The Measurement Vendor Landscape in 2027 — Who Consolidates, Who Disappears

I want to write the issue that is most awkward for me to write.

I run a company in the category I am about to describe. Everything below should be read with that in mind, and I have tried to earn the reader's trust by being specific about where my own company is exposed rather than only where everyone else is. If that section is not the most uncomfortable part of this issue, I have not done it properly.

Here is the claim.

The marketing measurement category is entering a compression. Between roughly now and 2029, a market that currently supports something like forty credible vendors will support meaningfully fewer, and the sorting will not happen on product quality alone. It will happen on three structural forces that most vendors in the category are still describing as marketing problems rather than existential ones.

I have watched enough categories consolidate to know that the interesting question is never who is best. It is who is structurally positioned to survive the thing that is about to happen. Those are different questions and they frequently have different answers.

One. The buyer changed, and the new buyer is harder to bluff.

I wrote in Issue #2 that measurement purchasing is migrating from the CMO to the CFO, and in Issue #4 that the question has escalated again to the board. That migration is now the single most consequential fact about this category's economics.

A vendor optimised for CMO-narrative selling beautiful dashboards, compelling story, strong brand, an emotionally resonant demo was well adapted to the buyer of 2019. That vendor is poorly adapted to a procurement led by someone whose professional instinct is to distrust confident numbers. CFOs ask what would have to be true for the number to be wrong. Most measurement products in this category have never been asked that question by anyone with the authority to walk away.

Two. Agents changed the product requirement.

A measurement layer built for a human to read once a week is a fundamentally different artefact from one an agent queries continuously. The human tolerates ambiguity, applies judgment, and knows to be sceptical of a number that moved thirty percent overnight. The agent does none of that.

This means the vendors who survive have to ship something most of the category has not built: machine-legible calibrated state, with uncertainty expressed in a form an agent can actually reason about, and provenance attached to every number so a recommendation can be traced back to the experiment that justified it. That is an engineering programme measured in years. A vendor discovering the requirement in 2027 is discovering it too late.

Three. The core product is being repriced toward zero.

This is the one that actually kills companies. Attribution as a standalone product is on a path to becoming a commodity feature, because the market is slowly learning that it costs almost nothing to produce an attribution number and therefore an attribution number carries almost no information. A business whose revenue is concentrated in a product the market is learning to value at zero has a limited number of quarters to become a different business.

I know this one from the inside. It is the decision I described in Issue #1, and it cost us roughly a fifth of our revenue across two quarters.

It is worth pausing on how a serious industry ended up with a product it now has to unwind, because the mechanism is not stupidity and it will happen again in the next cycle if nobody names it.

Rory Sutherland has been making an argument lately that I think explains it better than anything written from inside the measurement industry. His version is aimed at marketers, but it applies with more force to the people who sell them tools. The argument is that measurement is not neutral because it is far easier to produce evidence for some activities than others, a culture that funds in proportion to evidence will drift toward whatever is measurable rather than whatever works. He calls this, roughly, the logic trap.

Now apply that to vendors rather than marketers.

A measurement company sells what it can demonstrate. Attribution demonstrates beautifully it produces a number for every dollar, immediately, with no forgone revenue and no waiting. Incrementality demonstrates terribly by comparison. It requires the customer to deliberately not spend money, wait several weeks, and accept an answer with a confidence interval attached.

Given those two products and a sales quota, the entire category made the same choice, repeatedly, for fifteen years. Not because anyone believed attribution was more truthful. Because it was more sellable. The category optimised for demonstrability and called the result accountability.

That is the logic trap operating one level up the supply chain, and it produced a forty-vendor market built on a primitive that is about to be repriced.

I am going to describe kinds of companies rather than named ones. Partly because naming competitors in a letter like this is bad manners, and mostly because the archetype is the useful unit if you are evaluating vendors, what you want is to know which kind of company is in front of you.

The attribution incumbents. Largest revenue bases, strongest logos, most sophisticated go-to-market, and structurally the most difficult position in the category. They are the textbook Christensen case: the existing product is profitable enough that rebuilding underneath it feels premature every single quarter until it is too late. Some will make the pivot it requires a CEO willing to eat a revenue hole deliberately, which is rarer than it sounds. Most will not, and will be acquired for their customer relationships rather than their technology.

The methodology specialists. Companies that do one thing with real rigour — geo-testing, or Bayesian marketing mix modelling, and little else. I want to be careful here, because there is a lazy narrative that says point solutions lose. They do not necessarily. A specialist with genuine methodological credibility has three good outcomes available: stay deliberately small and premium, become the trusted engine inside somebody else's platform, or get acquired at a good price by a platform that needs the credibility. What a specialist cannot do is become a platform by adding two more methods quickly. That takes years and most attempts produce three mediocre methods instead of one excellent one.

The unified platforms. Companies attempting to run marketing mix modelling, incrementality testing, and calibrated attribution as one orchestrated system. Highest ceiling in the category and by some distance the hardest to execute. This is where Lifesight sits, so treat the following as self-assessment rather than analysis: the characteristic failure mode of this archetype is overclaiming. Three methods is three times the surface area on which to be wrong, and the temptation to describe an integration as more seamless than it is scales accordingly. The unified platforms that survive will be the ones that were boring about their limitations early.

The platform-native tools. Measurement products built by the advertising platforms themselves. These are not going anywhere, they are free, they are beautifully integrated, and they will retain a floor of the market forever. They also cannot be independent, because the entity reporting the results is the entity selling the media. That is not an accusation of bad faith. It is a structural conflict that no amount of good engineering resolves.

The consultancies. Covered in Issue #6. The six-month project producing a static model is being replaced by continuous measurement inside a platform. The talent is excellent and is already reallocating. The revenue model is contracting.

Here is the part of category consolidation that people who build products consistently underestimate, and it is worth being honest about because it cuts against my own interests.

There is an old advertising-industry adage that Sutherland has done more than anyone to popularise in the modern era: nobody ever got fired for buying IBM. His broader point is about defensive decision-making — that in most organisations, the person making a purchase is not optimising for the best outcome. They are optimising for the outcome they can most easily defend if it goes badly. Those two objectives produce different purchases, and the gap between them is where a great many superior products go to die.

Marketing measurement is an unusually pure case of this, because the buyer is purchasing a system whose entire function is to tell them uncomfortable things. A measurement platform that reports a wide confidence interval, tells the CMO her favourite channel is underperforming, and refuses to produce a number it cannot defend is doing its job correctly and creating career risk for the person who bought it. A platform that produces confident, flattering, precise-looking numbers is doing its job badly and is much easier to defend in a quarterly review.

So the compression will not be clean. Some vendors with genuinely inferior methodology will survive it comfortably on the strength of being the safe, familiar, defensible purchase. Some vendors doing the most honest work in the category will not make it. I would rather say that plainly than pretend the market is a meritocracy, because a CMO who understands this dynamic can consciously decide not to be governed by it and the ones who do are the ones I most enjoy working with.

If I were buying rather than selling, this is the list. It is built on a specific principle, which I will name after the questions.

One. What have you published that would embarrass you if it turned out to be wrong? A small number of vendors in this category publish live accuracy benchmarks against holdouts — dated, checkable records of how their predictions performed against what actually happened. It is rare, and it is the single cleanest signal available, because a vendor who publishes a number that can be checked has accepted a cost a weaker vendor could not afford. Ask what they have put in public. A vendor with nothing that could embarrass them has told you something.

Two. Show me a customer where your recommendation was to spend less. Every measurement vendor claims independence. The demonstration of independence is a documented case where the system told the customer to reduce spend on the channel they were emotionally committed to, and the vendor said it anyway. If the answer is a long pause, the independence is aspirational.

Three. What can your system not measure? I wrote in a recent issue that distinctive brand assets, creative quality, and category-creation spending largely resist measurement at actionable resolution. Any vendor who tells you their platform measures everything is either not thinking carefully or hoping you are not. The right answer is a specific list delivered without defensiveness.

Four. Can an agent read your output, or only a human? Ask to see the machine-legible form. Ask how uncertainty is represented in it. Ask whether a recommendation carries provenance back to the experiment that justified it. If the answer is a dashboard screenshot, you are buying a product built for the previous decade.

Five. Who signs off on the methodology — you or us? There is a meaningful difference between a vendor who hands you a model and a vendor who expects your team to interrogate and approve the assumptions. The second relationship is more work and produces measurement your CFO will actually defend.

The principle underneath all five is one I have borrowed from Sutherland and applied somewhere he has not.

His account of why advertising works leans on costly signalling — the idea, borrowed from evolutionary biology, that a signal is credible in proportion to how expensive it would be to fake. Extravagant advertising works partly because it is extravagant; the company is demonstrating a confidence it could not afford to fake.

Every one of those five questions is a test for a costly signal. Publishing a falsifiable accuracy number is expensive — you might be wrong in public. Telling a customer to spend less is expensive — it shortens their contract. Naming your own blind spots is expensive — it hands your competitor a talking point. Building machine-legible provenance is expensive — it is years of engineering with no demo value. Inviting the customer to audit your methodology is expensive — it is much harder than being trusted.

Cheap talk is cheap for everyone, which is why it tells you nothing. Ask what this vendor has done that a bad vendor could not have afforded.

I said at the top that this should be the uncomfortable part.

Lifesight is a unified platform, which means we carry that archetype's characteristic risk. Running three methods in orchestration is genuinely harder than running one well, and the honest statement of our position is that the reconciliation layer — the part that decides what to do when the methods point in different directions — is the part still being refined. There are edge cases where it does not hold confidently: small experiment samples, low-signal channels, brands at lower budget bands. We are better at saying this in a letter than our category is at saying it in a sales cycle, including us, and I would like that gap to be smaller.

We are also exposed to the defensive-purchasing dynamic I described above, and I do not think we should pretend otherwise. A platform whose entire pitch is we will tell you things you do not want to hear, with wider intervals than the tool you are replacing is a harder purchase to defend internally than the incumbent. We win those deals when there is a CFO in the room. We lose a meaningful number of them when there is not, and the ones we lose are frequently lost to a more confident, less honest number.

The strategic bet the company is making is that the CFO is in the room more often every year. Issues #2 and #4 are, in a sense, that bet written out. If I am wrong about the direction of the buyer, I am wrong about a great deal else.

One. Whether the first significant attribution-primary vendor announces a deliberate rebuild. Not a feature addition — a public statement that the core product is being repositioned and the revenue consequence is accepted. The first company to do it will take a beating in the short term and will be studied in business schools if it works. I expect it within eighteen months.

Two. Whether procurement templates start asking methodology questions. The RFP is where category shifts become irreversible. When the standard enterprise measurement RFP includes describe what your system cannot measure and provide a customer reference where you recommended reduced spend, the compression accelerates sharply, because those questions cannot be answered by a vendor who has not done the work. I have started seeing the first of these. Not many yet.

Three. Whether consolidation produces better products or merely larger ones. This is the outcome I am least confident about. Category consolidation frequently produces companies that are bigger and worse — three mediocre methods stapled together, sold as a platform, defended by a sales organisation. If the compression ends with the industry's honest engineering absorbed into companies that then stop publishing anything falsifiable, the market will have consolidated and gone backwards simultaneously. I do not think that is the likeliest outcome. I think it is the one worth naming out loud so it is harder to arrive at quietly.

Two things.

The first is that I am describing a market I compete in, and no amount of self-criticism fully corrects for that. The archetype framework above is mine, it happens to place my own company in the position with the highest ceiling, and the reader should apply a discount accordingly. I would encourage anyone using this issue as an evaluation aid to use the five questions rather than the archetypes. The questions work regardless of who wrote them. The archetypes are a vendor's map of his own market.

The second is that my timeline may be wrong in the direction of impatience. I have believed for two years that this compression was imminent, and it has been slower than I expected each year. Categories consolidate more gradually than the people inside them anticipate, because incumbency is worth more than challengers think and switching costs are higher than product people ever model. If you are a vendor reading this and feeling the pressure, you probably have longer than the argument suggests. If you are a buyer reading this, that is precisely why the questions matter — the market will not sort this out for you on a useful timescale, so you will have to.

If you have recently run a measurement RFP, I would genuinely like to see the questions you asked. I am trying to build a picture of how fast the buying process is actually changing, and the RFPs are the most reliable evidence available.

Thanks for reading the seventh one.

Tobin Co-Founder & CEO, Lifesight July 2026