More Selected AI Tool Categories Did Not Mean Clearer Attribution

Clear attribution was roughly unchanged between respondents selecting three to four AI tool categories and those selecting five or more. The five-plus group also reported weaker cost visibility, an association that does not show what adding or removing a tool would cause.

These figures come from the State of AI-Led Growth 2026 survey. The full report breaks down tool-category use, attribution, and spend visibility across respondents.

What the survey curve shows

The survey counted selected tool categories, not necessarily individual tools. Clear attribution was reported by 17 of 63 respondents selecting one or two categories, 33 of 88 selecting three or four, and 15 of 41 selecting five or more—27.0%, 37.5%, and 36.6%. The difference between the last two groups is less than one percentage point and does not establish a sweet spot or a decline after category four.

Respondents selecting five or more categories were no more likely to report clear attribution than those selecting three or four. The survey is cross-sectional, so it does not show what adding a category would do to the same team or whether any category was purchased to improve attribution.

Attribution is a system problem

Each added tool can create another data surface. Attribution becomes harder when enrichment, automation, analytics, and CRM systems disagree, and models break down when disconnected inputs accumulate. That is an operational explanation, not a mechanism tested by this survey.

A marketing ops lead at a Series D company put the failure mode to us plainly. Their CRM, automation, and analytics "all tell slightly different stories. AI helps summarize, but it doesn't reliably reconcile truth across systems yet."

A Salesforce survey of roughly 7,700 data and analytics leaders found that 84% said their data strategies needed an overhaul. That supports the broader data-readiness problem, but it does not show that adding a fifth tool causes attribution to worsen.

The invisible AI budget

Spend visibility was also limited. In our survey, 54% reported only a rough monthly AI-spend figure or no tracking at all: 40% rough and 14% none.

Only 51% of organizations can confidently evaluate AI ROI, and 57% still track that spend in spreadsheets. Spreadsheets haven't gone away. They just sit alongside native billing tools and third-party platforms instead of replacing them.

More selected categories were associated with weaker visibility

In our data, 61% of respondents selecting five or more categories lacked precise cost visibility, compared with 49% of those selecting three or four. These are different cohorts; the survey does not show that eliminating tools would reduce the rate or that trimming helps immediately.

AI products may use fixed seats, variable consumption, or hybrid pricing. Credits, tokens, API calls, and agent actions add meters that can make total spend harder to track when ownership is unclear.

The strongest vendor benchmark tells the same story. A survey of 218 IT leaders found 78% hit unexpected consumption or AI-pricing charges in the past year. The same index reported 393% growth in AI-native spend at organizations above 10,000 employees; it does not establish that larger stacks caused the growth.

The benchmark comes from a vendor selling into this problem, so its incentive matters. Our 61%-versus-49% result is an association in this survey, not evidence that category count caused weaker visibility.

The honest counterpoint on utilization

The fair objection to all of this is that utilization data actually looks better lately. Average martech stack utilization was 33% in 2023. By 2025 it had risen to 49% in Gartner's series, with only 15% of organizations qualifying as high performers. Those are separate survey editions with different respondent pools, and both characterize utilization as low, but the direction is up, not down, and we're not going to pretend otherwise.

Even so, at 49% more than half of stack capability still sits unused, and AI spend adds a new blind spot on top of the old one. Business AI adoption crossed 50% among businesses on Ramp in March 2026; within that customer base, monthly AI spend rose 4x year over year, and the median business already spending on AI directed nearly 15% of its software budget to it. These figures describe Ramp customers, not all companies.

Audit the stack before another purchase

Before another purchase, audit the four things the tools you already run should be doing.

Start with discovery. Inventory every AI tool in use, including free tiers, browser extensions, personal API keys, and AI add-ons metered onto platforms you already license. The inventory may surface overlap or unused access.

Then spend attribution. Map every meter, from seats to API calls, to a team and a pipeline stage, so an overage lands on an owner instead of arriving as a surprise invoice.

Then policy. Decide which tools may write events to the CRM and which stay read-only. A tool that can't write a governed event into the CRM can't participate in attribution, whatever its dashboard claims.

Then monitoring. Review the inventory and the spend map monthly, because consumption pricing moves costs without any procurement event to warn you.

The goal is a governed set of tools with clear CRM connections, a visible path from first touch to pipeline event, and one shared rule for a qualified touch. Protect first-party, CRM-anchored events because the CRM is where revenue is recorded. This is GrowthX's recommended architecture; the survey did not establish that it separated the category-count cohorts.

The full report has the attribution-by-tool-count breakdown and spend-visibility benchmarks behind this article. Download it, then use the four-step audit above before your next renewal conversation.