91% of marketers use AI. Only 41% can prove ROI — down from 49% the prior year. More adoption, less accountability. The average marketing team now pays for 12 to 15 overlapping AI tools. Here's the audit framework that tells you what to keep, what to cut, and what to consolidate.

Why AI adoption and AI ROI are moving in opposite directions — and what the gap reveals about stack design
The three-list framework that every stack audit should produce: keep, cut, consolidate
The 90-day cadence that prevents the audit from becoming an annual event nobody acts on

The Stack Grew Faster Than the Governance

(cite index="33-1">91% of marketers now use AI, yet only 41% can prove ROI — down from 49% the prior year. The martech landscape hit 15,384 tools in 2025, a 9% increase driven by AI proliferation. 76% of enterprises report negative outcomes from disconnected AI tools.</cite)

The inversion is the story. Adoption went up. Proven ROI went down. That's not a coincidence — it's a consequence of how AI tools get added to stacks. A content team keeps the legacy brief tool, adds an AI draft tool, adds an optimization assistant, and adds a separate QA workflow. Each addition seemed reasonable at the time. The aggregate is a stack of overlapping tools that individually underperform and collectively create more complexity than they solve.

(cite index="31-1">The 2026 twist is AI. Agencies are adding AI layers faster than they are retiring older systems. Instead of using AI to simplify the stack, they use AI to add another stack on top of the old one.</cite)

The signal that a stack needs auditing isn't one tool failing — it's the maintenance cost of the whole exceeding the value of any part. When the team spends more time switching between tools than using them, when outputs from one tool don't feed the next, when nobody can name who owns which tool's performance, the stack has outgrown its governance.

The Three Lists Every Audit Should Produce

(cite index="31-1">Every agency should walk out of a 2026 stack audit with three lists. The first is what to keep: systems with clear usage, clear ownership, reliable integrations, and measurable value. The second is what to cut: tools with weak adoption, overlapping functionality, poor ROI, or high maintenance relative to value. The third is what to consolidate: categories where two to five tools can be reduced into one primary system without reducing delivery quality.</cite)

Keep when: the tool has a named owner, documented active usage across more than 70% of eligible team members in the past 90 days, a clear integration with at least one other tool in the stack, and at least one measurable output metric tied to it. The usage threshold matters — most teams drastically overestimate how widely their tools are actually used. A tool that "everyone uses" often has five active users in a team of twenty.

Cut when: adoption is below 40% in the past 90 days, functionality overlaps significantly with another tool in the stack, there is no named owner and no one who would notice if it were cancelled, or the maintenance and administration overhead exceeds the time the tool is saving. (cite index="28-1">Put every renewal through the same ROI test you would use to approve a new purchase. If the primary reason a tool is in the stack is that nobody got around to cancelling it, that is the signal.</cite)

Consolidate when: two or more tools serve the same function for different teams or workflows, the integration cost of connecting them would be lower than the ongoing cost of running both, or a newer platform in the stack can absorb the function of an older one without reducing output quality. (cite index="31-1">The fastest savings usually come from AI tool overlap, reporting duplication, collaboration add-ons, light automation tools that duplicate CRM or project features, and legacy point solutions retained after a broader platform expanded.</cite)

The Audit Process: Four Phases

Phase 1: Inventory. List every tool the team has access to, regardless of whether it's in active use. Include free tools, trials that became subscriptions, tools added by individuals that never got centrally documented, and any AI capabilities bundled into platforms the team already uses but hasn't activated. Most teams discover 20 to 40% more tools than they thought they had. The inventory reveals the first category of cuts immediately: tools that exist in the list but that nobody can describe a use for.

Phase 2: Usage audit. For each tool in the inventory, collect 90 days of usage data. Most SaaS platforms expose this in admin dashboards. For tools that don't track usage natively, run a team survey with a single question: "In the last 90 days, how often did you use [tool] to complete a work task?" Four options: never, once or twice, weekly, daily. Tools in the "never" or "once or twice" category go to the cut list unless there's a specific, documented reason for low usage (a tool used only in specific campaign types, for example).

Phase 3: ROI mapping. For each tool that passes the usage threshold, define what output it produces and what that output is worth. (cite index="32-1">For marketing, tie tools to cost per qualified lead, speed to launch, campaign throughput, content production cycle time, funnel conversion, and sourced or influenced pipeline.</cite) A tool that reduces content production cycle time by two days per campaign is worth a different amount to a team that runs five campaigns per month than to one that runs fifty. The ROI calculation doesn't need to be precise — it needs to be directional enough to rank tools against each other and against their annual cost.

Phase 4: Integration assessment. For each tool on the keep list, assess its integration quality with the rest of the stack. (cite index="30-1">A content AI tool that doesn't share data with your SEO tool creates duplicate work. An email AI tool that doesn't connect to your CRM delivers generic personalisation. An analytics tool that doesn't integrate with your ad platforms produces incomplete attribution.</cite) Tools that fail the integration assessment either need to be integrated (and the integration cost added to their total cost of ownership) or moved to the consolidate or cut list.

The Specific Categories That Produce the Most Cuts

Two AI tool categories consistently underperform in 2026 ROI data. (cite index="35-1">AI video tools delivering just 1.1× to 1.6× ROI because production overhead remains high, and AI-generated paid social creative, which Meta, TikTok, and Google have quietly down-ranked in their 2026 algorithm updates.</cite)

The category that produces the most consolidation candidates is content production tooling. Teams that added a dedicated AI writing tool, then an AI SEO optimizer, then an AI content repurposing tool typically find that one of their existing platforms — often their CMS or their DAM — has expanded to cover one or two of these functions natively. The standalone tools stay on the subscription because the integration work required to switch to the native function seems like effort. The audit forces the question: is the standalone tool delivering enough value over the native function to justify both the subscription and the integration complexity?

The category that produces the most surprises in the usage audit is analytics and reporting. Teams often subscribe to AI analytics tools that were purchased to solve a specific problem — attribution, competitive intelligence, performance prediction — that has since been solved by a native feature in their ad platform or CRM. These tools sit at low adoption not because the team decided they weren't useful, but because the problem they solved was quietly solved elsewhere.

The 90-Day Review Cadence

(cite index="34-1">A strategy without a review cadence quietly falls apart. Set a recurring 90-day check-in where you revisit your tool audit, compare against your baseline, and cut anything that isn't earning its place.</cite)

The audit is not a one-time event. It is a quarterly operating discipline. The 90-day cadence serves three functions that an annual review can't: it catches new tool additions before they become entrenched, it surfaces usage drift before contracts renew, and it maintains the stack map as a current document rather than an archaeological artifact that nobody trusts.

The 90-day review is lighter than the initial audit: 30 minutes per quarter reviewing usage data for the keep list, flagging any tool whose usage has dropped below the threshold since the last review, and reviewing any new tool requests against the stack map to determine whether an existing tool can serve the need before a new subscription is added.

When the production infrastructure that manages creative workflows keeps tool usage data alongside project activity, the 90-day review generates automatically from the production record. The audit overhead drops from a multi-day project to a standing agenda item.

FAQ

How do you handle tools that are used by only one team member who is highly dependent on them? Separate the individual dependency from the organizational value. If one person uses a tool daily and considers it essential, first determine whether the function is available elsewhere in the stack. If yes, the tool is a consolidation candidate. If no, consider whether the function is strategic enough to justify the cost and the single-point-of-failure risk. Tools with one active user and no backup owner are a governance liability regardless of that user's engagement.

What's the right way to communicate tool cuts to the team? Announce the cut with the reason and the alternative in the same message. "We're cancelling [tool] because usage was below 20% in the past 90 days and the function is available in [existing tool]. Here's how to do [the specific workflow] in [existing tool] instead." The most common objection to stack cuts is that the replacement isn't as good — address this proactively by running the replacement workflow with anyone who used the cancelled tool before the cut takes effect.

How do you audit AI capabilities that are bundled into existing platforms rather than standalone tools? Include them in the inventory as capabilities rather than tools. The question is the same: is this capability in active use, is there a named owner for it, and is it producing measurable value? Bundled AI capabilities that nobody has activated are the most common source of hidden AI value in a marketing stack — and the most common reason a new standalone AI tool purchase is unnecessary.

What's the minimum viable stack for a creative team of 10? (cite index="35-1">Most high-performing teams run 3 to 5 integrated tools, not one platform.</cite) For a creative team of 10: a project management and workflow tool, an asset management and distribution system, a content production tool with AI assistance, and an analytics platform. Everything else should be justified against these four rather than added by default.

How long should the initial audit take? The inventory and usage audit take one to two weeks for a team of 10 to 20 with a mid-complexity stack. The ROI mapping and integration assessment take another one to two weeks. Four weeks total for the initial audit, then 30 minutes per quarter to maintain it. Teams that skip the initial audit because it "takes too long" typically spend more than four weeks per year managing the overhead of an ungoverned stack.

How to Audit Your AI Stack: Cutting the Tools That Don't Deliver

The Stack Grew Faster Than the Governance

The Three Lists Every Audit Should Produce

The Audit Process: Four Phases

The Specific Categories That Produce the Most Cuts

The 90-Day Review Cadence

FAQ

Sources

Other Posts

The Agency Reset: Why the Labour-Based Creative Model Is Breaking

The Creative Project Risk Register: How to Anticipate What Can Go Wrong

How to Build a Template Library That Teams Actually Use