Marketing Bought “AI” To Optimize What Was Already Fast. Martech Futurist | July 30, 2026
The last five days of enterprise marketing research and product releases converged on a single measurement problem. Marketing organizations report high AI deployment rates and unchanged launch dates. The reason sits in allocation. Teams applied AI capacity to drafting, ideation, and image generation, which were the fastest steps in the production chain before AI arrived. The steps that consume the calendar are assembly, review, and coordination across teams, and those received the least automation. A deployment percentage measured at the drafting layer describes activity. It says nothing about whether anything shipped sooner.
The July 28 edition of this newsletter argued that continuous agent operation creates a variable cost line marketing operations now has to carry on the P&L. This edition takes the next step. The return on that line depends on where the capacity was pointed, and the allocation data published this week shows it was pointed away from the constraint.
Adoption is being measured at a step that was never the bottleneck
Knak surveyed more than 300 enterprise marketing leaders and found 70% running AI in production, with 88% reporting that output still requires moderate to substantial human editing before use. In the same population, 85% missed at least one planned campaign launch date in the past twelve months, and one in ten missed more than five times a year. The causes respondents named are operational: approvals and sign-off at 47%, design and creative production at 38%, and cross-team coordination at 36%.
The allocation figures explain the gap. AI use concentrates on first-draft copy at 64% and image generation at 56%. Only 25% of teams apply it to building or coding the emails and landing pages where the production burden is heaviest. Sixty percent of surveyed companies involve four or more people in producing a single email, 69% require two to three revision rounds, 54% move the work across three to five separate tools, and 51% still run approvals through email, Slack, or Teams threads. Knak prices the internal labor at more than $300 per send using Bureau of Labor Statistics pay data.
It reminds me of what Hannah Elsakr of Adobe said when I interviewed her on The Agile Brand podcast: "[We] found that two-thirds of their time is pixel-pushing to turn something from a 1×1 to a 16×9 to a tower form. That is not inspiring work; let the AI do that." The drudgery she describes sits downstream of the draft. That is the step where the hours accumulate, and it is the step 75% of teams left alone.
In the AI capability framework, this is a pillar mismatch. Drafting assistance is Augmentation, which amplifies a human author. Assembly, routing, and approval are Orchestration, which coordinates action across systems and people. Marketing bought a great deal of Augmentation and needed Orchestration.
Concentration is the signal that capacity found real work
StackAdapt published usage data alongside its Ivy Studio launch that gives marketing operations a diagnostic worth borrowing. Its employees run more than 15,000 AI-assisted workflows per week across campaign management, reporting, troubleshooting, and optimization. Nearly 70% of Ivy Studio usage concentrates in the top 20 workflows.
That distribution shape matters more than the volume. When AI capacity lands on genuine repeated bottlenecks, usage clusters, because a small number of steps carry most of the recurring load. Usage spread thinly and evenly across dozens of workflows indicates that capacity went to work that was already cheap. Any marketing operations leader can run this check against their own telemetry this week: rank workflows by AI invocations, look at what the top 20 absorb, and compare that list against the steps that delay launches.
The check has a second use. It tells you which workflows deserve a cost ceiling and which deserve a redesign.
The controls shipping now put the unit of configuration at the workflow
Algolia added guardrails and cost controls to Agent Studio that are configured per agent. Retailers can set global and per-IP request limits, maximum tokens per response, conversation-depth limits, maximum steps per completion for tool-using agents, and approved domains, alongside input and output content guardrails with fallback messages. IDC's Heather Hershey framed the requirement in economic terms, arguing that agentic commerce cannot scale on intelligence alone and that teams need control over what an agent can say, where it operates, and how much it consumes.
Progress Software moved in a parallel direction on the content side, embedding Sitefinity agents inside the publishing workflow to handle analysis, optimization, and editorial review within the content lifecycle. The release includes conflict handling that prevents contradictory recommendations when multiple agents run at once, which is an operations problem that only appears once several agents share a workflow.
Both releases make the same structural move. They shift the configuration boundary from the tool to the workflow. That boundary is what makes allocation enforceable, because a per-workflow ceiling forces a decision about which workflows justify the spend.
Buyers are pressing on the same point from the demand side. It reminds me of what Bill Staikos, CX operator and consultant at Be Customer Led, said when I interviewed him on The Agile Brand podcast: "2026 is where people are now asking, where's the business outcome and the result? Show me hard metrics now. Last year many software companies got attention just for adding generative AI features; this year that bar is gone."
Featured Insights
PR Newswire (Knak) | Marketing teams adopted AI to move faster. 85% still missed a campaign launch date last year | July 28, 2026
Knak's Marketing Production in the Age of AI surveys more than 300 enterprise marketing leaders and locates the delay in the production layer between an approved idea and a launched campaign. The report separates deployment rate from launch performance and shows the two moving independently, with AI concentrated on creative inputs and largely absent from the build step. Knak counts Google, Amazon, Uber, Meta, and OpenAI among its customers, so the finding reflects sophisticated teams.
My takeaway: Report AI adoption as launches shipped on schedule. A deployment percentage measured at the draft stage will look strong while the calendar stays flat.
ExchangeWire (StackAdapt) | StackAdapt Introduces Ivy Studio, a New Hub for AI-First Advertising | July 28, 2026
StackAdapt launched a workspace combining planning, forecasting, analysis, optimization, and execution, where marketers state an intended outcome and agents surface context and recommend next steps while the marketer holds approval. The disclosed internal usage data is the more useful part for operations leaders: 15,000 AI-assisted workflows per week, with close to 70% of Ivy Studio usage sitting in the top 20 workflows.
My takeaway: Treat the concentration curve of your own AI usage as a diagnostic. Clustering suggests capacity found the bottleneck; a flat distribution suggests it did not.
BigDATAwire (Algolia) | Algolia Adds Governance, Cost Controls to Agent Studio | July 28, 2026
Agent Studio now supports per-agent configuration of request limits, token ceilings, conversation depth, maximum steps per completion, approved domains, and input and output guardrails. The feature set describes a specific production failure: an agent that answers accurately while consuming unbounded inference spend. IDC's Heather Hershey positions these controls as the precondition for moving from pilot to production.
My takeaway: Add per-agent cost and behavior limits to procurement criteria. Ask vendors to demonstrate token ceilings, step limits, and domain restrictions during evaluation.
Progress Software | Progress Software Introduces AI Agents Built into CMS Workflows to Reduce Content Bottlenecks | July 28, 2026
Sitefinity Generative CMS gained custom agents that execute content analysis, optimization, and editorial review inside the publishing workflow, plus page-level evaluation across content, metadata, and SEO properties, adaptive learning from user feedback, and conflict handling across concurrently active agents. Loren Jarrett of Progress described the shift as enterprise AI moving from assisted engagement into digital operations.
My takeaway: When evaluating content tooling, check where the agent runs. An agent inside the workflow removes a handoff; an agent beside it adds one.
Key Takeaways
Separate adoption from throughput in your reporting. Knak's 70% deployment figure and 85% missed-launch figure describe the same teams. Publishing both numbers side by side prevents an adoption rate from standing in for a result.
Audit the production chain before buying more generation capacity. Count the people, tools, and revision rounds required per asset, and price the internal labor. The 25% AI application rate at the build step is where the recoverable hours sit.
Rank your workflows by AI invocation volume. Compare the top 20 against the steps your team names as causes of delay. Divergence between those two lists is an allocation problem you can fix without new spend.
Move the configuration boundary to the workflow. Per-agent token ceilings, step limits, and depth limits let you set different bounds for a high-volume production workflow and a low-volume exploratory one. Tool-level budgets cannot make that distinction.
Name an owner for approval routing. Half of surveyed enterprise teams still run sign-off through chat threads, which is the mechanism behind the 47% approval bottleneck. No agent deployment addresses this until someone owns the routing design.
Expect the proof standard to keep rising. Buyers now open evaluations with outcome questions. Vendors and internal teams both need evidence tied to cycle time and launch performance.
A few more thoughts
Whether it was running two companies (and five buy/sell transactions) as CEO, or working as an advisor to Fortune 500s, the habit that transferred most directly into marketing operations is looking for the step where throughput actually breaks. Reporting almost never points there, because reporting gets built around the steps that are easy to instrument. Drafting is easy to instrument. Approval routing across four people and three tools is not, so it stays invisible in the dashboard and visible only in the calendar.
The seven Agile Brand Principles include continuously improving, which requires knowing when an evolutionary change suffices and when a structural one is needed. This week's data suggests marketing has been running evolutionary changes on the drafting step while the structural problem sits two steps later. The Book 2 framing holds here as well: analysis can tell you where the constraint lives, and choosing to move resources toward it stays a human decision.
The organizations that close this gap over the next two quarters will do it by measuring launches, ranking workflows, and setting bounds per workflow. None of that requires a new model.