The real pattern in all the noise
Scan those headlines and one theme cuts through: everyone is quietly admitting that the way we measure marketing is broken.
Misleading metrics. Cannibalization. AI content. Core updates. New ad surfaces like ChatGPT. Marginal ROI. Loyalty “working harder.” Conversion case studies. Agentic AI shopping. All of it points to the same operational problem:
Most teams are optimizing to the wrong numbers, in channels that are changing faster than their dashboards.
For CMOs, performance marketers, and media buyers, this isn’t an academic issue. It’s why:
- ROAS looks strong while contribution margin quietly erodes
- SEO “wins” don’t show up in revenue because of cannibalization and AI search
- Social engagement climbs while retention and LTV stall
- AI tools generate more content and tests than your measurement system can credibly evaluate
The job now is not to add more metrics. It’s to ruthlessly compress what you measure into a system that can survive:
- AI-driven changes in how people search and shop
- Rising media and logistics costs
- New walled gardens and “brand tax” channels
- Executive pressure for near-term profit, not just growth
The three failures of modern marketing measurement
Most teams are tripping over the same three issues.
1. Metric inflation: dashboards full of numbers that don’t matter
When Neil Patel says “Most marketing metrics are misleading,” he’s being polite. Many are worse than misleading; they’re operationally dangerous.
The usual suspects:
- Blended ROAS that hides channel freeloaders and brand search dependency
- CTR and CPC treated as success metrics instead of cost inputs
- “Engagement” that has no proven link to revenue or retention
- Open rates in an era of privacy changes and unreliable tracking
- Rankings and traffic that ignore cannibalization and intent
The pattern: we track what’s easy, not what’s causal.
2. Channel myopia: optimizing inside silos instead of across the P&L
Look at the headlines on SEO cannibalization, 8,000 title tag rewrites, Google Shopping ROAS, Instagram follower growth, and ChatGPT ads. Each is channel-obsessed.
Inside most orgs, it plays out like this:
- SEO chases non-brand traffic and “visibility” while paid search buys brand terms to hit ROAS targets
- Paid social optimizes for last-click ROAS, ignoring that top-of-funnel creative actually pays off in search and direct
- CRM/email is judged on open/click rates, not incremental revenue or churn reduction
- Brand spends are justified with soft metrics that never reconcile back to unit economics
Result: every channel looks “good” in isolation while CAC rises and marginal ROI falls.
3. AI chaos: more experiments, same brittle measurement
AI is now in everything: content generation, bidding, creative, targeting, even “agentic” shopping and AI search results. But the measurement stack hasn’t caught up.
Current reality:
- AI tools produce content at a pace your attribution model can’t meaningfully evaluate
- Search behavior is shifting to AI overviews and chat interfaces, but reporting still fixates on blue links and classic SERP rankings
- New placements (e.g., ChatGPT ads) are tested with shallow metrics and no clear incrementality framework
- AI “optimization” systems in ad platforms are black boxes, and most teams don’t run proper holdouts to validate uplift
The net effect: teams move faster, but not smarter. More tests, same confusion.
What leaders actually measure instead
The operators who are quietly winning in this environment have a different posture. They don’t chase every metric; they enforce a small, brutal hierarchy of numbers.
At the core, you’ll see three layers:
Layer 1: Non-negotiable business metrics
These are the scoreboard. Everything else is diagnostics.
- Contribution margin per order / per customer (after media, discounts, variable costs)
- Payback period (months to breakeven on CAC, by major cohort or channel)
- Incremental revenue / profit from marketing (via experiments or MMM)
- Retention and LTV by acquisition source and first product purchased
If your dashboards don’t have these front and center, you’re flying by vibes.
Layer 2: Channel-level “truth metrics”
These are the minimum viable set of metrics that actually drive performance in each major channel, tied back to the business layer.
Examples:
-
Search (paid + organic)
- Non-brand vs brand contribution margin
- Query-level profitability (not just keyword-level)
- Cannibalization rate: percent of organic traffic lost or shifted when paid is on
- Share of conversions where search is first or second touch, not just last click
-
Paid social
- Incremental conversions via geo or audience holdouts
- Creative-level MER (marketing efficiency ratio) over a realistic attribution window
- Downstream impact on branded search and direct traffic
-
SEO / content (including AI content)
- Revenue per 1,000 sessions by content cluster or intent
- Net-new query coverage vs cannibalized queries
- Assisted conversion value (via path analysis or MMM)
-
Email / CRM
- Incremental revenue vs holdout (not campaign “attributed” revenue)
- Churn impact by lifecycle program (winback, onboarding, etc.)
- Spam complaint and unsubscribe rates by segment and message type
Layer 3: Diagnostic metrics (on a short leash)
These are the “nice to have” metrics that help you troubleshoot, but they never get to drive strategy alone.
Examples:
- CTR, CPC, CPM, quality score, engagement rate
- Time on page, scroll depth, bounce rate
- Open rate, click-to-open, deliverability indicators
- Rankings, impressions, share of voice
The discipline: if a diagnostic metric moves but your business and channel truth metrics don’t, you don’t declare victory. You debug.
How to rebuild your measurement system in 90 days
This isn’t a multi-year transformation. You can materially improve decision quality in a quarter if you’re ruthless.
Step 1: Run a metric kill list
Pull your main dashboards. For each metric, ask two questions:
- Can we name a specific decision we’ve made in the last 90 days because of this metric?
- If this metric disappeared tomorrow, would we materially fly blinder on profit or growth?
If the answer is “no” to both, it goes on the kill list. Hide it from executive views. Keep it only in specialist dashboards as needed.
The goal: compress down to the 10-15 metrics that truly matter across the three layers above.
Step 2: Rebuild around marginal ROI, not averages
Marketing Week is right: marginal ROI is about to matter a lot more than blended averages. Rising media, logistics, and fulfillment costs (see Amazon’s surcharges) make “good enough” ROAS deceptive.
At a minimum:
- Plot marginal CAC vs spend for your top 3-5 channels
- Identify the spend point where incremental CAC breaks your payback or margin guardrails
- Shift planning conversations from “What’s our ROAS?” to “Where does the next dollar go?”
This is how you decide whether to test ChatGPT ads, double down on Shopping, or pull back on Instagram: not “is it working?” but “is it beating our next-best dollar?”
Step 3: Put incrementality experiments on a calendar, not a wishlist
With AI-driven bidding and opaque algorithms, incrementality is no longer nice-to-have. It’s survival.
Make it operational:
- Pick 2-3 channels where spend is material (paid social, search, CRM)
- For each, schedule a recurring test: geo holdout, audience holdout, or on/off test
- Decide in advance: what metric, what timeframe, what decision threshold?
Example: “We’ll pause prospecting in 10% of zip codes for 4 weeks. If contribution margin per new customer in holdout falls by >8% vs control, we treat prospecting as incremental at current settings.”
This is how you cut through AI black boxes and get real answers.
Step 4: Treat AI and SEO as one system, not separate fiefdoms
Headlines about AI content, AI search, Google Web Guide, and agentic shopping all point to one thing: “SEO” is no longer just about ranking blue links.
Operationally, that means:
- Measure topic-level performance, not page-level vanity metrics
- Track query families and intent clusters, not individual keywords
- Monitor AI overview and chat visibility where possible, and correlate to brand search and direct traffic
- Include AI-generated and human-generated content in the same performance framework: revenue per 1,000 sessions, assisted conversions, and cannibalization
The point is not “Is AI content bad for SEO?” The point is “Does this content, in this cluster, improve profitable demand capture or not?”
Step 5: Rewrite your creative and brand briefs around measurable problems
Search Engine Land nailed it: if you can’t say what problem your brand solves, AI won’t either. That’s also true for your media mix and your measurement.
Update your briefs so every campaign answers three questions:
- What problem are we solving for the customer? (In plain language.)
- What behavior change do we expect? (Search more, trial, repeat, refer, upgrade.)
- What is the smallest set of metrics that will prove or disprove that behavior change?
Then wire those metrics into your dashboards and experiments before you launch, not after.
What this looks like for real operators
When you talk to teams that are actually hitting numbers in this environment, a few patterns show up:
- Media buyers talk in P&L terms. They know contribution margin, payback, and marginal CAC by heart. ROAS is a diagnostic, not the headline.
- SEO and paid search share a single query map. Cannibalization is measured and managed, not discovered after the quarter closes.
- AI tests are scoped like product experiments. Each one has a clear hypothesis, holdout, and decision rule tied to profit, not just clicks.
- CRM is judged on incremental revenue and churn. Open rate screenshots no longer make it into board decks.
- Dashboards got smaller. Exec views fit on one screen. Channel specialists have deep views, but everyone agrees on the small set of “truth metrics.”
None of this requires a new buzzword or a seven-figure martech overhaul. It does require the one thing AI can’t give you: the discipline to stop worshipping the wrong numbers.