You can only buy what you’re able to specify. Martech Futurist | August 26, 2026
Your leverage in the next martech renewal equals the precision of the specification you bring to it.
I argued on Monday that marketing leaders now hold contractual leverage, because other organizations write the underlying conditions your stack depends on. That argument has a precondition I skipped. Contractual leverage requires you to name what you want. Most marketing organizations cannot name it, because they have never measured what their current systems actually produce or what it costs to fix the output.
This week's research puts numbers on that gap.
Price the correction work before you price the license
TechTarget published a feature on August 19 covering what Glean's Work AI Index calls botsitting: the untracked labor of checking AI outputs, feeding models missing organizational context, debugging errors, and rerunning prompts. Workers in the study save roughly 11 hours a week using AI and spend 6.4 hours a week making the output usable, leaving a net gain of 4.6 hours. Frequent botsitters are 73% more likely to be looking for new jobs.
Two findings matter more than the productivity math. Forty-one percent of workers have delivered AI-generated outputs they could not explain if asked, and 12% knowingly delivered output they believed was wrong. Those figures describe an organization operating without an acceptance standard. Nobody wrote down what "good enough to ship" means for a given workflow, so each person set the bar privately, under deadline pressure.
Christian Chung of Fueled, quoted in the piece, put the operational risk plainly: adoption metrics look healthy while a meaningful share of the paid-for return leaks back out as cleanup time that appears in no report.
It reminds me of what Srikrishnan Ganesan, CEO and Co-Founder at Rocketlane, said when I interviewed him on The Agile Brand podcast: "The human brings the deep context and verifies what the AI delivers. Then they iterate, teach the system, and automate more of the work over time." Verification is real work with a real cost. Cost it, staff it, and put its acceptable level in writing.
This sits squarely in the Trust and Verification layer of my AI capability framework. Trust under Augmentation means calibration: knowing when a human should override the contribution. You cannot calibrate against a standard you never wrote.
Keep notes on what you decided to ignore
Forrester published a blog post this week on AI use case sprawl in B2B digital commerce. Its analysts describe an expanding catalog of options open to any team: intelligent search, personalization, product recommendations, digital assistants, content generation, pricing optimization, agentic experiences. Finding opportunities has become trivial. Choosing among them has become the difficult part, and Forrester's analysts argue that the hardest decision facing these organizations is which use cases to leave alone.
Growing portfolios of pilots, competing priorities, and fragmented ownership follow directly from an unwritten selection standard. When no one records why the team passed on a use case, the same idea returns at the next executive meeting with a new vendor attached.
It reminds me of what Meenal Nalwaya, Head of Product at Reka, said when I interviewed her on The Agile Brand podcast: "This tech is cool, but still anchor on the customer problem; your product will really be defined by the success metric." The success metric is the specification. Write it before the pilot, and the pilot tells you something.
The Scrum practice that translates best here is the Definition of Done: an explicit, agreed statement of the conditions an increment must satisfy before anyone calls it complete. Apply it per AI workflow. Name the accuracy bar, the review step, the escalation path, and the person accountable for the call.
Move your buying criteria from features to controls
MarketScale published an analysis on August 23 of Gartner's Hype Cycle for Digital Marketing, 2026. Gartner's abstract describes CMOs facing a trilemma of flat budgets, aggressive growth targets, and disruption from answer engines, and positions autonomous marketing as the operating model taking shape in response. The framing pushes evaluation criteria from features toward controls: metering models for AI features, audit trails for content decisions, notification terms when a supplier swaps underlying models, and named human gates for claims, pricing, and regulated language.
Every one of those is a specification you write yourself and then hold a supplier to.
Here is what the arithmetic looks like on a single workflow. Say your team produces 40,000 AI-assisted content items a year across email, product detail copy, and ad variants. The platform quotes $180,000 annually. Twelve people touch that workflow, and applying the Glean ratio, each spends 6.4 hours a week on correction across 48 working weeks: 3,686 hours. At a fully loaded $75 an hour, correction costs $276,000. Total annual cost lands near $456,000, and the license accounts for 39% of it.
Now specify. You set a first-pass acceptance rate of 85% for product detail copy, measure the current rate at 60%, and write both figures into the Statement of Work with a quarterly review. Closing a third of that correction gap returns about 1,229 hours, worth roughly $92,000 a year. That single number changes what you are willing to pay and what you are willing to accept.
You reach that number by measuring first. The vendor cannot hand it to you.
Staff the people who write the spec before you cut them
Jodie Cook reported in Forbes on August 20 that roughly 10,000 marketing roles have disappeared, with AI agents absorbing the work. WPP moved from 108,044 people to 98,655 in a year. McKinsey cut 3,000 to 4,000 positions while its workforce came to include 20,000 AI agents. Forrester revised its forecast from 7.5% of US agency jobs automated by 2030 to 15% automated by the end of 2026. Gartner's CMO Spend Survey shows where the reductions landed: 23% of agencies cut junior copywriting roles in 2025 with 31% planning further cuts, and 19% cut junior design roles with 24% planning more.
The junior layer performed most of the verification. Those people read the draft, caught the wrong product name, flagged the claim legal would reject, and learned the brand standard by applying it a thousand times. Cutting that layer while output volume rises moves the correction burden onto senior staff who cost more per hour and have less time.
Book 2 holds a commitment that applies directly. Humans stay accountable for direction even when analysis runs fully automated. Intelligence tells you what is and what is likely. Deciding what the organization should accept is a normative act, and it belongs to a person with a name and a budget. Agile Brand Principle 4 says the same thing from the values side: respect employees and their time. Unpriced correction labor disrespects both.
Reorganize before you cut. Move two people from execution into a specification and quality function, give them the acceptance standards to write and enforce, and measure them on first-pass acceptance rates by workflow.
Key insights
Correction labor is a line item. Multiply people by hours by loaded rate and put the result next to the license fee. The ratio will change your renewal position.
An acceptance standard is a per-workflow artifact. Name the accuracy bar, the review step, the escalation path, and the accountable person for each workflow separately.
Controls belong in the SOW. Metering model, audit trail access, model-change notification terms, and human-gated decision categories go in writing before signature.
Selection decisions need a written reason. Record why you passed on a use case, and the same pilot stops reappearing under a new vendor name.
Verification needs a home. Assign the people who enforce the standard before you reduce the layer that was quietly doing it.
What to do this week
Pick your highest-volume AI workflow. Measure its first-pass acceptance rate across 50 recent outputs. Write that number down along with the rate you require. Bring both to your next vendor conversation.
That is the whole discipline. Your leverage in the renewal equals the precision of the specification you bring to it, and precision starts with a measurement you took yourself.
A closing note
I have written more than a few books, and (counting curated and co-written books) will have published four this year alone. The ones that went well started with an outline specific enough that I knew what each chapter had to accomplish before I wrote a sentence of it. The ones that went badly started with a topic and a hope, and took much longer to shape and ultimately craft into a cohesive whole. The difference showed up months later, in rewrites, and the rewrite cost always exceeded what the outline would have cost.
Marketing organizations are running the second version right now at considerable scale. The specification work looks slow next to a pilot you could launch tomorrow. It is the only thing that makes the pilot mean anything.