STOP WORSHIPPING THE STACK: HOW ELITE MEDIA BUYERS REALLY USE AI & AD SPY TOOLS TOGETHER

Most marketers don’t have a media strategy problem. They have a stack addiction.

Walk into almost any performance team today and the conversation starts with tools: “What AI copywriter are we using?” “Which CTV optimization platform?” “What’s in our spy stack?” Every plateau, every missed target, every creative slump gets translated into the same question: “What should we buy next?”

That’s stack‑first thinking—and it’s quietly destroying your edge.

Martech and AI tools were supposed to make decision-making sharper. Instead, by centering strategy around the stack, teams have created exactly what MarTech calls “more noise than capability”: overlapping dashboards, redundant analytics, and a dozen different “single sources of truth” that never quite agree. You end up spending more time reconciling reports than reading the market.

The core mistake is sequence. The way elite buyers work is simple:

• Strategy defines the problem.
• An audit reveals the gap.
• Only then does a tool earn the right to exist in the stack.

Most teams flip that order. They chase capabilities first—creative AI, predictive scoring, audience discovery, CTV optimizers—then reverse‑engineer a “strategy” to justify the spend. As MarTech warns, this is how you end up with three analytics platforms parsing the same data in slightly different ways. Those tools don’t triangulate truth; they fragment it. People inevitably gravitate to whichever dashboard best confirms what they already wanted to believe.

AI makes this even more seductive. A model that promises “agentic optimization” or “self‑driving funnels” sounds like a strategy upgrade when it’s really just another execution engine. In streaming, research cited by AdExchanger shows buyers overwhelmingly want AI to improve planning and optimization, yet only a small minority are willing to hand campaigns over to full autonomy. That tension is the tell: sophisticated teams know AI is powerful, but they don’t trust it to decide what “winning” means. Tools should accelerate judgment, not define it.

Inside most organizations, though, AI is being bolted on in silos. One specialist uses ChatGPT for ad hooks, another leans on a different model for emails, media leans on a DSP’s algorithm, and the CTV team runs an entirely separate “AI layer” on top of that. As MarTech points out, nothing connects; value doesn’t compound. You don’t get a coherent learning system; you get scattered, shallow outputs with wildly inconsistent quality.

That scatter has real costs. Every new AI or ad‑spy tool becomes:

• Another place for data to drift and definitions to change.
• Another interface your team has to learn—usually without real training.
• Another interpretation of performance, giving stakeholders ammunition to cherry‑pick numbers.

The result isn’t faster optimization; it’s a slower, more political decision cycle. Leaders spend precious time arguing whose dashboard is “right” instead of asking better questions about audiences, offers, and messaging.

Meanwhile, the external environment is getting harder, not easier. AI is moving from tactical execution into the planning layer. Research summarized by illumin shows organizations are starting to embed AI into scenario modeling, forecasting, and decision intelligence long before launch. That shift favors teams that know exactly what they’re trying to simulate and measure. It punishes those that are still asking their stack to tell them what their strategy should be.

Stack‑first thinking also blinds you to risk. When teams are told to “just use AI,” they prioritize speed over judgment, pasting internal strategy docs and customer data into public models and hoping for the best. As MarTech has warned, that’s not just sloppy workflow; it’s a long‑tail brand and security risk that hits you before a single ad goes live.

The uncomfortable truth is this: if your first instinct in a slowdown is to shop for another tool—another AI assistant, another spy platform, another optimization layer—you’re signaling that you don’t really understand your problem yet. Elite media buyers don’t start with “What stack should we use?” They start with “What edge are we trying to build?” and treat every tool, AI or otherwise, as a downstream decision in service of that answer.

Before elite buyers touch a spy tool, an AI planner or a DSP login, they do something unfashionable: they write the problem down in plain language.

Not “our CPMs are high” or “CTR is down.” Those are symptoms. The real exercise is defining, with painful specificity, what you’re trying to change in the business and what you believe is blocking it. Only then does the stack become relevant.

This is the quiet superpower behind high-performing teams. They don’t start with, “How do we use AI?” They start with, “What decision are we trying to make better, faster, or at greater scale?” As one analysis of martech bloat put it, tools are a downstream decision; strategy defines the problem, the audit identifies the gap, and only then does a tool earn a place in the stack, not the other way around, as.

Top buyers operationalize that principle with a simple sequence.

1. Define the business outcome, not the channel metric

Elite media buyers frame objectives in language a CFO would sign: incremental revenue from a segment, reduced CAC on a specific product line, lift in qualified leads from a priority region. “Increase ROAS” is not a strategy; “reduce blended CAC by 15% in 90 days for first‑time purchasers in Tier‑1 cities” is.

This matters for how you’ll actually use AI and spy tools. If the outcome is, say, “steal share from two specific competitors in streaming TV,” that immediately narrows the questions you’ll ask your ad intelligence stack: which publishers they over‑index on, what creative hooks they lean on, how aggressively they rotate offers on CTV versus social. The tools become lenses on a clearly defined problem instead of shiny dashboards begging for attention.

2. Translate the outcome into concrete decision questions

Once the business result is clear, top buyers break it into decision questions that AI can assist with and that ad spy tools can illuminate. For example:

• Which audiences are most likely to deliver this outcome at our target economics?
• Which messages and offers are most likely to move those audiences now?
• Which placements and formats are underpriced for those audiences?

Notice what’s missing: “Which AI platform should we use?” The platform question only shows up after you’ve articulated the decisions you need help with.

This is exactly where AI is strongest. Research highlighted in illumin’s overview of AI advertising trends shows AI moving upstream into decision intelligence: scenario modeling, forecasting, and planning before a single dollar is spent. Elite buyers exploit that by feeding clear decision questions into their models instead of vague prompts like “make a media plan.”

3. Audit your current signal before adding new tools

Once the decision questions are defined, top buyers perform a brutally honest audit: What signal do we already have? Where is it strong? Where is it unreliable, delayed, or missing?

That audit spans:

• Performance data (by cohort, channel, creative concept, and journey stage)
• Competitive signal (what competitors are spending on, what they’re testing, how often they rotate)
• Consumer insight (what people say, search, and actually do)

The goal isn’t to justify the stack you already own. It’s to pinpoint the specific gaps that are blocking better decisions. As one critique of stack‑first thinking noted, piling on tools before this audit only adds “more conflicting versions of the story” and more places where data drifts, making it harder to see what’s actually working, a dynamic.

When elite buyers do bring AI into this audit, they treat it as an analyst, not an oracle. They’ll ask an AI model to summarize cross‑channel performance for a segment, or to cluster creative concepts by theme and outcome, but they maintain ownership of interpretation. That human‑first posture mirrors what FreeWheel’s research into streaming TV buyers found: only a small minority are willing to hand over fully autonomous control, with most preferring AI that surfaces insights and automates grunt work while they retain strategic oversight, as AdExchanger’s analysis of AI in streaming TV described.

4. Decide how AI and spy tools will help — before choosing which ones

Only after the problem and gaps are clear do top buyers define roles for tools:

• “We need AI to simulate budget reallocation scenarios across CTV, YouTube, and paid social before we change anything live.”
• “We need a spy tool to monitor competitor creative rotation weekly for our three priority audiences, categorized by offer and hook.”
• “We need AI to cluster competitive ads into messaging territories we’re under‑leveraging.”

Notice that these are job descriptions, not brand names. This is how you stay out of “adopt AI because we’re supposed to” mode — the trap where individual team members quietly bolt random tools onto their workflows and create fragmented, low‑quality outputs that don’t connect, a pattern MarTech’s critique of siloed AI usage flagged across many organizations.

By the time an elite buyer logs into an AI planner or pops open an ad spy dashboard, the hard work is already done. The strategy has defined the problem. The audit has defined the gaps. The tools are there to answer specific, pre‑defined questions — not to decide what the questions should be.

AI’s real job on an elite media team is painfully unsexy: shrink the distance between a decision and the execution of that decision. It is there to move hands faster, not to move brains out of the loop.

When teams confuse those roles, they either let AI overstep into “autopilot strategist” territory or underuse it as a glorified spell-checker. Both mistakes show up the same way in the numbers: sluggish testing velocity, shallow learnings, and a stack that looks impressive while ROAS quietly stalls.

Look at how serious buyers actually talk about AI. In streaming, most buyers say the biggest upside is in planning and optimization, not in handing off the steering wheel. Nearly half call out campaign setup, pacing, and performance monitoring as the domains where AI can immediately help, yet only a small minority are comfortable with fully autonomous campaign management, according to research cited by AdExchanger. That’s the right posture: AI as an execution accelerator that sits under a human strategy, not above it.

Practically, that means your prompts, agents, and workflows should start from decisions you have already made:

• “We are going to test three new angles against women 25–44 in CTV this week.”
• “We need variant hooks for this winning UGC format that speak to price sensitivity instead of convenience.”
• “We believe mid-funnel frequency is too low; build a schedule that pushes average exposures from 2 to 4 without blowing budget.”

Once those calls are made, AI’s job is to compress everything between the idea and the live campaign. A capable assistant can turn that one sentence into audience lists, structured naming conventions, platform-ready asset plans, QA checklists, and even pre-written briefs in minutes. The buyer still decides what to test; the system takes over the “how fast can we stand this up?” problem.

This is exactly where AI is already strongest. Modern tools excel at automating repeatable tasks, surfacing optimization opportunities, and running scenarios across messy datasets. As illumin notes, AI can now ingest business objectives, historical performance, audience insights, and budget parameters to model different campaign paths before a single dollar is spent. That doesn’t mean you let the model pick your plan; it means you stop wasting a strategist’s time on spreadsheet gymnastics and let them judge tradeoffs across scenarios instead.

But this only works if you resist the urge to “adopt AI” for its own sake. Teams that are told to “just use AI” invariably end up with random pockets of automation and no shared idea of what the system is for. One person uses a chatbot for headlines, another spins up a separate tool for email, and none of it connects back to the actual business problem. As MarTech warns, that siloed, tool-first mentality produces fragmented outputs and shallow quality. You’re not accelerating execution; you’re just producing mediocre assets faster.

Elite buyers flip the order. They define the problem in human language, decide the strategy, then deliberately assign AI to the slowest or most error-prone steps between decision and deployment:

• Turning a measurement framework into platform-specific tracking setups.
• Translating a winning insight from an ad spy tool into 20 structured test variants.
• Cleaning and stitching performance data so you can see creative, placement, and audience in one coherent view by the next morning.

They also draw a hard line: AI does not get to invent business objectives, redefine brand guardrails, or silently change targeting priorities because “the model found something.” As AdExchanger’s coverage of buyers makes clear, the market wants intelligent assistants, not black-box agents running campaigns on their own.

The net result is a team that looks slow only at the moments that matter—when you are deciding what game to play—and blisteringly fast at everything that comes after. AI and ad spy tools become force multipliers precisely because they are not allowed to make the calls; they’re forced to make the calls cheaper, cleaner, and faster to execute.

Spy tools are not cheat codes. They’re binoculars.

Elite buyers don’t open a spy platform looking for “the winning ad” to rip. They open it to answer a much narrower question: “What does the battlefield actually look like right now?”

That shift—from copy hunting to competitive intelligence—is everything.

Copy-paste behavior is exactly the “workslop” problem Greg Kihlstrom described, where teams crank up volume with AI and tools instead of tightening thinking, flooding channels with low-quality, derivative work that doesn’t move the numbers, as he argued in MarTech. Spy tools become accelerants for that mediocrity when they’re treated as vending machines for hooks, angles, and creatives. Elite teams refuse to use them that way.

Instead, they define a question first, then go in to collect evidence.

For example:

• “Who’s clearly overpaying to own this audience, and how are they structuring their funnels?”
• “What storytelling patterns keep resurfacing in my category over the last 90 days?”
• “Which competitors have quietly pivoted offers, and what does that say about pressure on pricing or margins?”

A good spy session for a senior buyer looks more like a reconnaissance briefing than a swipe file. You’re mapping:

• Channel and format priorities – Who’s clearly leaning into short-form video versus static, or CTV versus social? When FreeWheel’s research found buyers care most about AI that tightens planning and optimization, this is what they meant: better visibility into where money is flowing so you don’t plan in a vacuum.
• Offer architecture – Trials vs. discounts vs. bonuses, guarantees, bundling. You’re not stealing the offer; you’re inferring unit economics, cash-flow needs, and risk tolerance.
• Narrative territory – Which benefits are overcrowded, which objections are going unanswered, and where white space exists for your brand to own a distinct story.

Then AI comes in—not to hallucinate strategy, but to digest what you’ve just seen.

You can drop a batch of competitor ads into an AI assistant and ask it to cluster by angle, identify recurring claims, map objections addressed versus ignored, or build a grid of “who is talking to which psychographic.” This is the kind of upstream decision intelligence that AI is increasingly enabling, where it helps model scenarios and surface patterns before you spend a dollar, as illumin described in its breakdown of AI as a strategic partner in planning.

Notice what’s happening there:

• Spy tools give you observable reality in the market.
• AI gives you compressed analysis of that reality.
• You, the buyer, still make the calls.

You’re not asking AI, “Write me ads like Brand X.” You’re asking, “Given these 200 competitor creatives, what are the dominant promises and what’s missing?” That keeps AI in its proper role—speeding up synthesis and execution—without outsourcing judgment, the same human-over-autopilot posture buyers expressed when only 22% said they wanted fully autonomous campaign management, according to.

The practical payoff is twofold.

First, you avoid the sameness trap. If everyone is watching the same ads and lazily cloning them, any temporary advantage evaporates. Your CPAs rise in lockstep with your competitors’ because you’ve voluntarily entered a race to the middle.

Second, you build a more durable sense of what actually matters in your category. Over time, disciplined reconnaissance plus AI-assisted analysis gives you your own proprietary “market memory”: how competitors respond to seasonality, what messages fatigue fastest, how often offers cycle. That institutional knowledge is something no single tool can sell you, and no stack can replace.

Spy tools can absolutely fuel ideation. But on an elite team, they serve the strategy you already clarified, not the other way around. They are there to inform your positioning, not erase it.

Strategy comes first or nothing else matters.

Elite buyers don’t open their spy tools, prompt their AI, or touch a bid until they’ve answered three questions in plain language:

• Who are we trying to move?
• What are we trying to make them do next?
• What constraints are non‑negotiable? (budget, margins, inventory, brand risk)

This is the upstream work that too many teams try to outsource to “the stack.” As one martech analyst argued, tools should be a downstream decision, not the place where you invent a plan after the fact; when you reverse that order, you just add dashboards without adding clarity about what’s working, a point driven home in a piece urging marketers to stop building strategy around your martech stack.

Once that spine is defined, the combined workflow kicks in.

First pass is Spy.

You go into Meta’s Ad Library, TikTok Creative Center, or a paid spy tool with a tightly scoped brief: “Show me high‑frequency offers for mid-ticket supplements targeting women 35–54 in the last 30 days.” You’re not looking for “the winner” to copy; you’re mapping realities:

• Channels your competitors are leaning on
• Dominant angles and offers (trials, bundles, guarantees)
• Landing page archetypes that seem to sustain spend
• Creative structures (UGC first-person hooks vs polished brand spots)

The output is a short competitive field report, not a swipe file. It might say, “Category is over‑indexed on discounts and under‑indexed on social proof,” which becomes a hypothesis: “Let’s test a proof‑heavy control against a value‑stack variant.”

Now AI shows up as a force multiplier for that hypothesis.

You feed it structured inputs: your brief, your field report, your constraints, and your brand rules. You don’t ask, “Write me a funnel.” You ask:

• “Generate 10 hook variations that dramatize risk reversal without mentioning discounts.”
• “Turn this long‑form testimonial into three 15‑second scripts tuned to skippable video.”
• “Propose three budget allocations that prioritize Roku and YouTube in Q4 under a $150 CPA ceiling.”

This is where AI begins to behave less like an after‑the‑fact optimizer and more like a planning partner. Industry research on AI advertising notes that models can ingest objectives, historical performance, and market conditions to model different campaign scenarios before media dollars are spent. The human still chooses the path; the machine makes it possible to see five paths instead of one.

Then you execute and measure.

Here, agentic and automation capabilities shine: pacing checks, anomaly alerts, and continuous bid/creative reallocation. In streaming and CTV specifically, buyers are already using AI to reduce manual effort across campaign setup, pacing, and performance monitoring so decisions can be made faster and at finer granularity, as recent FreeWheel research summarized. Crucially, that same research found that only a small minority of buyers want fully autonomous management; they want augmented execution with human strategy in the loop.

So your measurement loop is deliberately human‑interpreted, machine‑accelerated:

1. AI consolidates platform data, flags outliers, and surfaces patterns.
2. You interpret those patterns against your original strategy (and reality: margins, inventory, LTV, brand).
3. You decide which levers are allowed to move: pricing, creative angle, audience mix, channel weight.

Then you refine, in that order.

Sometimes refine means “do more of what’s working.” But at an elite level, it usually means re‑running the entire workflow at a tighter altitude:

• Strategy: “Our TikTok CTR is great but CAC is off; we’re attracting the wrong cohort.”
• Spy: “Who’s profitably scaling on TikTok with similar AOV? What hooks and landers are they using?”
• AI: “Generate variants of our best hooks that target an older, higher‑LTV segment; suggest landing page rewrites that emphasize substance over novelty.”
• Measure: “Did LTV‑indexed CAC improve, even if top‑line CPA barely changed?”
• Refine: “Shift 20% of budget to the new cohort and spin up a CTV test with adapted creative.”

Behind the scenes, this discipline has a side benefit the stack‑addicted teams ignore: it naturally pushes you to rationalize your tools. When every campaign runs Strategy → Spy → AI → Measure → Refine, bloated point solutions that don’t clearly serve one of those steps become obvious dead weight, echoing the push toward stack rationalization and core tool governance described in a piece urging marketers to look past AI to see where martech is going.

The result isn’t a “full AI stack” or a “full spy stack.” It’s a repeatable operating system where strategy drives the question, spy tools ground you in the market, AI multiplies your options, measurement tests your judgment, and refinement keeps the loop turning.

Before you open a spy tool or brief an AI assistant, you need a simple, ruthless constraint: “This is what success looks like, and this is the box we have to play in.”

Everything else hangs off that.

Start with the economic target: CPA or ROAS

The stack can’t save a bad unit economics story. You need a clear, non-negotiable performance target before you let any platform “optimize.”

• Target CPA if you buy on cost-per-action: “We can afford $80 to acquire a trial user.”
• Target ROAS if you’re closer to revenue: “We need at least a 3.5x blended ROAS on first-order revenue.”

Don’t pull these numbers from a dashboard. Pull them from the business model: contribution margin, payback windows, allowable CAC by cohort, and inventory risk. Modern AI planning tools can help you pressure-test these targets up front. As illumin explains, AI can now model different campaign scenarios against budget and objectives before you spend, instead of only tweaking bids after launch. Use that: feed your historical CPAs, margins, and seasonality into your AI copilot and have it outline realistic CPA/ROAS ranges by channel. Then you, not the model, decide what is acceptable.

Your target becomes the ruler you use later when spy tools tempt you with “cheap” traffic that never pays back.

Define the audience as a behavior, not a persona slide

Elite buyers don’t stop at “25–44, urban, high income.” They define who is moveable and how they behave in the wild:

• New-to-category vs. switchers
• High-intent searchers vs. passive scrollers
• Lapsed customers vs. active fans

Because algorithms increasingly control who actually sees your ads, you’re not hand-picking micro-segments as much as you’re teaching the system what “good” looks like. As one MarTech analysis points out, black-box optimization has blurred direct cause-and-effect: platforms decide who sees what, when, based on downstream signals. Your job is to define which signals matter.

Translate your behavioral audience into optimization events and guardrails:

• “Optimize for completed onboarding, not just leads.”
• “Exclude serial refunders and coupon-only purchasers.”
• “Prioritize users who hit feature X in their first 7 days.”

Write that down before you ever search a spy tool. Then, when you scan competitor ads, you’re not asking “Who are they targeting?” but “Which behaviors are they clearly trying to trigger?” You’ll see why their creative, hooks, and offers look the way they do.

Nail the offer and positioning before you look at competitors

Spy tools are for context, not identity. If you haven’t already decided your offer and angle, you’ll unconsciously conform to whatever is saturating the feed.

Codify three items first:

1. Core offer: What exactly are you selling in this campaign? Subscription, bundle, entry-level SKU, demo, lead magnet?
2. Value equation: Why this, why now, and why at this price or commitment level?
3. Positioning constraint: Where won’t you go? No fake urgency, no bait-and-switch discounts, no MLM-style earnings claims, etc.

This is where AI as “strategic partner” is actually useful. Instead of asking a model to “write 10 Facebook ads,” have it stress-test your offer structure and positioning. According to illumin’s research, AI is increasingly being used to evaluate multiple strategic paths and forecast outcomes before execution. Use that upstream power to compare, for example, “30-day free trial” vs. “$1 trial” vs. “annual plan with 2 months free,” and see which scenarios are likely to hit your CPA/ROAS targets given historical conversion and churn.

You still choose the path—but you’re not guessing.

Lock in constraints: channels, formats, and compliance

Platforms and AI agents will happily chase cheap impressions into any crevice of the internet if you don’t fence them in. High performers set hard constraints at the strategy stage:

• Channels: Where will you not run, even if CPMs look great? Maybe you cap streaming inventory until attribution and incrementality tests catch up. Even in CTV, buyers prefer AI as an assistant under human control, with only 22% open to full autonomy, according to AdExchanger’s survey of streaming TV stakeholders. Your plan should reflect that same bias toward human oversight.
• Creative formats: Decide upfront which formats are in play (UGC-style vertical, high-polish video, long-form advertorial, native, CTV spots) and how many distinct “angles” you’ll support. That keeps AI image/video tools from proliferating endless low-quality variants—the exact “workslop” trap Greg Kihlstrom warns about.
• Compliance and brand safety: Spell out forbidden claims, sensitive categories, and geo/legal nuances before you prompt a model or brief freelancers. For regulated verticals, encode these as explicit rules that every AI prompt references.

These constraints do two things. First, they prevent your AI tools from wandering into high-risk or low-value territory in pursuit of vanity metrics. Second, they give your ad spy workflow a lens: you’re not envious of every clever ad you see; you’re scanning only for tactics that are compatible with your economic targets, behavioral audience, offer structure, and risk tolerance.

Once this strategic box is real—not a vague deck—then it makes sense to open your spy tools and AI assistants. Otherwise, you’re just letting the stack define your strategy for you.