion data. It analyses all the conversion paths in your history, compares them to the paths that did not convert, and identifies which touchpoints statistically increased the probability of conversion.
The result is an attribution model that reflects how your specific buyers actually behave rather than how a generic model assumes they behave. For a company with sufficient conversion data, DDA produces materially more accurate channel credit than any rule-based model.
Data-driven attribution is not smarter than rule-based attribution by default. It is smarter only when it has enough data to learn from. Below the data threshold, it is less reliable, not more.
What the platforms actually require
Most content on this topic, including earlier drafts of this one, overstates DDA as gated behind a hard technical minimum. It’s worth being precise about what’s actually documented versus what’s practitioner shorthand.
// Table 01 · Data-driven attribution requirements by platform
Platform | What Google actually states | Additional context |
|---|---|---|
Google Ads (new conversion actions) | Eligible regardless of conversion or interaction volume | 200 conversions + 2,000 ad interactions in 30 days is Google’s recommended volume for better model performance, not a requirement |
Google Ads (switching an existing action off a legacy model) | 300 conversions + 3,000 ad interactions in 30 days to become eligible | Drops to 200 conversions + 2,000 interactions per 30 days to remain eligible once switched |
GA4 | No specific conversion-volume threshold is documented by Google | The widely repeated "~400 conversions a month" figure is industry shorthand tracing to older Universal Analytics guidance, not a current GA4-specific Google statement |
B2B MTA platforms (Dreamdata, Ruler Analytics) | Varies by vendor | Typically needs 90 to 180 days of CRM integration, offline conversion data, and UTM consistency to produce reliable account-level output |
These figures for Google Ads and GA4 are drawn from Google’s own Ads Help documentation on data-driven attribution and Google Analytics Help’s attribution settings page. Worth reading directly: Google is more permissive about eligibility than most secondary content suggests, and more quiet about a hard GA4 volume rule than most secondary content assumes.
How to check whether your account actually benefits from DDA
Since eligibility isn’t the real gate, the better question is whether your volume is high enough for the model to find signal instead of noise. Take your average number of monthly trackable conversion events in GA4. For B2B, the relevant conversion events are form fills, demo requests, and any other online action tagged as a key event. Closed deals are offline conversions and will not appear in this count unless you have set up Measurement Protocol.
There’s no official line to check against, but a useful practical signal is stability: if channel credit swings meaningfully week to week with no corresponding change in spend or channel performance, that’s a sign the model is finding patterns in noise rather than signal, regardless of what the eligibility status says.
The conversion volume problem is structural for high-ACV B2B firms. A company closing $100,000 deals at $20M ARR closes roughly 200 deals per year. They will rarely produce the volume of trackable events that gives a machine-learning model enough to work with, even though nothing technically prevents them from turning the model on. Data-driven attribution is not the right goal for this business. The right goal is the most accurate rule-based model with the best available self-reported data supplement.
What happens when you run DDA without enough data
Running DDA below a reliable data volume does not return a neutral result or an error, because there usually isn’t a hard eligibility gate stopping it. It returns a result that looks like a confident attribution model but is based on patterns that are statistically noise rather than signal. The model finds correlations in insufficient data and assigns credit accordingly. Those credits look precise, are expressed in specific percentages, and are wrong in ways that are not visible from the output.
A budget decision based on low-volume DDA output is not better than a decision based on position-based attribution. It is likely worse, because position-based attribution is transparent about the fact that it is an assumed distribution, while low-volume DDA presents noise as data-derived insight.
What to use when your volume is too thin to trust DDA
→ Position-based attribution as the primary model : HubSpot’s own version of this model, commonly called U-shaped, gives 40% credit to the first interaction and 40% to the interaction that created the lead, with the remaining 20% split across everything in between, not first-and-last touch as it’s sometimes loosely described elsewhere. Requires UTM consistency and CRM stage data. Works at any conversion volume. Produces materially better channel credit than last-touch without requiring machine learning. Since GA4 removed position-based natively in November 2023, implement it via Ruler Analytics, Dreamdata, or HubSpot’s custom attribution reporting.
→ Self-reported attribution on every discovery call : Captures dark social, word-of-mouth, podcast, and other channels invisible to any technical model. Two questions per call run consistently for two quarters produces pattern data that supplements the technical model.
→ CRM source data as a sanity check : HubSpot’s original source field and most recent source field give you first-touch and last-touch data. Cross-reference with self-reported data to find discrepancies that indicate misattribution.
Monthly conversions | Primary model | Supplement | Avoid |
|---|---|---|---|
Under 20 | Linear + self-reported | CRM source field | Any algorithmic model |
20 to 100 | Position-based (external tool) | Self-reported + CRM | Data-driven attribution |
100 to 400 | Position-based (external tool) | Self-reported, test GA4 DDA | Treating Google Ads DDA output as stable |
400 to 1,000 | Position-based + GA4 DDA test | Self-reported | Relying on DDA alone |
Over 1,000 | Google Ads DDA | Self-reported for dark social | Rule-based as primary model |
Worked example: why a B2B firm abandoned DDA
Kessler Digital is a B2B marketing consultancy closing an average of 12 deals per month at $22,000 each. Total annual conversion count: 144 deals. They enabled data-driven attribution in GA4, which was technically available to them, nothing blocked it, even though their conversion volume was well below what the model needed to find reliable patterns.
For three months, the DDA model produced channel credit distributions that changed week to week. LinkedIn went from 8% to 31% to 14% credit across consecutive months with no corresponding change in LinkedIn spend or performance. The model was finding different patterns in the noise each week and expressing them as precise percentage credits. A budget decision made in month two increased LinkedIn spend significantly, followed by a budget reversal in month three when LinkedIn credit fell back to 14%.
After switching to position-based attribution via a dedicated tool, channel credits stabilised. LinkedIn settled at a consistent 24% contribution as the interaction that first brought the buyer in. Paid search settled at 29% contribution as the interaction that most often converted the lead. Budget decisions became more stable because the model was no longer chasing statistical noise, and results improved over the subsequent two quarters.
Data-driven attribution on insufficient data does not produce a more accurate picture. It produces a more precisely wrong one. Precision is not the same as accuracy.
Not sure whether your conversion volume supports DDA or which attribution approach fits your current pipeline volume? Also read the attribution models guide for the full comparison of rule-based alternatives, and the GA4 attribution setup guide for how to configure your property correctly. Book a 360 GTM audit with mrktrs →
Frequently asked questions
What is data-driven attribution?
Data-driven attribution is an attribution model that uses machine learning to analyse your historical conversion path data and assign credit to each touchpoint based on how it statistically influenced conversion. Unlike rule-based models which apply a fixed credit formula to every deal, DDA learns from your actual data which touchpoints correlate with conversion for your specific business. It is the most accurate attribution model available when sufficient data is present.
Do you need a minimum number of conversions for data-driven attribution?
It’s more nuanced than a hard minimum. Google states most new Google Ads conversion actions are eligible for DDA regardless of volume, and recommends 200 conversions plus 2,000 ad interactions in 30 days for better model performance. A stricter 300-conversion, 3,000-interaction threshold applies only when switching an existing conversion action off a legacy model. GA4 has no publicly documented volume minimum at all; the "~400 a month" figure often cited is industry shorthand, not an official Google rule. What actually matters for B2B firms with high-value deals is that the model needs meaningful volume to find real signal, whether or not it’s technically "eligible" to run.
What is the difference between data-driven and algorithmic attribution?
Algorithmic attribution is the broader category: any model that uses an algorithm to assign credit rather than a fixed rule. Data-driven attribution is a specific type of algorithmic attribution that uses machine learning trained on your own historical conversion data. Other algorithmic approaches such as Shapley value attribution and Markov chain models are also forms of algorithmic attribution.
Can data-driven attribution work for B2B?
Yes, at sufficient conversion volumes. For B2B firms that generate enough monthly trackable conversion events for the model to find genuine patterns, DDA produces more accurate channel credit than any rule-based model. For most B2B firms at $1M to $20M ARR with high-value deals and long sales cycles, that volume threshold is rarely achievable, and position-based attribution combined with self-reported attribution is more trustworthy than DDA running on thin data.
What is the best alternative to data-driven attribution for B2B?
Position-based attribution, giving 40% credit to the first interaction and 40% to the interaction that created the lead, with 20% split across the middle, combined with self-reported attribution from discovery calls. Position-based handles the trackable digital touchpoints. Self-reported captures dark social, podcast, word-of-mouth, and other untracked touchpoints that influence B2B deals but are invisible to any technical model. Together they produce a more complete picture than DDA on insufficient data.
Does your volume actually support DDA?
One question. One personalised recommendation for your conversion volume. No sales call required.