# How to Measure the Halo Effect Between Marketing Channels

TV and brand campaigns make your other channels convert better. That amplification lives between channels, not inside any touchpoint, which is why MTA can't see it and MMM can.

- Canonical: https://mbuzz.co/articles/cross-media-synergy-halo-effects
- Published: 2026-07-23
- Author: Holly Mehakovic, mbuzz (https://mbuzz.co)

---

> **TL;DR:** Channels amplify each other when one channel changes how well another converts. A brand or CTV campaign makes people search your name, so the search ad they'd have scrolled past gets clicked and converts. That extra output is an interaction between two channels, not a touchpoint, so it has no row in the event log for a multi-touch model to credit. MTA splits a fixed 100% of credit across the touches it observed; it has no mechanism to make the pie bigger when two channels run together. Marketing mix modeling estimates synergy directly by adding an interaction term (channel A spend times channel B spend) to a regression on aggregate sales. Use MTA for the journey-level credit split, MMM for the channel-interaction question, and never let one tool claim to answer the other's question.


## The question attribution can't answer

Attribution asks one question: which touchpoint gets credit for this conversion? Baked into that question is an assumption most people never notice. It assumes each channel's contribution is independent and additive, so the whole job is slicing a fixed pie among the touches that appeared in the journey.

Synergy breaks the assumption. When a connected-TV or brand campaign runs, it doesn't only produce its own conversions. It changes how well every other channel converts. Someone sees the brand on TV, searches the brand name a day later, clicks the paid ad they would have scrolled past, and converts at a higher rate than a cold searcher would. The search channel looks more effective. The cause sits upstream, in a channel the search click never recorded.

Here is the plain version of the problem. An attribution model can only credit touchpoints it observed. Synergy is an interaction between two channels, not a touchpoint of its own, so there is nothing in the event log for the model to credit. The TV exposure that made the search click convert usually isn't in the data at all: no click, no server event, no row. And even when both touches are captured, MTA distributes a fixed 100% of credit across them. It has no mechanism to say "these two together produced more than they would apart." That extra output has nowhere to live.

Marketing mix modeling works from the other direction. It's a top-down regression on aggregate spend and outcomes over time, so it never needs to see an individual journey. You add an interaction term, channel A spend times channel B spend, and the regression estimates whether the two together move sales beyond their separate contributions. The coefficient on that term is the synergy estimate.

So the honest one-line version is this. MTA counts the touches it saw and splits a fixed pie. Synergy is the pie getting bigger when two channels run together, and only a model that works on totals instead of touches can see it.

<div class="not-prose my-8 bg-white border border-slate-200 rounded-lg overflow-hidden">
  <div class="px-5 py-3 border-b border-slate-200 bg-slate-50">
    <p class="text-xs font-bold text-slate-500 tracking-[0.15em]">TWO DIFFERENT QUESTIONS</p>
  </div>
  <div class="px-5 py-5">
    <div class="grid grid-cols-2 gap-3">
      <div class="bg-indigo-50 border border-indigo-200 rounded-md p-4">
        <div class="text-[10px] font-bold tracking-[0.15em] text-indigo-500 mb-2">MTA ASKS</div>
        <p class="text-sm text-slate-700 leading-relaxed">Within a converting journey, how should the credit I observed be split across the touches?</p>
        <p class="text-xs text-slate-500 mt-3">Works on touches. Splits a fixed 100%.</p>
      </div>
      <div class="bg-emerald-50 border border-emerald-200 rounded-md p-4">
        <div class="text-[10px] font-bold tracking-[0.15em] text-emerald-600 mb-2">MMM ASKS</div>
        <p class="text-sm text-slate-700 leading-relaxed">If I turn this channel up or down, what happens to total sales, including its effect on the other channels?</p>
        <p class="text-xs text-slate-500 mt-3">Works on totals. Can let the total grow.</p>
      </div>
    </div>
    <p class="text-sm text-slate-700 mt-4">Synergy is only visible to the second question. Selling MTA as if it answered the second question is the category's core dishonesty.</p>
  </div>
</div>

This piece is not "MMM good, MTA bad." Both tools answer a real question well and the other question not at all. The mistake is asking one of them to do the other's job.

## What the evidence says

The claim that upper-funnel exposure lifts downstream channels is not a vendor invention. It shows up across sources with different incentives, and it's worth separating the disinterested evidence from the interested evidence before you weigh any number.

WARC analysis of cross-media effectiveness found that TV advertising improved generic search effectiveness by around 8%, online video by around 20%, and paid social by around 31%. WARC is a credible, method-agnostic source, though it sits behind a paywall, so treat the exact figures as widely-cited rather than independently reproduced here.

A 2021 Nielsen study commissioned by Google looked at what happens when brands add upper-funnel effort. Brands that added upper-funnel to an existing mid-funnel program raised ROI by around 70%. Brands that added upper-funnel to a program already covering mid and lower funnel raised ROI by around 13%. The two numbers describe different starting conditions and are easy to conflate, so the useful takeaway is the direction and the diminishing return, not the precise percentages.

The single most defensible anchor is academic and disinterested. Veverka and Holy, publishing in Applied Stochastic Models in Business and Industry, modeled minute-by-minute organic website visits against TV ad airings for an e-commerce company. They found measurable, immediate visit spikes after airings, varying by time of day, channel, and creative. That study grounds the whole argument. An upstream exposure that produced no click still produces downstream behavior, and a click-based log will file that behavior under search or direct.

<div class="not-prose my-8 bg-white border border-slate-200 rounded-lg overflow-hidden">
  <div class="px-5 py-3 border-b border-slate-200 bg-slate-50">
    <p class="text-xs font-bold text-slate-500 tracking-[0.15em]">HOW TV LIFTS OTHER CHANNELS (WARC ANALYSIS)</p>
  </div>
  <div class="px-5 py-5">
    <div class="grid grid-cols-3 gap-3">
      <div class="bg-slate-50 border border-slate-200 rounded-md p-4 text-center">
        <div class="text-3xl font-bold text-slate-700">+8%</div>
        <div class="text-xs text-slate-500 mt-2">generic search effectiveness</div>
      </div>
      <div class="bg-slate-50 border border-slate-200 rounded-md p-4 text-center">
        <div class="text-3xl font-bold text-slate-700">+20%</div>
        <div class="text-xs text-slate-500 mt-2">online video effectiveness</div>
      </div>
      <div class="bg-slate-50 border border-slate-200 rounded-md p-4 text-center">
        <div class="text-3xl font-bold text-slate-700">+31%</div>
        <div class="text-xs text-slate-500 mt-2">paid social effectiveness</div>
      </div>
    </div>
    <p class="text-xs text-slate-500 mt-4">Source: WARC cross-media effectiveness analysis. Figures are widely cited by third parties; confirm against the primary before quoting exact numbers.</p>
  </div>
</div>

Vendor research points the same way, and it should be read as directional and labelled as vendor research. An MNTN analysis published via Search Engine Land found paid search conversion rates rose around 2% within 30 days of adding CTV and around 22% by 90 days, while paid social rose around 4% at 30 days and around 9% at 90 days. MNTN sells CTV, so it has an interest in the result. The more useful part of that finding is the shape, not the exact percentage. The lift compounds over time as the brand exposure accumulates, rather than landing all at once.

## Why MMM can see what MTA can't

The mechanism MMM uses to capture synergy is old, standard econometrics, not a proprietary trick. In a marketing mix model you write sales as a function of each channel's spend, then add a product term for the pair of channels you suspect interact. The coefficient on that product term is the synergy estimate. Ashok Charan's Marketing Analytics covers it as textbook interaction effects, the same construction used for promotion-times-display interactions, and Chen and colleagues formalize it for cross-channel effects in their 2023 arXiv framework paper.

The reason this matters for a measurement-literate reader is that it moves synergy out of vendor-claim territory and into established statistics. You are not being asked to believe a marketing story. You are being shown a regression term that has a confidence interval attached to it.

Here is what that looks like as a decision. An Ekimetrics example, from an MMM consultancy so read it as an illustrative modeled case rather than a universal constant, shows search returning around €3.20 per euro on its own. Run TV concurrently, and in the model search's effective return rises to around €4.10 per euro. That extra 28% is a synergy bonus attributable to the interaction, not to search itself. The point is not the exact figure. The point is that the same search spend has two different measured returns depending on whether the model accounts for the TV interaction, and that difference changes where the profitable search ceiling sits.

<div class="not-prose my-8 bg-white border border-slate-200 rounded-lg overflow-hidden">
  <div class="px-5 py-3 border-b border-slate-200 bg-slate-50">
    <p class="text-xs font-bold text-slate-500 tracking-[0.15em]">THE ALLOCATION FLIP (MODELED EXAMPLE)</p>
  </div>
  <div class="px-5 py-5 space-y-3">
    <p class="text-sm text-slate-700"><strong class="text-slate-900">Same search spend, measured two ways:</strong></p>
    <div class="grid grid-cols-2 gap-3 mt-4">
      <div class="bg-slate-50 border border-slate-200 rounded-md p-3">
        <div class="text-[10px] font-bold tracking-[0.15em] text-slate-500 mb-1">SEARCH READ STANDALONE</div>
        <div class="text-3xl font-bold text-slate-700 mb-2">€3.20</div>
        <div class="text-xs text-slate-500">return per euro, ignoring the TV interaction</div>
      </div>
      <div class="bg-emerald-50 border border-emerald-200 rounded-md p-3">
        <div class="text-[10px] font-bold tracking-[0.15em] text-emerald-600 mb-1">SEARCH WITH TV RUNNING</div>
        <div class="text-3xl font-bold text-emerald-700 mb-2">€4.10</div>
        <div class="text-xs text-emerald-700">a <strong>28%</strong> synergy bonus that belongs to the interaction, not to search</div>
      </div>
    </div>
    <p class="text-sm text-slate-700 mt-2">On the standalone read you cap search where marginal return hits your floor. Account for the TV interaction and the profitable ceiling moves up, and TV earns credit for lift that shows up in the search line. Two different optimal budgets from one dataset.</p>
    <p class="text-xs text-slate-500">Source: Ekimetrics illustrative MMM example. An MMM consultancy, so read as a modeled case, not a constant.</p>
  </div>
</div>

## Three ways the credit goes to the wrong place

### Search steals the credit CTV earned

A direct-to-consumer brand runs CTV for the first time. Before CTV, its branded paid search converts at 6% and delivers around 1,000 conversions a month. During CTV, branded search converts at around 7.3%, roughly the 22% relative lift the MNTN shape suggests, and volume rises too because more people now search the brand name. MTA sees more branded-search conversions and credits the search channel.

<div class="bg-slate-50 border-l-4 border-slate-400 p-5 my-8 rounded-r-md not-prose">
  <p class="text-xs font-bold text-slate-500 tracking-[0.15em] mb-3">A WORKED EXAMPLE &mdash; THE REALLOCATION ERROR</p>
  <p class="text-sm text-slate-700 leading-relaxed mb-3">The marketer reads the dashboard as "search is crushing it" and shifts budget out of CTV and into search. CTV gets cut. Branded search quietly falls back toward 6% and lower volume, because the thing feeding it is gone.</p>
  <p class="text-sm text-slate-700 leading-relaxed">MTA had no row for "CTV made search convert better," so it recommended defunding the actual cause. An MMM interaction term, CTV spend times search spend, would have flagged the search lift as CTV-dependent and stopped the cut.</p>
</div>

### The invisible upstream touch in B2B

This one is closest to home for a software buyer. A SaaS company runs a LinkedIn thought-leadership push alongside Google Ads. A buyer sees three LinkedIn posts across two weeks. Those are impressions with no click and nothing in the server log. Then the buyer searches the category, clicks a Google ad, and books a demo.

Every attribution model in the tool sees a journey that begins at the Google Ads click. First-touch, last-touch, Markov, Shapley: each one distributes 100% of credit across the observed touches, and the LinkedIn brand exposure is not one of them. The models will disagree with each other about how to split the Google, organic, and email touches, and that spread is exactly what a multi-model tool exists to show. But none of them can manufacture credit for a touch that produced no event.

<div class="not-prose my-8 flex flex-wrap items-stretch gap-2">
  <div class="flex-1 min-w-[130px] bg-pink-50 border border-pink-200 rounded-lg p-3">
    <div class="text-[10px] font-bold tracking-[0.15em] text-pink-500 mb-1">WEEKS 1&ndash;2</div>
    <div class="text-sm font-semibold text-slate-900">3 LinkedIn posts</div>
    <div class="text-xs text-slate-500 mt-1">Impressions, no click. No server event. No row in the log.</div>
  </div>
  <div class="flex items-center text-slate-300 text-xl">&rarr;</div>
  <div class="flex-1 min-w-[130px] bg-indigo-50 border border-indigo-200 rounded-lg p-3">
    <div class="text-[10px] font-bold tracking-[0.15em] text-indigo-500 mb-1">LATER</div>
    <div class="text-sm font-semibold text-slate-900">Category search</div>
    <div class="text-xs text-slate-500 mt-1">Clicks a Google ad. First recorded touch.</div>
  </div>
  <div class="flex items-center text-slate-300 text-xl">&rarr;</div>
  <div class="flex-1 min-w-[130px] bg-emerald-50 border border-emerald-200 rounded-lg p-3">
    <div class="text-[10px] font-bold tracking-[0.15em] text-emerald-600 mb-1">CONVERSION</div>
    <div class="text-sm font-semibold text-slate-900">Books a demo</div>
    <div class="text-xs text-slate-500 mt-1">MTA splits 100% across Google, organic, email. LinkedIn earns nothing.</div>
  </div>
</div>

This is the honest boundary. Model choice changes how observed credit is split. It does not recover an unobserved interaction. Recovering the interaction is an MMM job.

## Where this argument has to stay honest

The reason this article can make its case to a skeptical reader is that it doesn't overclaim on either side. Four caveats keep it honest.

First, MTA does not measure incrementality, and nothing here should imply it does. MTA splits observed credit. It cannot tell you what would have happened without a channel. Synergy and incrementality both require a counterfactual that MTA does not have.

Second, MMM estimates synergy, it doesn't observe it. An interaction coefficient is a modeled inference from correlated aggregate spend. It's vulnerable to collinearity, because channels that always run together can't be cleanly separated. It's sensitive to the priors and regularization you choose and to how you specify the model. The €4.10 number is not a fact of physics. Reading it that way would be the same sin the whole argument accuses MTA of. "MMM captures synergy" should always read "MMM can estimate synergy, with confidence intervals, if the data supports it."

Third, the size of the effect is genuinely contested by serious people. Professor Byron Sharp of the Ehrenberg-Bass Institute has publicly called multi-channel amplification "a bit of a myth" and argues channels behave far more additively than the synergy literature claims. Felipe Thomaz of Saïd Business School at Oxford disputes him, saying the interaction of channels has been established in the literature for a very long time. A measurement-literate reader may already hold Sharp's view, and that's fine, because the measurement point survives either way. Even if synergy is smaller and additive-leaning, MTA still can't see cross-channel effects of any size, additive or multiplicative, because they don't live in the touch log. The size of synergy is contested. MTA's blindness to it is not.

<div class="not-prose my-8 bg-white border border-slate-200 rounded-lg overflow-hidden">
  <div class="px-5 py-3 border-b border-slate-200 bg-slate-50">
    <p class="text-xs font-bold text-slate-500 tracking-[0.15em]">THE OPEN DISAGREEMENT</p>
  </div>
  <div class="px-5 py-5">
    <div class="grid grid-cols-2 gap-3">
      <div class="bg-amber-50 border border-amber-200 rounded-md p-4">
        <div class="text-[10px] font-bold tracking-[0.15em] text-amber-700 mb-2">SHARP (EHRENBERG-BASS)</div>
        <p class="text-sm text-slate-700 leading-relaxed">Multi-channel amplification is "a bit of a myth." Channels behave closer to additively than the synergy literature claims.</p>
      </div>
      <div class="bg-slate-50 border border-slate-200 rounded-md p-4">
        <div class="text-[10px] font-bold tracking-[0.15em] text-slate-500 mb-2">THOMAZ (OXFORD SAID)</div>
        <p class="text-sm text-slate-700 leading-relaxed">The interaction of channels has been established in the published literature for a very long time.</p>
      </div>
    </div>
    <p class="text-sm text-slate-700 mt-4">The point that holds under both views: whatever the size, cross-channel effects don't live in the touch log, so MTA can't see them. That's a measurement fact, not a marketing opinion.</p>
  </div>
</div>

Fourth, a lot of the halo numbers come from interested parties. WARC is credible. MNTN sells CTV and Ekimetrics sells MMM, so both are measuring methods they profit from. That's why the disinterested academic anchor, Veverka and Holy, carries the load-bearing claim that upstream exposure moves downstream behavior. Everything else is corroboration around that anchor.

## Which tool for which question

The practical close is not "buy a synergy tool," because MTA is not one and honest MMM won't pretend its estimate is a measurement. The practical close is to know which question you're asking and pick the tool that can answer it.

| Question you're actually asking | Tool that answers it | What it gives you |
|---|---|---|
| Within this converting journey, how should observed credit split across touches? | MTA | The credit split, and the spread between models on that split |
| Did channel A make channel B convert better? | MMM interaction term | A synergy estimate with a confidence interval, if the data supports it |
| Did this channel cause net-new conversions that wouldn't have happened? | Geo-lift / holdout | Incrementality, the causal counterfactual neither of the above gives you |

Use MTA for the tactical, journey-level credit split and channel diagnostics. Use MMM for the strategic allocation and interaction question at the total-spend level. Triangulate the two, and where the stakes are high enough, validate with a geo-lift test. Don't expect either tool to do the other's job, and be suspicious of any vendor that says one tool does both.

An MTA tool's honest role here is narrow and worth stating plainly. It shows you the spread between models on the credit it can see, and it's explicit that cross-channel synergy sits outside what any MTA, including the good ones, can measure. That candor is the point. Pretending otherwise would be the "trust our number" move that the whole problem started with.

## Further reading

- [MTA vs MMM](/articles/mta-vs-mmm): the two methods side by side, and where each one is strong
- [Triangulating MTA, MMM and incrementality](/articles/mta-mmm-incrementality-triangulation): how to reconcile three measurement reads without picking a favourite
- [What is multi-touch attribution?](/articles/what-is-multi-touch-attribution): the credit-split question MTA is built for
- [Budget reallocation from attribution](/articles/budget-reallocation-attribution): how a borrowed-credit read can quietly defund the channel doing the sourcing

## Key takeaways

- Synergy is an interaction between channels, not a touchpoint, so there is no event in the log for an attribution model to credit.
- MTA distributes a fixed 100% of observed credit; it has no mechanism to make total output grow when two channels run together.
- MMM estimates synergy by adding a product term (channel A spend times channel B spend) to a regression on aggregate sales.
- Upper-funnel and brand spend often shows up as extra credit in the search line, which makes search look better than it is on its own.
- How big synergy is stays contested; that MTA cannot see cross-channel effects of any size is not contested.


## FAQ

**If my MTA shows paid search is my best channel, how do I know that's real and not brand or TV halo feeding branded search?**

Check whether branded search volume and conversion rate move when you turn upper-funnel spend up or down. If search softens whenever brand or CTV pauses, some of search's credit is borrowed from the channel feeding it. An MMM interaction term (upper-funnel spend times search spend) isolates that dependency. MTA can't, because the upstream exposure never produced a click or a server event to credit.

**Can't I just add my TV or LinkedIn impressions into my MTA as touchpoints and let the model handle it?**

No. Impressions with no deterministic link to the converting user aren't real touchpoints, so the model can't place them in a journey. And even if you could, MTA still only splits a fixed 100% of credit across observed touches. It has no interaction mechanism that lets two channels together earn more than they would apart. Adding impressions doesn't grow the pie; it just re-slices it.

**What's the actual difference between 'MTA can't see synergy' and 'MTA can't measure incrementality'? Aren't those the same problem?**

Same root cause, different questions. Both need a counterfactual that MTA doesn't have. Incrementality asks whether a channel caused net-new conversions that wouldn't have happened otherwise. Synergy asks whether channel A made channel B convert better. MTA answers neither. Incrementality is approached with geo-lift and holdout tests; synergy is approached with MMM interaction terms.

**How much synergy is actually there? Is this a 5% effect I can ignore or a 30% effect that flips my budget?**

It's contested and it varies by mix. WARC analysis put TV's lift on search effectiveness around 8%, online video around 20%, and paid social around 31%. A modeled Ekimetrics example showed a 28% synergy bonus on search once TV ran alongside it. Byron Sharp of the Ehrenberg-Bass Institute argues the effect is smaller and channels behave more additively. The honest answer is that you have to estimate it for your own mix, and the estimate comes with error bars.

**If I already run MMM, do I still need MTA, or the other way round?**

They answer different questions, so most measurement-literate teams run both and triangulate. MMM handles the strategic allocation and interaction question at the total-spend level. MTA handles the tactical, journey-level credit split and channel diagnostics. Neither does the other's job well, so don't ask it to.