Draw a customer data stack on a whiteboard and it comes out as three boxes joined by arrows: something that captures events, a CDP and a data warehouse. Buying committees that draw it argue about which box sits at the center. The question that decides more is which box gets the event first. It sounds like plumbing. The answer decides what the other two boxes ever see. No later layer can put back a consent choice nobody recorded at the click, or an event the browser dropped. Capture sets the ceiling.

Customer Context Infrastructure is the layer built for that first write: each click, page view or app event is captured, checked against consent and tied to a person inside the company's own cloud. The CDP keeps the durable profile and the audiences built on it, and the warehouse keeps the history that analysts and models read. Most CDP Institute members now run a CDP, and a falling share of their owners say the platform pays off. Settle the first job first. Every later layer inherits its result.

This page collects the primary-source statistics on each layer, from the CDP Institute's member research and Scott Brinker's martech surveys to Google's and Apple's documentation and MetaRouter's customer results. It was last updated on September 30, 2026.

Key statistics on CDPs, data warehouses and Customer Context Infrastructure

Among CDP Institute members, CDPs are the norm, their payoff is uneven, and several of the problems behind it begin before the CDP receives an event.

• 68% of respondents report a deployed CDP in their organization, in a survey of CDP Institute members (CDP Institute, 2024).
• The share of deployed CDPs returning significant value is 64% and falling (CDP Institute, 2024).
• Only 17% of marketers report high utilization of their CDP (Gartner's Lizzy Foo Kune, via AdExchanger, 2024).
• As the named center of the martech stack among B2C and hybrid companies, CDPs fell from 26.9% to 17.4% between 2024 and 2025 (Chiefmartec, 2025).
• 55% of data practitioners spend time maintaining or organizing data sets, the most common use of their time (dbt Labs, 2024).
• In 2025, automated traffic made up more than 53% of all web traffic, up from 51% (Imperva, 2026).
• Allegro increased usable advertising data by 30% after moving collection server-side in its own Google Cloud (MetaRouter customer).
• At a Fortune 50 retailer using MetaRouter, data variability across third-party tools fell from an average of 35% to 1% (MetaRouter customer).

Several lines on this list describe a shortfall or a cost. The two that describe gains both come from a change made at capture.

What is Customer Context Infrastructure?

Customer Context Infrastructure (CCI) is the layer that captures a customer's behavior, identity and consent at the first write, inside the company's own cloud, before a CDP, a warehouse or an ad platform receives the event. MetaRouter calls its category Customer Context Infrastructure and describes it in a launch post by CEO Nikhil Raj. The CDP Institute's definition of a CDP covers the job that comes next.

• A CDP is software that creates and maintains "a persistent, unified customer record" that other systems can reach (CDP Institute).
• In the Institute's FAQ, a CDP "ingests data from all of an organization's sources" (CDP Institute).
• Identity belongs to the CDP too: it "assumes primary responsibility for defining and maintaining customer identity and customer record structure over time" (CDP Institute).
• Three architectures now deliver CDP services: packaged, warehouse-native ("composable") and dual-mode (CDP Institute).
• Where a CDP "stores and pools customer profiles," Customer Context Infrastructure "runs earlier" and "never stores your data" (MetaRouter FAQ).
• Running "inside your own cloud," MetaRouter's Customer Context Infrastructure connects identities across touch points "into one ID graph that you own" (MetaRouter launch post).

The Institute's definition starts from intake and puts identity over time with the CDP. CCI puts identity at the moment of capture, so the profile starts from an event already tied to the right customer. The CDP can be only as sure of a person as that first record allows. When both layers try to do the first job, the company runs two collectors and gets two answers to who the customer is.

CDP adoption and value statistics

Among CDP Institute members, owning a CDP is now the usual case. The open question is whether it pays off, and the Institute's own survey shows the share of CDPs returning significant value falling.

• 68% of respondents reported a deployed CDP (CDP Institute, 2024).
• A unified customer database is reported by 57% of respondents, the first time more than half have said so (CDP Institute, 2024).
• Only 64% of deployed CDPs are returning significant value, and that share has been falling over time (CDP Institute, 2024).
• High CDP utilization is reported by only 17% of marketers (Gartner's Lizzy Foo Kune, via AdExchanger, 2024).
• On Gartner's Hype Cycle, CDPs sit "squarely in the trough of disillusionment," in Foo Kune's words (AdExchanger, 2024).
• 54.9% of companies report a gap between their hopes for a martech payoff and the results (The CMO Survey, Fall 2024).

About one deployed CDP in three is not returning significant value, by its owners' account. Input quality is the part an architect can fix. A CDP can only work with what the first mile delivers: events dropped in the browser, stitched to the wrong person or sent without a consent state never produce an error it can see. The platform takes the blame for a ceiling that was set upstream.

CDP vs data warehouse at the center of the stack

Chiefmartec asked martech and marketing operations leaders which platform sits at the center of their stack, in 2024 and again in 2025. Among B2C and hybrid companies, the CDP lost ground to the warehouse on one side and to engagement platforms on the other:

• Among B2B companies, "Other," which includes custom-built martech platforms, rose from 2% to 10% as the named center of the stack (Chiefmartec, 2025).
• CDPs "appear to have been squeezed between two layers of the stack," the data layer and the engagement layer (Chiefmartec, 2025).
• In Scott Brinker's reading, CDP capabilities "seem to be moving to either the warehouse layer ('composable CDPs') or the engagement layer" (Chiefmartec, 2025).
• Packaged CDPs are still more common than warehouse-native (composable) approaches (CDP Institute, 2024).

The CDP's jobs are being split across layers: history and composable profiles move to the warehouse, while audiences stay in a CDP or move into engagement platforms. The warehouse, the CDP and the engagement platform were each built around what happens after an event arrives. Unless someone assigns the first write, it goes to whichever vendor's tag loads first. Wherever the profile ends up, it starts from the record capture wrote.

Browser and bot limits on client-side data collection

Every layer downstream receives only what the browser lets a script keep, and it takes a bot's events as readily as a shopper's. Each script also adds to page load, and MetaRouter's Fortune 50 case study names third-party JavaScript as a key contributor to page latency.

• Safari's Intelligent Tracking Prevention deletes cookies created in JavaScript and other script-writeable storage after 7 days without user interaction (WebKit).
• In 2025, automated traffic accounted for more than 53% of all web traffic, up from 51% the year before (Imperva, 2026).
• 37% of all internet traffic came from malicious bots in 2024 (Imperva Bad Bot Report, 2025).
• A 0.1-second improvement in mobile site speed lifted retail conversions by 8.4% (Deloitte, 2020, commissioned by Google).
• After the same 0.1-second improvement, retail average order value rose 9.2% (Deloitte, 2020).

A bot session that reaches the CDP can become a profile and an audience member, and a Safari shopper who stays away for more than a week can come back as a stranger. Both are decided when the event is first written, and nothing downstream flags them. Every tag moved off the page also gives back load time, which the Deloitte figures count in conversions.

Customer Context Infrastructure results from MetaRouter customers

MetaRouter's published customer results show what changes when collection moves server-side into the company's own cloud and every other layer reads from one governed stream. In each customer case the ceiling was raised at capture.

• Allegro, one of Central and Eastern Europe's largest online marketplaces, increased usable advertising data by 30% after deploying MetaRouter (MetaRouter customer).
• Conversions across paid campaigns rose by up to 9.7% at Allegro (MetaRouter customer).
• Allegro recovered more than $1 million in annual advertising spend with MetaRouter and Google Cloud (MetaRouter customer).
• Inside its own Google Cloud environment, Allegro runs MetaRouter and feeds the stream into BigQuery, Vertex AI and Gemini (MetaRouter customer).
• A Fortune 50 retailer moved most third-party tags off the page in a private Google Cloud deployment and improved page load time by about 700ms (MetaRouter customer).
• Across its third-party tools, data variability at the same retailer fell from an average of 35% to 1% (MetaRouter customer).
• Event volume and storage fall by 20% to 30% when an Adobe implementation collects only the events its use cases need, according to MetaRouter's own modeling of Adobe implementations, a figure MetaRouter describes as unaudited (MetaRouter estimate).

Allegro's 30% means its ad platforms had been bidding on a smaller sample than its real traffic, and no CDP or warehouse downstream could restore the missing events. The Fortune 50 variability figure means the same event had been counted differently in each tool. Both were fixed at collection, and every later layer read one version of each event.

Consent and privacy rules for customer data collection

Consent decides what an event may be used for and which tools may receive it. In the EEA, ad personalization needs the person's consent, and in California people must be able to opt out of targeted ads.

• Since March 2024, ad personalization features require consent from end users in the EEA (Google Campaign Manager 360 Help).
• Advertisers who don't send consent signals lose ads personalization capabilities (Google Ads Help).
• Whatever the technique, advertising tracking based on browser or device information must rely on the user's informed choice (CNIL).
• Administrative fines under the CCPA run up to $7,988 for each intentional violation from January 1, 2025 (California Privacy Protection Agency, 2024).
• In July 2025, $1.55 million from Healthline became the largest CCPA settlement at the time, after the company failed to let consumers opt out of targeted advertising (California Attorney General, 2025).
• Server-side tagging needs one HTTP request per event from the browser, with the server dispatching the vendor-specific requests (Google Tag Manager Help).
• Before data reaches marketing partners, teams can remove personally identifiable information in the server container (Google Tag Manager Help).

When consent is applied once, before the event fans out, each destination receives only what the person allowed and an auditor has one place to look. When each tool applies consent on its own, every vendor added to the stack adds another consent implementation that can drift from the rest. The consent state captured at the click is the one every destination inherits, so that is where it has to be right.

Data warehouse and data quality statistics

A warehouse stores what it is given, and it never sees the consent banner or the session that produced an event. The data team repairs identity and event definitions after the fact.

• 55% of data practitioners report spending time maintaining or organizing data sets, the most common use of their time (dbt Labs, 2024).
• Platform and infrastructure maintenance takes time from 26% of practitioners (dbt Labs, 2024).
• 69% of enterprises cite preparing data for analysis as the most time-consuming part of the analytics process (ISG Research, 2024).
• Data teams spend more than half their time on repetitive or manual tasks, say 64% of organizations surveyed (Matillion).
• Over a quarter of organizations estimate losses of more than $5 million a year from poor data quality (IBM Institute for Business Value).
• Data accuracy or bias is a leading barrier to scaling AI for 45% of business leaders (IBM Institute for Business Value).
• Among 500 US AI professionals, 81% report significant data quality issues at their company (Qlik).
• At least 30% of generative AI projects were predicted to be abandoned after proof of concept by the end of 2025, with poor data quality among the causes (Gartner prediction, 2024).

A warehouse-native CDP reads those same tables, so it waits on the same cleanup. Part of that maintenance time goes to events that could have arrived complete. MetaRouter's Databricks integration, for example, writes events into Delta tables with identity resolved and consent already applied, so the warehouse and any composable CDP reading from it start from the same governed record. Cleanup repairs definitions and identity after the fact, and an event or consent state the first write never recorded stays missing.

Check the first write in your customer data stack this quarter

Follow one event through the stack to find where its ceiling sits.

Take one high-value event, such as checkout start, and pull one week of it from every place it lands: the CDP, the warehouse and each ad platform. Compare three numbers per destination: the event count, the share of events carrying a consent state, and the share carrying a known customer ID. With bot traffic filtered the same way in every destination, most of the remaining gap should be events from shoppers who declined consent or opted out. A larger gap means events are being lost or rewritten between layers, and a destination applying consent itself means two layers are doing the same job. Either finding marks where to draw the boundary and name one owner.

MetaRouter is the customer context infrastructure that captures identity, behavior, and consent in real time, inside your own cloud, before the moment is lost. The platform runs in the customer's own AWS, Google Cloud or Azure environment and feeds the CDP and warehouse already in place, so every layer after it starts from the same record.