Field notes — Enterprise media

Tests I'd want to run.

I don't know Anthropic's media strategy, and I'm not going to guess at it and get it wrong. But a few things hold no matter what the strategy is. They come down to B2B first principles, and most teams never actually test them. Here's how I would.

Max Massengill  ·  for the Senior Media Manager role  ·  a point of view, not a plan

The assumption underneath everything

Almost all B2B media rests on one belief nobody audits: that the targeting data works. That you can actually reach the people who matter, whether that's IT decision-makers, developers, or a line-of-business buyer, at the companies you care about and with real intent. Not just pay a premium for a list that says you did.

If that belief is wrong, everything built on top of it is theater. The segmentation. The personalized creative. The account tiers. All of it.

So before optimizing anything, I'd test the substrate. Then, and only then, the strategy.

The experiments
T1 GATE: none — run this first

Hypothesis

The audience data most B2B media runs on is materially less accurate than the industry assumes. Accurate enough that broad proxies can rival "precision" targeting.

Why I believe it There's an old targeting study I've never forgotten: reaching a broad demographic hit IT decision-makers about as well as buying third-party "ITDM" data did. That was years ago and the tech has moved, so it needs revisiting. But pick your audience, developers, security leaders, finance buyers, and the story tends to repeat. In twelve years I've watched match rates and audience graphs get bought, optimized against, and almost never audited. That's a strange thing to build a whole discipline on.

▶ How I'd test it

Three methods, triangulated. Two prongs to the data question: is the graph real (deterministic), and does the modeled audience actually contain who it claims (probabilistic)?

  1. Control-group match test (deterministic). Upload a real contact list from Clay-style enrichment, and alongside it a batch of fake and misspelled emails. Push both into the platforms and onboarders (Google, Meta, LinkedIn, LiveRamp). If the fakes match at any real rate, the graph is inflating matches, and your live match rates are suspect. It's a placebo arm for an ad platform.
  2. Account-overlap test (probabilistic). Run tightly targeted campaigns against a small set of named accounts, say ten Fortune-scale companies, and drive on-site form fills. Then check how many captured companies actually sit inside those ten. Leakage is the signal. Run it across providers and you get a ranking, not an opinion.
  3. In-ad survey test (probabilistic). Serve rich-media in-banner surveys to a modeled audience: "which of these companies do you work for?" With enough volume you get a real statistical read on how accurate the audience is.

What would prove it

Fakes matching. Low target-account overlap on conversions. Survey-confirmed accuracy below a bar we set going in.

I'd size all three for statistical power before spending a dollar. An underpowered test is just an expensive guess.

The decision it informs

If the data is unreliable, the whole media go-to-market changes. You stop hyper-segmenting on the media side and shift to broad reach plus a message-driven approach. If it's tolerable, you've earned the right to run T2. This is a media call, not a sales one. Tight segmentation can still be exactly right for CRM and sales outreach, where you're working known contacts, not a probabilistic ad graph.

T2 GATE: run only if T1 ⇒ data is usable

Hypothesis

When creative is held constant, broad, algorithm-led targeting performs at least as well as hyper-segmentation, and often better.

Why this needs a test, not an opinion This is one of the most contested questions in B2B, and almost nobody actually tests it. The research leans one way: LinkedIn's B2B Institute, and Binet and Field's long-and-short-of-it work on reach and brand-building, keeps finding that broad reach beats narrow targeting. My gut leans the other way, because B2B audiences are genuinely niche and the way you speak to a vertical, a persona, an intent stage really does matter. So there's the tension. The evidence says go broad. The instinct says go narrow. And the reason B2B stays narrow anyway isn't evidence, it's career risk. Broad targeting is almost impossible to get signed off, because if it doesn't work you look like you forgot how to do your job. That is exactly why it deserves a clean test instead of a strong opinion.

▶ How I'd test it

Design

Two arms, the same creative in both, personalized at the segment level (vertical, persona, intent stage), not one-to-one. That holds creative constant so the only variable is targeting. Arm A: hyper-segmented (size, vertical, intent, title). Arm B: a broad seed, and let platform optimization decide who to serve.

What would prove it

A statistically significant difference in qualified-pipeline efficiency between the arms, measured on downstream quality, ie SQLs and pipeline, not CPL. Optimize to CPL and you'll only prove the algorithm is better at buying cheap junk leads. That's the trap, not the finding.

The decision it informs

If broad wins or ties, you finally have the evidence to make the move the research has pointed to for years, and the cost savings that come with it. If hyper-segmentation wins, you keep it, but now you're funding it on proof, not fear. Either way the real follow-on question becomes T3.

Why run these in order Because each answer decides whether the next test is even worth running. If T2 shows hyper-segmentation clearly wins, I stop. That's the answer, and I never need T3. If it doesn't, T3 becomes the interesting question: is it the targeting that mattered, or the creative? Testing in sequence keeps it cheap and fast, and means I'm never spending budget on a question the last result already settled.
T3 GATE: run only if T2 ⇒ segmentation did not win

Hypothesis

Under broad targeting, one broad, vertical-agnostic message performs at least as well as a full library of segmented creative. If it does, the whole go-to-market gets radically simpler and cheaper.

Why I believe it If the algorithm is choosing who sees the ad, the question stops being who and becomes what. Does message precision still earn its cost, or is it complexity we're paying for out of habit?

▶ How I'd test it

Design

Broad targeting in both arms. Arm A: a single broad message. Arm B: the full segmented creative library, served by the platform.

What would prove it

No meaningful lift from segmented creative means you collapse to broad and save the production cost. Meaningful lift means you keep the creative variety but drop the targeting overhead.

The decision it informs

This is the difference between a lean media operation and an expensive one. It tells you whether creative complexity is an investment or a tax.

How the answers fork the strategy
Start · T1Is the targeting data actually real?
NO ↓
Broad reach, message-led.Stop paying for precision that isn't there. Reallocate to coverage and a message that works without micro-targeting.
YES ↓
T2Does hyper-segmentation beat the algorithm?
YES ↓
Fund the classic playbook.Hyper-segmentation is earning its keep. Keep segmenting, and invest in it with confidence, not habit.
NO ↓
T3Does segmented creative still win under broad targeting?
YES ↓
Let the algorithm target. Invest in creative.Drop the targeting overhead, keep a rich, segmented creative library.
NO ↓
Collapse to broad and broad.Broad targeting, one strong message. The simplest, cheapest engine that actually works.

None of this is the strategy. It's how I'd pressure-test the assumptions any strategy would sit on.

Answer these three and you're not guessing about how to reach enterprise buyers anymore. You know. That's the work I'd want to do here.

— Max Massengill