42% of In-Store Sales Are Created Online
By
Izzy Gregorio
·
13 minute read
Measurement · CPG
42% of In-Store Sales Are Created Online
Channel-level reporting credits that demand to the wrong place, or to nothing at all. Which means a budget allocated on channel ROAS is allocated on a number that was never built to answer the allocation question.
By Izzy Gregorio · Updated August 2026 · 16 min read
In short
Integrated marketing services measure and manage online and offline channels as one system instead of separate reports. It matters because Analytic Partners' ROI Genome analysis finds online channels drive 42% of in-store sales and offline channels drive 40% of online sales. Channel-level reporting credits that demand to the wrong place, or to nothing.
Key takeaways
Five things worth carrying into your next planning meeting
Nielsen's Marketing ROI Blueprint, released October 2025, found 85% of marketers are confident they can measure marketing ROI while only 32% measure it holistically across traditional and digital channels together.
ROI Genome finds integrated campaigns are 31% more effective than non-integrated campaigns.
45% of the sales created by Amazon display advertising happen somewhere other than Amazon, per the same database of more than 1,000 brands.
Attribution, media mix modeling, and incrementality testing answer three different questions. Running one and calling it measurement produces a media plan priced on partial information.
Google Meridian, Meta Robyn, and PyMC-Marketing are open source, which removes the six-figure vendor engagement that used to gate media mix modeling to the largest brands.
The gap
85% of marketers say they can measure ROI. Only 32% actually do.
express confidence in their ability to measure marketing ROI
measure it holistically across traditional and digital together
Nielsen, The Marketing ROI Blueprint, October 9, 2025
Both numbers are true at the same time. That is the problem, stated in full.
A channel report is not broken. It answers exactly the question it was built to answer: what did this platform, this campaign, this keyword appear to produce inside its own reporting window, against its own baseline. It was never designed to tell you whether the sale would have happened anyway, or whether the advertising that created the demand ran somewhere the platform cannot see.
Stack eight of those reports side by side and you have eight correct answers to eight small questions, and no answer to the one your CFO is asking.
The gap does not surface as a measurement complaint. It surfaces as a budget decision.
Channels that report cleanly get funded. Channels whose contribution lands inside somebody else's report get trimmed. Run that for three or four planning cycles and the mix quietly reshapes itself around what is easy to see rather than around what works.
The stakes
What happens to a budget that cannot prove itself
Gartner's 2026 CMO Spend Survey, with 401 respondents and a majority above a billion in revenue, puts marketing budgets at 7.8% of company revenue, roughly 18% below where they sat four years ago. In the same survey, 56% of CMOs say they lack the budget to deliver their 2026 strategy, and 62% say missing growth expectations would trigger cuts.
The misses are already showing. Gartner's 2026 data has 20% of CMOs missing customer acquisition goals, up from 13% the prior year, and 13% missing ROI goals outright. Gartner also reported in February 2026 that more than 40% of CMOs pushing for larger budgets stand to lose C-suite influence if they cannot prove return.
“Every marketing dollar is now under the microscope.”
Marta Cyhan-Bowles, Chief Communications Officer and Head of Global Marketing COE, NIQ, November 2025
For CPG specifically, the money is still large and the growth is not. EMARKETER forecasts roughly $59 billion in US CPG ad spend for 2026, and reports that growth slowed to 4.6% in 2025, falling below the national average for the first time in three years.
Flat budget. Higher bar. A CFO who now reads the marketing line the way she reads cost of goods. Measurement stopped being an analytics project somewhere in there and became a job-security project, and pretending otherwise does not help you.
The shopper
Omnichannel shoppers broke your measurement model
More than 90% of US grocery shoppers now buy both in store and online, per FMI and NIQ. Those shoppers spend about 1.5 times more than single-channel shoppers and are roughly three times more loyal. The customer stopped separating channels years before the reporting did.
| Shopper behavior | Figure | Source |
|---|---|---|
| Buying both in store and online | More than 90% | FMI and NIQ, 2026 |
| Omnichannel spend and loyalty | 1.5x spend, about 3x loyalty | FMI and NIQ, 2026 |
| Monthly grocery spend by behavior | $1,043 both channels, $659 online only, $669 in store only | Grocery Doppio, Incisiv, FMI, 5,100+ shopper study |
| Bad experiences tolerated before churn | 4.2 for dual-channel, 1.3 for digital-only | Grocery Doppio, Incisiv, FMI |
| Self-reporting as omnichannel buyers | 84% | Circana, 52 weeks ending January 1, 2026 |
| Retailers the average household shops annually | 39 | Circana, April 2025 |
| Share of wallet given to a retailer | 31% dual-channel, 26% in-store only | McKinsey, State of Grocery North America 2026 |
Now put two numbers next to each other. Physical stores still account for roughly 80% of grocery-related sales, per NIQ. Online accounted for roughly 75% of total grocery dollar growth in 2025, per FMI and NIQ. Online share moves from 18% in 2024 to a projected 25.5% in 2028, growing at an 11.6% compound rate against 0.6% for in-store.
Stores are where the sales are. Online is where the growth is. A measurement system that keeps them in separate reports cannot see the thing that is actually changing.
And there is no easy exit. McKinsey and NIQ put 2025 US grocery growth at 1.2% in sales, driven by 2.2% in price with volume down 1.0%. Price carried the number and volume went backward. The next point of growth has to be earned through demand, which means through media, which means the measurement has to be right or the money goes to the wrong place.
The headline number
How much of in-store sales are driven by digital advertising?
of in-store sales driven by online channels: search, display, digital video
of online sales driven by offline channels: out-of-home and television
Analytic Partners ROI Genome, distributed through Think with Google. Integrated campaigns in the same database run 31% more effective.
Read the first number again with your own profit and loss in mind. If online media creates something close to 42% of your shelf velocity, and you evaluate online media against online conversions, you are grading it on a fraction of what it produced. Not slightly. Structurally, and in a known direction.
The CPG proof point is sharper still. The ROI Genome database, built across more than 1,000 brands and 25 years of analysis, finds that 45% of the sales created by Amazon display advertising happen off Amazon, along with 23% of the sales created by Amazon sponsored search. Nearly half the value your retail media display budget produces lands somewhere other than the platform writing the report on it.
Search gets misread the same way. The same database finds roughly 30% of paid search clicks are actually driven by other advertising, mostly video. Your most efficient-looking line item is partly a collection point for demand that something upstream created and never got credit for.
There is a documented cost to optimizing around that blind spot. Analytic Partners finds that optimizing only to short-term ROI runs up to 20% less efficient over the long term.
Google and WARC's effectiveness work, as compiled by Coupler.io, puts short-term profit return at £1.87 per pound and £4.11 once sustained effects are counted, places peak effectiveness with brand building at 40 to 60% of total investment, and documents one European retailer losing 44% of its return within two quarters after over-weighting performance. Treat those last figures as directional, since they arrive through a secondary compilation.
The conclusion is not that digital is overrated or that brand is underrated. It is narrower and more useful. A budget allocated on channel-level ROAS is allocated on a number that was never built to answer the allocation question. Attribution and media mix optimization are not a reporting upgrade. They are a repricing of the media plan.
Definition
What are integrated marketing services?
Integrated marketing services plan, run, and measure paid, owned, and earned channels as a single system resolving to one business metric, rather than as separate teams optimizing separate metrics. In practice that means one media plan, one measurement stack, and one number every layer reports against. For a CPG brand, that number is retail sales, not platform conversions.
Most agencies sell integrated as a staffing arrangement. Same account team, more channels, one invoice. That is coordination, and coordination is fine, but it does not change a single budget decision.
The version that changes budget decisions is a measurement arrangement.
One metric that every channel, agency, and retail media network resolves to.
One model that reconciles all closed-loop reporting against actual shipped or scanned revenue.
One test cadence that checks whether the model is describing reality.
One plan built from marginal return, not from last year's plan plus or minus ten percent.
If your integrated program does not change how money moves, it is an org chart, not a system.
Keep going
Measurement analysis with the caveats left in.
What the research actually supports, where vendor data is doing the work, and which findings will not survive the next twelve months. Written for people who have to defend a number in a room.
No spam. Unsubscribe anytime.
The arithmetic
Why platform-reported conversions exceed actual sales
Because every retail media network, platform, and measurement partner counts conversions inside its own attribution window, against its own baseline, with no visibility into any of the others. A brand running eight to twelve closed-loop reporting environments will find that credited sales exceed actual sales. That is not fraud. It is arithmetic.
The money moved fast enough to make this urgent. EMARKETER's December 2025 forecast puts US retail media ad spend at $60.32 billion in 2025, rising to $71.09 billion in 2026, up 17.8% year over year. WARC Media sizes the global market at roughly $200.4 billion in 2026 and projects retail media reaching 55.8% of category media investment for alcoholic drinks and 54.9% for food by 2027. FMI reports 71% of brands actively using retail media networks in 2026.
Concentration compounds it. EMARKETER estimates roughly 89% of incremental 2026 US retail media dollars, about $9.42 billion of $10.53 billion, go to Amazon and Walmart.
The people spending the money already know. Skai and Stratably's 2026 State of Retail Media Measurement found 75% of advertisers naming incrementality as their biggest measurement challenge, while only 15% feel very or extremely effective at measuring it. That is vendor survey data, so treat the exact percentages as directional. The direction is not in dispute.
You have moved the majority of your category's media investment into environments that grade their own homework, and you have no independent scorer.
Three instruments
Attribution, mix modeling, and incrementality answer different questions
These are not competing methodologies. They are three instruments reading three different things, and a media plan needs all three readings.
| Layer | The question it answers | Cadence | Best used for |
|---|---|---|---|
| Multi-touch attribution | Which campaign, creative, or keyword got credit for this conversion? | Daily or weekly | In-channel optimization on trackable digital media |
| Media mix modeling | What is the marginal return on each channel, including offline and untrackable ones? | Quarterly or annual | Budget allocation across the full mix |
| Incrementality testing | Would this sale have happened anyway? | Per test, 4 to 8 weeks | Validating and calibrating the other two |
Adoption is climbing but uneven. EMARKETER and Snap survey data puts US marketers using mix modeling at 53.5%, while Forrester's 2023 survey put business-to-consumer usage closer to 30%. Compilations assembled by Digital Applied across 1,200-plus teams show multi-touch attribution at 47%, up from 31% in 2023, mix modeling at 26%, up from 9%, and only 33% running an explicit hybrid. That sample skews business-to-business, so read it as a trend line rather than a CPG benchmark.
The objection worth taking seriously: media mix modeling is quarterly and your buying decisions are weekly. Correct. That is the argument for running all three rather than picking one.
Attribution moves this week's spend inside a channel. Mix modeling prices next quarter's allocation across channels. Incrementality tells you whether either one is describing something real. A team running attribution alone optimizes with great precision inside a plan that was priced wrong.
What changed
Why mix modeling finally fits a mid-market budget
Media mix modeling used to require a six-figure vendor engagement, which is why it stayed with the largest CPG advertisers for two decades. Google Meridian, Meta Robyn, and PyMC-Marketing are now open source and production-grade, which removes the price floor rather than lowering it.
Meridian matters most for CPG specifically. It is free, it has more than twenty certified measurement partners, and it now supports non-media variables including price and promotion, plus channel-level contribution priors and improved modeling of long-term and upper-funnel effects. Trade promotion is a primary driver of CPG volume. A model that cannot hold it as a variable is not modeling your business.
NIQ is a certified Meridian partner, which connects syndicated CPG sales data directly to open-source modeling. That combination did not exist at this price two years ago.
Two items belong on your radar with a caution attached to each. Google previewed Meridian GeoX in May 2026, an open-source geo experimentation layer designed to feed causal test results straight into the model. It is announced, with testing expected later in 2026, and it is not yet proven.
Separately, EMARKETER has reported that Google reduced its incrementality test minimum from roughly $100,000 to about $5,000. That figure has not been independently audited, so treat it as directional and confirm it against your own representative before you plan around it.
The uncomfortable part
What happens when you test what the dashboards claim
Uncomfortable, usually. In ecommerce datasets where geo-holdout testing is common, measured incremental return typically runs well below platform-reported return, and the gap is widest on branded search and retargeting.
Vendor benchmark compilations from Eightx and Haus put measured incremental return roughly 30 to 60% below platform-reported figures. Haus's public case documentation shows overstatement in the 1.5 to 3 times range. Common Thread Co.'s proprietary geo-holdout database puts median incremental return on Google branded search at 0.27 and on Facebook acquisition campaigns at 1.14.
Every one of those numbers comes from a vendor-owned database skewed toward direct-to-consumer and ecommerce. They are directionally useful and they are not CPG norms.
Do not walk into a planning meeting and present 0.27 as your branded search reality. Walk in and say that in datasets where holdout testing is routine, branded search consistently tests far below what the platform reports, and that you intend to find out what yours is.
The closest thing to a CPG read comes from Lifesight case documentation on a major grocery chain: a geo holdout on non-branded paid search returned zero percent sales lift across twelve test markets, and the budget was reallocated to connected television. One case, one chain, vendor-documented. It is not proof of a category rule. It is proof that the test is worth running, because the result was not close.
This is the part of the analysis that makes paid media look worse, which is exactly why it stays in. A measurement partner who only shows you the numbers that flatter the current plan is not measuring anything.
The next one
The next blind spot is already open
Roughly half of consumers now use AI as a primary or preferred product research source, per McKinsey data. Contentsquare puts AI-referred sessions at about 0.2% of retail web traffic. Both cannot be describing the same reality.
They are not. The Digital Bloom's February 2026 analysis of 446,405 visits found 70.6% of AI-referred sessions arrive with no referrer header and get filed as direct traffic. An Attrifast analysis of 41.2 million sessions put 71% of ChatGPT visits landing in analytics as direct. Similarweb finds a 2.5 times lift in likelihood to visit a brand's site after an AI recommendation, arriving mostly through branded search rather than a referral link. These are vendor datasets, so hold the decimals loosely and the pattern firmly.
Google has since added a dedicated AI Assistant channel and medium dimension in Analytics, and Adobe added a conversational AI referrer type. The instrumentation exists now. Most brands have not turned it on.
It is the same failure in a new channel: demand created in one place, credited to another, and underfunded as a result.
Generative Engine Optimization is the work of being the source those answers cite. If you fixed offline-to-digital attribution, you already know how this story ends.
The rollout
How to build integrated measurement in 90 days
You do not need a new technology stack to start. You need one metric, one baseline model, and one honest test.
- 1
Inventory the reporting overlap
Add up every conversion credited by every platform and retail media network for one month. Compare it to actual shipped or scanned revenue. The difference is the size of your problem, expressed in dollars your CFO already understands.
- 2
Pick one metric the whole model resolves to
Retail sales, not platform conversions. Every layer, every agency, every network reports against it. Anything that cannot be tied to it becomes a diagnostic rather than a scorecard.
- 3
Stand up a baseline model
Two to three years of weekly data by geography, with price and promotion included as variables. In CPG, a marketing mix model without trade promotion is malpractice.
- 4
Run one geo holdout on the channel you trust most
Branded search or retargeting. Choose the channel you are most confident about, because a test on a channel you already doubt teaches you nothing you will act on.
- 5
Calibrate the model, then reallocate
The test corrects the model. The model prices the plan. Re-test quarterly, rotating channels.
- 6
Fix the AI attribution gap while you are already in there
Set up a dedicated channel grouping for AI assistants now, so the next channel shift is not invisible to you for two years.
Steps one and two cost an analyst about a week. Step three is where most programs stall, and it stalls for organizational reasons rather than technical ones, because somebody has to own the number.
Common questions
Integrated measurement, answered
What is the difference between media mix modeling and multi-touch attribution?
Multi-touch attribution assigns credit for individual conversions across the trackable digital touchpoints that preceded them, on a daily or weekly cadence. Media mix modeling uses aggregate historical data to estimate the marginal return of every channel, including offline and untrackable media, on a quarterly or annual cadence. Attribution optimizes inside channels. Mix modeling allocates across them.
What is incremental ROAS and how is it different from ROAS?
Reported return divides attributed revenue by ad spend using a platform's own attribution rules. Incremental return measures only the revenue that would not have occurred without the advertising, established through a controlled test such as a geo holdout. In ecommerce datasets where holdout testing is common, measured incremental return typically runs well below platform-reported figures, with the widest gaps on branded search and retargeting.
Is Google Meridian free?
Yes. Meridian is an open-source marketing mix modeling framework available at no license cost, with more than twenty certified measurement partners available for implementation support. It supports non-media variables including price and promotion, which matters for CPG brands where trade promotion drives volume. Meta Robyn and PyMC-Marketing are additional free, production-grade alternatives.
How often should a CPG brand rebuild its marketing mix model?
Refresh the model quarterly with new data and rebuild the specification annually, or sooner if the business changes structurally. A new retail partner, a major pricing change, a distribution shift, or a new channel entering the mix all invalidate prior assumptions. Between rebuilds, calibrate the existing model with incrementality test results rather than waiting for the next full refresh.
What data do you need to run a marketing mix model?
At minimum: two to three years of weekly sales data at the geographic level you plan to model, spend and impressions by channel over the same period, and non-media variables including price, trade promotion, distribution, seasonality, and competitor activity where available. For CPG, syndicated sales data from a provider such as NIQ or Circana usually supplies the sales side.
How do you attribute in-store sales to digital advertising?
Not through click-based attribution, which cannot follow a shopper into a store. You use media mix modeling to estimate each channel's contribution to total sales including in-store, then validate with geo holdout tests that compare sales in exposed markets against controlled markets. Retailer data partnerships and loyalty-card matchbacks add precision but do not replace the causal test.
What is the best attribution model for a CPG brand?
No single model. CPG demand crosses channels that click tracking cannot follow, so the working answer is a three-layer system: multi-touch attribution for in-channel digital optimization, media mix modeling for cross-channel budget allocation including offline, and incrementality testing to calibrate both against reality. Any single-model answer to this question is selling you the model it happens to sell.
Start here
Find out what your reporting is missing
Start with step one. Inventory the reporting overlap for a single month and compare credited conversions to actual revenue. It costs an analyst a day, and it sizes the problem before you spend anything fixing it.
Then close the newest gap. The AI visibility audit scores your brand across six dimensions of generative engine visibility, shows which questions your category is asking ChatGPT, Claude, Gemini, and Perplexity, and identifies which brand gets cited when they ask. You get a scored baseline before anyone touches your site.