We audited 1,110 B2B SaaS websites. 84% of sites with ad pixels had no detectable dedicated attribution tool.

Key takeaways

• 84% of sites with detected ad pixels had no detectable dedicated attribution tool (Strict Blind Spot Rate 84.4%, 95% CI 81.2–87.2%; n = 552 sites with ad pixels). If HubSpot, Marketo and Pardot also count as attribution layers, 41.7% still had none detected (Broad Blind Spot Rate).
• More ad channels, only slightly more attribution. Among sites with 5+ ad pixels, only 37% had a dedicated attribution tool detected. Each additional ad pixel is associated with 35% higher odds of having one (logistic OR 1.35, p < 0.001, controlling for company size).
• The LinkedIn Insight Tag appears on 73% of sites with ad pixels, and 80% of those sites had no detectable dedicated attribution tool.
• 78% of ad-pixel detections were invisible in page source. They were found only inside public Google Tag Manager container configuration, so an audit that reads only HTML would miss about three in four.
• Nearly one in three sites using GTM still has Universal Analytics configured inside its container (167 of 534 GTM sites, 31.3%), years after Google stopped processing standard UA data in July 2023.
• Tracking Overlap Rate: 38.5%. Nearly four in ten sites with analytics had two or more analytics tools detected. 55% of sites with product analytics also had GA4.
• Startups and enterprises run different tools. PostHog appears on 36% of 1–10 person startups and 1.5% of 201+ companies. Marketo shows the reverse: 0% to 43%. Within size bands, the two aren’t significantly associated: the split tracks company size.
• Larger companies run larger stacks. The median site has 3 tracking tools detected: 1 at 1–10 employees, 10 at 201+ (Spearman ρ = 0.47, p < 10⁻⁶²).

Scope: a snapshot of 1,110 sites, weighted toward Y Combinator startups and supplemented with 156 selected large companies. It is not a representative survey of all B2B SaaS, and “detected” means found in public page source or configuration, not confirmed firing or spend. Detections from tools with brand-name or error-prone signatures were re-fetched and checked by hand against the matched code. See Methodology and Limitations.

Introduction

Every B2B marketing team says it wants “full-funnel attribution”. Very few say what’s actually on their website.

Most of what we know about B2B measurement comes from two kinds of evidence: surveys, where marketers describe their stack, and vendor benchmarks, built from the customers of a single platform. Both are useful. Neither shows what companies have actually installed.

So we read the stacks directly. This is an exploratory, cross-sectional audit of the tracking tags on the homepage and pricing page of 1,110 B2B SaaS companies, from 3-person seed startups to public companies. We parsed page source, public Google Tag Manager containers, Segment settings and first-party JavaScript bundles, then tested what we found. Because it’s a single snapshot, it shows differences between groups of companies and associations between tools. It can’t show how stacks change over time.

Research questions

• RQ1 · Adoption. Which analytics, attribution and advertising tags do B2B SaaS companies run, and how does adoption vary with company size and category?
• RQ2 · Stack depth. How many tracking tools does a typical site run, and what drives that number?
• RQ3 · Co-occurrence. Which tools travel together, and which almost never meet?
• RQ4 · Measurement gaps. How many sites with ad pixels show no dedicated attribution tag, and how many have more than one analytics tool detected?

Working propositions

This is exploratory research: the propositions below were written to organise the analysis and were not pre-registered. We test each one and report the result, but they should be read as descriptive checks, not confirmatory hypothesis tests.

• P1. Larger companies have larger tracking stacks. (RQ2)
• P2. Tools co-occur in distinct “kits” rather than at random, and startup kits differ from enterprise kits. (RQ3)
• P3. Most sites with ad pixels have no detectable dedicated attribution tool. (RQ4)
• P4. Sites with more ad pixels are more likely to have a dedicated attribution tool. (RQ4)

Definitions

Terms used throughout this report

• Tracking tool: any of the 68 analytics, attribution, advertising, session-replay, visitor-identification, tag-management or marketing-automation products in our signature library.
• Ad pixel: Google Ads, LinkedIn Insight, Meta, Reddit, X, Microsoft UET (Bing), TikTok, Quora, AdRoll, Floodlight, or G2/Capterra conversion tracking.
• Site with ad pixels: a site with at least one ad pixel detected (n = 552). A pixel shows advertising infrastructure, not current campaigns or spend.
• Dedicated attribution tool: a product whose vendor positions marketing or revenue attribution as a core product, not just a report inside a broader tool (e.g. Usermaven,Bizible/Adobe Marketo Measure, HockeyStack, Dreamdata, Factors.ai, Cometly, Fibbler, Ruler, Attributer). Multi-layer platforms that meet this rule count in each layer they serve but once in the tools-per-site total. Under this rule Usermaven, the publisher, counts as both analytics and attribution. GA4, Adobe Analytics and Amplitude include attribution reports but are classified as analytics.
• Strict Blind Spot Rate: share of sites with ad pixels where no dedicated attribution tool was detected.
• Broad Blind Spot Rate: share of sites with ad pixels where neither a dedicated attribution tool nor a marketing-automation platform with built-in attribution reporting (HubSpot, Marketo, Pardot) was detected.
• Tracking Overlap Rate: share of sites with at least one web or product analytics tool detected that have two or more detected (legacy Universal Analytics excluded).
• GTM-only detection: a site–tool detection whose only source was the site’s public Google Tag Manager container configuration (not page HTML, a first-party bundle or Segment).
• Detected means present in public page source or configuration. It doesn’t prove a tag fires, and absence of a tag doesn’t prove a company lacks the capability: attribution can also run server-side, inside a CRM, or in GA4 (see Limitations).

Related work

The closest public benchmark is the HTTP Archive Web Almanac, whose Third Parties chapter measures third-party scripts across millions of pages and identifies googletagmanager.com as one of the top third-party domains on the web [1]. It describes the whole web, not B2B software, and it does not classify tools by their role in a marketing stack. Commercial technology-lookup services also report which tools a site uses; we did not compare our detection against them. Vendor benchmarks from HockeyStack Labs [2], Dreamdata [3] and Amplitude [4] report funnel and spend metrics from their own customer bases, so by design they describe companies that already run an attribution or analytics platform.

This study adds three things: a B2B-specific sample stratified by company size, parsing of public GTM container configuration (where we found most ad tags), and formal tests of which tools co-occur.

Full method, the tool classification list and a description of the published data are at the end.

Finding 1: Attribution is the thinnest layer

Google tools form the base of most stacks, and dedicated attribution is the least common layer

What the data shows. 84.4% of sites have at least one tracking tag detected. The typical site has 3 distinct tracking tools (mean 4.4; 90th percentile 11; maximum 20). Google dominates the base layer: Google Analytics 4 is on 52.9% of sites, Google Tag Manager on 48.1% and Google Ads on 38.9%. The LinkedIn Insight Tag (36.4%) is the second most common ad pixel, well ahead of Meta (17.7%). On these B2B sites, LinkedIn, not Facebook, is the most common paid social pixel.

What’s missing matters as much as what’s there:

• Dedicated attribution tools appear on just 8.2% of sites. The most common are Bizible/Adobe Marketo Measure (3.3%), HockeyStack (2.3%), Dreamdata (1.8%) and Factors.ai (1.2%). Ecommerce-style attribution tags (Hyros, Northbeam, Rockerbox, Triple Whale) were almost never detected, as you’d expect in a B2B sample.
• Product analytics (PostHog, Amplitude, Mixpanel, Heap, Pendo) appears on 32.6% of sites, overwhelmingly PostHog (25.2%).
• Session replay appears on 21.2%. Microsoft Clarity (12.4%) is now well ahead of Hotjar (7.2%).
• Privacy-first web analytics appears on 14.8%, almost all Cloudflare Web Analytics (12.3%). Plausible (1.6%), Fathom and Matomo (0.5% each) remain niche in B2B SaaS.

Stack layer	All sites	1–10	11–50	51–200	201+	χ² (df=3)	Cramér’s V
Any tracking tag	84.4%	77.0%	86.3%	89.4%	91.5%	27.1***	0.16
GA4	52.9%	28.7%	60.5%	71.6%	71.4%	145.8***	0.36
Tag manager / CDP	50.3%	18.0%	55.6%	75.2%	84.4%	291.5***	0.51
Any ad pixel	49.7%	26.1%	53.2%	70.2%	73.9%	157.4***	0.38
Marketing automation / CRM	34.9%	9.9%	39.5%	46.1%	65.8%	200.5***	0.43
Product analytics	32.6%	37.1%	35.4%	29.1%	21.1%	17.6***	0.13
Visitor ID / ABM	23.2%	11.2%	20.4%	17.0%	55.8%	154.3***	0.37
Session replay	21.2%	6.5%	23.3%	33.3%	36.7%	91.4***	0.29
Privacy-first analytics	14.8%	19.6%	12.9%	7.8%	14.1%	13.6**	0.11
Dedicated attribution	8.2%	1.0%	4.7%	12.8%	25.6%	116.8***	0.32

n = 383 / 387 / 141 / 199 sites by size band. ** p < 0.01, *** p < 0.001. Reading Cramér’s V: about 0.1 is a small effect, 0.3 moderate, 0.5 large.

Why it matters. Across the sample, dedicated attribution is the least common layer (8.2%), far behind ad pixels (49.7%). Ad pixels outnumber dedicated attribution in every size band, from about 26 to 1 at 1–10 employees to about 3 to 1 at 201+. Ad pixels appear on 70% of 51–200 employee sites; dedicated attribution reaches only 26% even at 201+. Tag management shows the largest size effect of any layer (V = 0.51), which matters for Finding 3. Because this is a snapshot, these figures describe prevalence, not the order in which companies install tools.

Caveat. Cloudflare Web Analytics can be switched on at the CDN level, so part of its adoption may reflect infrastructure defaults, not a deliberate analytics choice.

Finding 2:  Bigger companies, bigger stacks

Larger companies run larger stacks, and developer-tool companies run lighter ones

What the data shows. Tools per site are higher in each successive size band.

Size (employees)	Mean tools	Median	75th percentile
1–10	2.0	1	3
11–50	4.3	4	7
51–200	5.7	5	8
201+	8.6	10	13

The difference between size bands is large and significant (Kruskal–Wallis H = 250.2, p < 10⁻⁵³). Spearman’s ρ between size band and tool count is 0.47, a moderate positive association: larger companies tend to have more tools detected. P1 is supported. This compares different companies at one moment; it doesn’t track any company’s stack as it grows.

Category matters too (Kruskal–Wallis H = 52.6, p < 0.001):

• Sales, Marketing and HR & Recruiting software companies have the heaviest stacks (median 5 tools) and the highest ad-pixel rates: 63–68% have at least one.
• Developer-facing categories (Engineering & Product, Infrastructure, Security) have a median of 2, lean on PostHog and Vercel or Cloudflare analytics, and have lower ad-pixel rates (40–42%).

Why it matters. Martech sellers run the most martech, which is unsurprising. The more useful point is that developer-tool companies show a different, lighter stack centred on product analytics. A single “standard B2B stack” benchmark hides two different markets.

Caveat. Category is partly confounded with size, because the large-company cohort is not spread evenly across categories. We report category differences as descriptive, not causal.

Finding 3: The hidden ad stack

Most of the B2B ad stack is invisible in page source

What the data shows. Half of all sites (49.7%) have at least one ad pixel detected, and 26.8% have three or more. Among sites with ad pixels, Google Ads is on 78.3%, the LinkedIn Insight Tag on 73.2%, Meta on 35.7% and Reddit on 31.2%. Among all 201+ employee sites, Reddit (42.7%) is more common than Meta (31.2%) and Microsoft Ads/Bing UET reaches 55.3%, against 1.0% among 1–10 person companies. Bing UET shows one of the strongest size effects in the dataset (Cramér’s V = 0.54).

The central methodological result: 77.9% of ad-pixel detections were found only inside the site’s public Google Tag Manager container, with no trace in page HTML, bundles or Segment. That is 1,344 of 1,726 site–ad-pixel detections whose only source was the container. Across all tool types, 51.5% of detections (2,537 of 4,927) were GTM-only.

Why it matters. Studies that search homepage HTML for “fbevents.js” or “insight.min.js” will badly undercount B2B ad pixels. In this sample, most ad tags were deployed through GTM, so a credible audit, internal or external, has to read the container. A review of a site’s source code alone won’t show most of its ad tags.

Caveat. A tag in a GTM container may be paused or limited to certain pages. GTM detection shows configuration, not confirmed firing.

Finding 4: Startup kits vs. enterprise kits

Tools co-occur in distinct kits, one common at startups and one at enterprises

What the data shows. We tested all 378 pairs among the 28 tools found on at least 30 sites. For each pair we computed the phi coefficient (φ) and lift and tested independence with a 2×2 chi-square test, or Fisher’s exact test for the 102 sparse pairs where an expected cell count was below five. We applied Benjamini–Hochberg correction across all 378. 241 of 378 pairs (63.8%) co-occur significantly (adjusted p < 0.05). As a check for size effects, we re-tested every pair within company-size bands (Cochran–Mantel–Haenszel). 192 pairs stay significant with the same direction.

Pair	Sites with both	φ	Lift	Conditional overlap
GTM + LinkedIn Insight Tag	371	0.66	1.9×	92% of LinkedIn sites use GTM
GTM + Google Ads	386	0.66	1.9×	89% of Google Ads sites use GTM
GA4 + GTM	462	0.65	1.6×	79% of GA4 sites use GTM
Google Ads + LinkedIn Insight Tag	326	0.65	2.1×	81% of LinkedIn sites also have Google Ads
Bing UET + Marketo	75	0.54	5.1×	79% of Marketo sites have Bing UET
Marketo + 6sense	49	0.53	7.3×	52% of Marketo sites have 6sense
Marketo + Bizible	29	0.46	9.2×	78% of Bizible sites have Marketo

All adjusted p < 10⁻²⁴ (Fisher’s exact test for Marketo + Bizible, chi-square for the rest). All seven stay significant within size bands.

Three clusters emerge:

1. The Google + LinkedIn core: GA4, GTM, Google Ads and LinkedIn Insight. 87 sites have exactly these four, and none of the other three, of the seven most common tools.
2. The enterprise ABM kit: Marketo, 6sense, Bizible, Bing UET, Floodlight and G2/Capterra tracking, with strong positive φ and high lift (about 5–9× for Marketo’s pairings).
3. The developer/PLG kit: PostHog and Vercel Analytics. Across the whole sample these are negatively associated with the enterprise kit: PostHog vs Marketo φ = −0.16, PostHog vs 6sense φ = −0.14, and PostHog and Floodlight never appear together. But within size bands none of these negative associations is significant. The split mostly reflects that PostHog is common at startups and the enterprise tools at large companies. HubSpot vs Marketo is different: it stays negative within size bands (φ = −0.15; only 2 sites have both). Sites rarely run two marketing-automation platforms.

P2 is partly supported. Co-occurrence is far from random, and the positive clusters hold within size bands. The startup and enterprise kits do differ, but that difference is largely a company-size effect, not two stacks that exclude each other.

Why it matters. B2B tracking stacks look less like individual choices and more like kits. Sites with Marketo are much more likely to also have ABM tools, Bing and review-site pixels, even compared with companies of the same size. For vendors, the kit a prospect already runs is a strong signal of what else is likely to be in its stack.

Caveat. Co-occurrence is association, not sequence. We can’t tell from one snapshot which tool came first. Pooled associations can also reflect company size and the collection method. GTM pairings are especially affected, because parsing a GTM container is how we discover many of the other tools.

Finding 5: The attribution blind spot

84% of sites with ad pixels had no detectable dedicated attribution tool

What the data shows. We measure the gap between detected ad pixels and detected attribution tooling with the Blind Spot Rate (see Definitions), reported in a strict and a broad version.

	Sites with ad pixels	Strict Blind Spot	Broad Blind Spot
All	552	84.4% (81.2–87.2%)	41.7% (37.6–45.8%)
1–10 employees	100	98.0%	78.0%
11–50	206	91.3%	43.7%
51–200	99	81.8%	40.4%
201+	147	67.3%	15.0%

The lower end of the strict confidence interval (81.2%) sits far above 50%. P3 is supported for detectable tools. Dedicated attribution among sites with ad pixels varies significantly with size (χ² = 54.4, df = 3, p < 10⁻¹¹, V = 0.31, a moderate effect). Even so, two-thirds of 201+ employee sites with ad pixels had no detectable dedicated attribution tool. The confidence interval reflects sampling error only; it doesn’t correct for the sample’s selection or for tags we couldn’t detect.

More channels, only slightly more attribution. Dedicated attribution is more common at each ad-pixel count, from 2% of single-pixel sites to 37% of sites with 5+ pixels. Still, 75.2% of sites with 3+ ad pixels had no dedicated attribution tool detected. A logistic regression on the 552 sites with ad pixels (McFadden pseudo-R² = 0.15) finds:

• each additional ad pixel is associated with 35% higher odds of a dedicated attribution tool (OR 1.35, p < 0.001), holding size constant;
• relative to 1–10 person companies, the odds are 7.5× higher at 51–200 employees (p = 0.009) and 9.2× higher at 201+ (p = 0.004). The 11–50 band’s difference (OR 4.0) is not statistically significant (p = 0.07).

P4 is supported, but the effect is modest. Even among sites with many ad pixels, most had no dedicated attribution tool detected. This is an association across companies, not evidence of when companies add channels or measurement.

By category, the Broad Blind Spot Rate ranges from 28% (HR & Recruiting) to 56% (Engineering & Product). Even among marketing-software companies, which sell to marketers, 35% of sites with ad pixels showed neither a dedicated attribution tool nor HubSpot, Marketo or Pardot.

Why it matters. Most sites with LinkedIn, Google, Meta or Reddit pixels show no dedicated layer for tying those touchpoints to pipeline. Unless they attribute through GA4, a CRM, or server-side tools we can’t see, they rely on each platform’s own reporting, where every platform can claim credit for the same deal. The gap is widest among the smallest companies (1–10 employees: 98% Strict, 78% Broad).

Caveat. The Strict and Broad rates are two classification rules, not bounds on real measurement capability. GA4’s attribution reporting, Salesforce campaign influence, server-side or warehouse-based attribution, and in-house reporting aren’t captured, and an installed tool may go unused. “No dedicated tool detected” is not the same as “no attribution”. Counting Usermaven as attribution, as our classification rule requires, lowers the Strict rate by 0.4 points. Removing a false Rockerbox detection raises it by 0.2.

Finding 6: Overlapping tools, legacy tags

Nearly four in ten sites with analytics have multiple analytics tools detected, and nearly a third of sites using GTM still have Universal Analytics in their container

What the data shows. The Tracking Overlap Rate is 38.5% (327 of 850 sites with analytics; 95% CI 35.3–41.8%), and 6.9% have three or more analytics tools detected. 55.2% of sites with product analytics also have GA4, and 15.3% of sites with session replay have two replay tools detected (e.g. Hotjar and Clarity). Overlap is fairly flat across size bands (35–43%): multiple analytics tools aren’t only a big-company pattern.

Legacy tags. Standard Universal Analytics properties stopped processing data on 1 July 2023, and Google shut down UA, including UA 360, from 1 July 2024. Yet UA configuration was found inside the GTM container on 167 of the 534 sites using GTM (31.3%) [5]. Counting every source, UA was detected on 190 sites (17.1% of all sites), including 179 GTM sites (33.5%).

Why it matters. Each additional tag adds page weight, another vendor in consent and privacy reviews, and another set of numbers that may not match the others. Multiple tools can be deliberate, so overlap is a reason to review, not proof of waste or double spending. Legacy UA configurations suggest that some GTM containers may contain outdated tracking setups that warrant review.

Caveat. Some overlap is deliberate (for example, a product analytics tool for in-app behaviour alongside GA4 for acquisition), and Cloudflare Web Analytics can be switched on by default at the CDN level, which may inflate overlap. We didn’t measure software costs. Some UA configuration may be paused or, where set up as a connected site tag, still pass data to GA4.

Implications for practice

1. Check for a blind spot before adding the next channel. If you run two or more ad pixels, make sure you have a way, a dedicated tool or otherwise, to tie paid touchpoints to signups, pipeline and revenue before you launch Reddit or Bing. (Finding 5)
2. Audit your GTM container, not just your HTML. In our sample, most ad tags lived there. Remove legacy Universal Analytics tags and unused or paused tags, and document who owns each one. (Findings 3 and 6)
3. Review overlapping analytics. If GA4 and a product analytics tool both count your visitors, check whether each has a distinct job. If not, choose one source of truth for traffic and conversions, or consider a tool that covers web analytics, product analytics and attribution in one place. (Finding 6)
4. Benchmark against your kit, not the market. A PLG developer tool and an enterprise ABM company shouldn’t compare themselves to the same “average B2B stack”. (Findings 2 and 4)
5. Treat every tag as a privacy decision. Each additional pixel is another vendor receiving visitor data, which matters for GDPR and consent coverage. (Findings 1 and 6)

Methodology

Sampling frame. 1,166 B2B software companies:

• 1,010 from the Y Combinator public company directory [9] (industry = B2B, status Active or Public, with a reported team size and a B2B sub-industry). Stratified by team size: random samples of 400 from 1–10 and 400 from 11–50 (seed 42), plus all 146 companies at 51–200 and all 64 at 201+.
• 156 public and late-stage private B2B SaaS companies (e.g. Salesforce, HubSpot, Datadog, Snowflake, Gong, Rippling), compiled by the research team and assigned to the 201+ band.

Categories follow YC sub-industries, mapped to 13 groups; the 156 large companies were categorised manually. 1,110 sites (95.2%) returned a usable homepage.

The other 56 failed through timeouts, HTTP errors or bot protection: 17 of 400 at 1–10 employees (4.3%), 13 of 400 at 11–50 (3.3%), 5 of 146 at 51–200 (3.4%) and 21 of 220 at 201+ (9.5%; 5 YC and 16 selected large companies). Failures were concentrated among the largest companies, mostly through bot protection, so the 201+ band (n = 199) may slightly under-represent sites with the strictest security set-ups.

Pages audited. For each site, the homepage and the pricing page, found from a homepage link containing “pricing” or by requesting /pricing. A pricing page was found on 64.5% of sites.

Detection. 68 tools were identified by script signatures (script URLs, initialisation calls, pixel endpoints) in four places:

1. raw HTML of both pages (script tags, inline scripts, <noscript> pixels, preconnect hints);
2. the public Google Tag Manager container for every GTM ID found (515 containers). We parsed only the container’s tag configuration (built-in tag types and custom HTML) and excluded GTM’s own library code, after an earlier pass showed that library code inflated Floodlight detections;
3. Segment destination settings for every Segment write key found;
4. up to 10 first-party JavaScript bundles per homepage, to catch SDKs installed via npm (e.g. posthog-js), using high-specificity endpoint signatures only.

Validation. On 19 October 2026 we re-fetched every site with a detection from the tools whose signatures include brand names or patterns we judged prone to false matches (all dedicated attribution tools, Usermaven, 6sense, Snowplow, Leadfeeder, Pardot, Adobe Analytics, Lucky Orange, RB2B, Floodlight and Matomo). We recorded each matched snippet with its context and checked them by hand. Of 361 site–tool detections, 339 were confirmed as tags or tag configuration. 20 were removed: logos, mentions or list entries (seven 6sense, two Snowplow, one HockeyStack, one Rockerbox, one RB2B), Next.js or other files that merely ended in “sp.js” (six Snowplow, one of them no longer reproducible), and two Adobe Target preconnects counted as Adobe Analytics. Two detections whose original GTM match could not be reproduced, but whose pages still reference the tool, were kept (one Bizible, one Factors.ai). All figures in this report reflect the corrected data. Other tools are detected by URL, endpoint or initialisation-call signatures.

Statistics.

• 95% confidence intervals for proportions: Wilson score method.
• Size effects: chi-square tests of independence (df = 3) with Cramér’s V, and Spearman’s ρ.
• Differences in tool counts: Kruskal–Wallis.
• Pairwise co-occurrence: phi coefficient and lift; 2×2 chi-square test (no continuity correction), or Fisher’s exact test where any expected cell count was below five (102 of 378 pairs) [6]; Benjamini–Hochberg false-discovery-rate correction across all 378 pairs and across per-tool size tests. Using chi-square for every pair would give 246 significant pairs instead of 241.
• Size sensitivity: Cochran–Mantel–Haenszel test of each pair stratified by size band, with Benjamini–Hochberg correction.
• Attribution adoption: logistic regression on ad-pixel count and size band.
• Analysis in Python (pandas, SciPy, statsmodels).

Limitations and threats to validity

Measurement validity.

• Detection is mostly a lower bound. Tags that load only after consent, only on deeper pages, through server-side tagging, or from bundles we didn’t scan will be missed. A site showing “no tags” may still track. In the other direction, signature matching can produce false positives. Our hand validation removed 20, and tools with URL- or endpoint-specific signatures weren’t hand-checked.
• Presence is not usage. A tag in a GTM container may be paused or misconfigured; a bundled SDK may not be initialised. An ad pixel doesn’t confirm current campaigns or spend. We did not validate detections against a rendered-browser sample.
• Classification choices. Some tools have attribution features we didn’t classify as attribution (e.g. GA4’s attribution reports [7]). Classifying multi-layer platforms takes judgement; our rule and its effect on the headline are stated in Definitions and Finding 5. Reporting both Strict and Broad Blind Spot Rates shows how much the result depends on that choice, but neither rate is a bound on true measurement capability.

External validity.

• The small-company sample is weighted toward venture-backed startups (YC). Results describe that population and shouldn’t be read as the whole B2B software market.
• Category effects are partly confounded with size, and so are pooled tool associations. The size-stratified test addresses the second.
• Confidence intervals and p-values reflect sampling error only. They don’t correct for selection bias in the sample or for undetected tags.

Design.

• Cross-sectional data. We observe each site once, so we report differences between groups and associations between tools, not adoption sequences, causes or growth trajectories.
• Exploratory analysis. The working propositions were not pre-registered, and many tests were run. We applied Benjamini–Hochberg correction to the co-occurrence and per-tool size tests, but other results should be read as descriptive.

Competing interests. This study was produced by Usermaven, which sells web analytics, product analytics and attribution software [8]. Usermaven is among the 68 tools in the signature library and was detected on 2 sites. It is classified by the same rule as every other tool.

Temporal validity.

• One point in time. Fieldwork took place on 15 October 2026, with validation re-fetches on 19 October 2026. Stacks change, and a repeat audit would be needed to measure change over time.

Appendix A: Summary of working propositions

Proposition	Test	Result	Verdict
P1. Larger companies have larger stacks	Spearman ρ; Kruskal–Wallis	ρ = 0.47, H = 250.2, p < 10⁻⁵³	Supported
P2. Tools co-occur in distinct kits	Pairwise χ² or Fisher, φ, BH correction; size-stratified CMH	241 of 378 pairs significant; 192 also within size bands; startup–enterprise split largely explained by size	Partly supported
P3. Most sites with ad pixels lack a detectable dedicated attribution tool	Wilson 95% CI	84.4% (81.2–87.2%)	Supported (detectable tools only)
P4. More ad pixels, more likely to have attribution	Logistic regression	OR 1.35 per pixel, p < 0.001	Supported, modest effect

Appendix B: Tool classification

Category	Tools
Tag manager / CDP	Google Tag Manager, Segment, RudderStack, Tealium, Adobe Launch
Analytics (web and product)	Google Analytics 4, Universal Analytics (legacy), Mixpanel, Amplitude, Heap, PostHog, Plausible, Fathom, Matomo, Usermaven, Pendo, Adobe Analytics, Vercel Analytics, Cloudflare Web Analytics, Simple Analytics, Snowplow, Koala
Session replay	Hotjar, Microsoft Clarity, FullStory, LogRocket, Mouseflow, Crazy Egg, Smartlook, Lucky Orange
Marketing automation / CRM	HubSpot, Marketo, Salesforce Pardot / Account Engagement, Intercom, Drift, Customer.io
Visitor ID / ABM	Clearbit, 6sense, Demandbase, RB2B, Warmly, Leadfeeder / Dealfront, Apollo Website Visitors, ZoomInfo WebSights
Ad pixel	Meta Pixel, LinkedIn Insight Tag, Google Ads, Google Floodlight / DV360, X (Twitter) Pixel, Reddit Pixel, Microsoft Ads (Bing UET), TikTok Pixel, Quora Pixel, AdRoll, Capterra / G2 tracking
Dedicated attribution	Usermaven (also counted under analytics), Dreamdata, HockeyStack, Cometly, Triple Whale, Fibbler, Factors.ai, Bizible (Adobe Marketo Measure), Hyros, Ruler Analytics, Attributer, Northbeam, Rockerbox, Wicked Reports

“Product analytics” in Table 1 means PostHog, Amplitude, Mixpanel, Heap or Pendo. “Privacy-first analytics” means Plausible, Fathom, Matomo, Simple Analytics, Cloudflare Web Analytics or Usermaven.

References

1. HTTP Archive (2024). Web Almanac 2024: Third Parties. https://almanac.httparchive.org/en/2024/third-parties
2. Google. Google Analytics 4 has replaced Universal Analytics. Analytics Help. https://support.google.com/analytics/answer/11583528
3. SciPy. scipy.stats.chi2_contingency (documentation on expected frequencies). https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.chi2_contingency.html
4. Google. Attribution reporting (GA4). Analytics Help. https://support.google.com/analytics/answer/10596866
5. Usermaven. Marketing attribution software.
   https://usermaven.com/multi-touch-attribution
6. Y Combinator company directory, via yc-oss API. https://yc-oss.github.io/api/companies/all.json