Final month, a knowledge crew and a advertising and marketing crew sat in the identical room and talked previous one another for 45 minutes.
The engineers spoke in Delta tables and medallion structure. The entrepreneurs spoke in journeys, segments, and ship time optimization. Identical buyer. Identical information. Two utterly totally different languages. No person was unsuitable. They only could not hear one another.
That useless area — between the place information lives and the place it prompts — is the place tens of millions of {dollars} in buyer worth go unrecovered each week, at corporations with world-class information infrastructure and world-class advertising and marketing ambitions. They’ve the rocket ship. They forgot the launchpad.
Scott Brinker’s analysis report, The New Martech “Stack” for the AI Age, revealed in partnership with Databricks, places the proper title on what wants to exchange the structure that created this drawback. He calls it the composable canvas. The analysis is precisely proper, and the course it factors is precisely the place this partnership is constructed to go.
What the composable canvas really means for entrepreneurs
When you’ve ever exported a CSV out of your information warehouse, emailed it to the marketing campaign crew, and waited three days for it to get uploaded into your advertising and marketing platform — that is the outdated stack mannequin failing you.
For 20 years, martech was inbuilt inflexible vertical layers: information on the backside, methods of engagement within the center, campaigns on the high. Every layer was its personal field. Getting these bins to speak to one another required pipelines, connectors, sync jobs, and groups of individuals whose whole job was transferring information from one system to a different. As Brinker’s report notes, integration remained a top-three problem for almost all of selling organizations surveyed — not in 2015, however in late 2025. After thirty years of distributors promising to resolve it.
The composable canvas is the architectural different: a unified information basis the place each instrument, out of your buyer engagement platforms, to your AI brokers, to your analytics, operates on the identical shared substrate, with out information ever having to maneuver. No middleware. No lag. No Sunday evening spreadsheet ritual.
Bryce Peake, former VP of Advertising and marketing Choice Sciences at Domino’s, places it plainly:
We have been making an attempt to change the identical martech stack we have had because the web began interneting. People, we’ll must construct a brand new one.
Trendy information platforms like Databricks present an open, shared basis that entrepreneurs, brokers, and apps can all function on collectively. And the enterprise case is concrete: “Pace: From months to minutes. Immediately, including a brand new advertising and marketing functionality means integration initiatives, information pipelines, and IT tickets. Within the composable canvas, new instruments and AI brokers plug right into a shared information basis immediately.”
The structure: 5 rings, one middle of gravity
Brinker’s framework organizes the composable canvas into 5 concentric rings, every with a definite function.
- Knowledge Core: the unified basis of buyer, firm, content material, code, and management information; the gravitational middle of every little thing
- Semantic Layer: shared definitions that make information constant and significant throughout each system that touches it
- CaaS (Context-as-a-Service): platforms like CDPs that bundle related information and context for the apps and brokers that want it
- Decisioning: the place AI engines optimize next-best actions and resolve competition when a number of brokers wish to attain the identical buyer concurrently
- Apps & Brokers: the outermost ring, the place buyer experiences really get delivered
The payoff of this construction is a dramatic discount in integration complexity. Within the outdated point-to-point mannequin, ten methods may require as much as 45 integrations. Within the composable mannequin, every new functionality joins a coherent shared ecosystem reasonably than an online of brittle pipelines. As Rick Schultz, CMO of Databricks, places it:
All CMOs try to resolve the identical three issues: effectiveness, effectivity, and self-service.
The composable canvas is how all three change into achievable concurrently.
Elizabeth Dobbs, AVP of Advertising and marketing Expertise at Databricks, describes the affect from firsthand expertise:
As soon as all of our advertising and marketing information was centralized within the lakehouse, we revisited fashions that had by no means totally leveraged that richness… The rebuilt mannequin transformed at 4X the earlier fee. From there, we shortly replicated the identical strategy to launch a extremely predictive account scoring mannequin in simply 4 weeks.
The final mile: the hole between your first-party information platform and a stay marketing campaign
Brinker is express that this report is a North Star, not a step-by-step implementation information, and that is precisely what makes it precious. What it deliberately opens up is the practitioner query: how does a unified information basis really change into a triggered marketing campaign? How does a decisioning layer get constructed for a advertising and marketing crew that does not communicate information engineering? How do the 5 rings transfer from an structure diagram to a buyer expertise that fires on the proper second?
These questions are the place the true work of implementation lives — what Sew calls the final mile.
Here is what the hole between the place to begin and the end line of the final mile really seems like in follow:
- The autonomous brokers with nowhere to go. A advertising and marketing chief at a significant model lately described a maddening state of affairs: her engineering crew had simply deployed an autonomous agent system on Databricks that might cause throughout their whole buyer information set in actual time. But it surely had no concept learn how to set off a marketing campaign. In the meantime, her advertising and marketing crew’s lifecycle journeys have been nonetheless being fed by a batch file up to date as soon as a day. Two gifted groups constructing the long run, with no visibility into one another’s work.
- The propensity rating no one acted on. A gaming model’s information science crew constructed a classy churn mannequin. The propensity rating fired appropriately, flagging clients in danger. It went right into a dashboard no one was checking. The client felt nothing from the model in the mean time they wanted to really feel one thing… and left. As a result of the final mile by no means bought constructed.
- The Sunday evening spreadsheet. A nationwide grocery retailer had a Databricks surroundings filled with provide eligibility information and a buyer engagement platform constructed for precisely this sort of personalization. The hole between them? 4 hours, each Sunday evening, and a CSV. Marketing campaign prep time: hours. The information was there. The activation functionality was there. No person had constructed the bridge. As soon as related, with provide eligibility flows mechanically into buyer profiles and campaigns, triggering in actual time. These 4 hours turned minutes, each week, completely.
As Brinker notes,
The martech stack does not sit on high of information. It’s information.
Architecturally, that is true. Operationally, for many manufacturers in the present day, the info and the marketing campaign nonetheless stay in utterly totally different worlds.
What closing the final mile requires
The manufacturers making the composable canvas actual aren’t doing it by shopping for higher instruments. They’re doing it by constructing the bridge — and that requires deep fluency in each the info platform and the advertising and marketing execution layer concurrently.
Advertising and marketing information structure constructed for activation. There’s a significant distinction between a knowledge basis constructed for analysts and one constructed for entrepreneurs. The previous is optimized for queries, insights, and governance. The latter is optimized to set off a marketing campaign on the proper second, enrich a buyer profile in actual time, and supply an AI agent with the context it must decide. Getting there requires understanding each what Databricks can do and what the engagement layer really wants from it.
Self-service analytics for advertising and marketing groups. Brinker describes pure language interfaces as a core function of the composable mannequin — and for good cause. That is achievable, however most advertising and marketing groups are nonetheless submitting tickets to get primary marketing campaign efficiency information.
The chance: construct analytics environments scoped to actual advertising and marketing use circumstances, so non-technical groups can reply their very own questions with out SQL, and not using a queue, and with out ready on information engineering. It is one of many highest-leverage implementations out there proper now.
AI brokers that run natively on the info layer. Essentially the most highly effective agentic advertising and marketing workflows aren’t brokers bolted on high of the present stack. They’re brokers that stay inside the info basis and execute by the engagement layer — pulling segments, validating information high quality, triggering campaigns, and measuring outcomes with out the CSV, with out the guide handoff, with out the lag. Kumar Ram, VP and World Head of Advertising and marketing Knowledge Sciences at HP, frames the strategic precept exactly:
We wish to personal the core assemble of the info and infrastructure. If we modify businesses, all we have to do is flip the activation layer on the high. The muse stays with us.
Migrations that construct towards composability, not away from it. For manufacturers nonetheless operating on inflexible legacy advertising and marketing platforms, the transfer to a composable structure isn’t merely a platform swap; it is an architectural resolution. Completed proper, it positions Databricks as the info layer beneath an engagement stack that may flex and evolve as AI capabilities change. Elizabeth Dobbs captures the objective:
We have designed our stack like modular Lego blocks so the enterprise can transfer quicker. As new applied sciences and AI brokers emerge, we’ve got the flexibleness to undertake the perfect instruments with minimal elevate.
3 ways to begin constructing towards the composable canvas in the present day
Brinker is obvious that it is a 3-5 12 months architectural journey, not a rip-and-replace mission. The excellent news: each step delivers worth by itself phrases. Listed below are three beginning factors that constantly unlock essentially the most speedy affect.
1. Get your advertising and marketing information into Databricks — really into it.
The commonest model of “we’ve got Databricks” is: the info engineering crew has Databricks, and advertising and marketing has a dashboard that often displays it. That is not a unified information basis. That is a reporting layer.
The primary transfer is consolidation: getting marketing campaign efficiency information, buyer behavioral alerts, loyalty information, and engagement historical past flowing into the identical lakehouse the place your transactional and operational information already lives. That is plumbing work — connectors, pipelines, information high quality checks — nevertheless it’s the muse every little thing else will depend on.
The speedy payoff is unified reporting: one model of reality that advertising and marketing, gross sales, and finance are all . As Brinker’s report places it:
When everyone seems to be trying on the similar numbers, calculated the identical method, conferences shift from debating information to deciding what to do about it.
That alone is well worth the funding. But it surely additionally ends the info tax that slows each marketing campaign down — the tickets, the exports, the ready.
The place to begin: Map the place your advertising and marketing information at present lives versus the place it must be. Establish the 2 or three information sources that may most change what advertising and marketing can do in the event that they have been unified — sometimes transaction historical past, behavioral occasions, and provide or loyalty information — and construct towards these first.
2. Construct one self-service analytics use case in your advertising and marketing crew
Most advertising and marketing groups are sitting one query away from an perception that may change how they run their campaigns. The issue is not the info. It is the friction between the marketer and the reply.
Brinker’s composable canvas places self-service analytics — particularly, pure language interfaces that allow non-technical customers question information straight — on the middle of the mannequin. Instruments like Databricks AI/BI Genie make this achievable in the present day: a marketer sorts a query in plain English and will get a solution in seconds, with out SQL, and not using a ticket, and not using a three-day wait.
The hot button is constructing for a particular use case reasonably than making an attempt to offer advertising and marketing entry to every little thing without delay. Begin with the query your crew asks most frequently and takes the longest to reply: marketing campaign conversion by phase, provide redemption by retailer, churn alerts by buyer cohort. Construct the Genie area round that call. Let the crew use it. Watch what occurs when friction disappears.
One model’s marketing campaign supervisor went from a four-day turnaround on a primary efficiency query to a solution in seconds — after which requested 12 extra questions over the following 20 minutes, surfacing an perception about push notification timing that had been hiding within the information for 1 / 4. The friction was gone, and so was the ceiling on what she may discover.
The place to begin: Establish the only most typical information query your advertising and marketing crew escalates to the info crew. Construct a ruled, use-case-focused analytics surroundings round that query first. Develop from there.
3. Decide one AI agent use case and ship it
The largest mistake advertising and marketing groups make with AI is treating it as a method reasonably than a follow. The composable canvas makes AI brokers genuinely operational — however provided that you begin constructing them on actual information, in actual workflows, towards actual outcomes.
The simplest first agent use circumstances are slender, high-frequency, and related on to a marketing campaign final result. A QA agent that validates each e mail earlier than it sends. A personalization agent that pulls real-time stock or loyalty alerts and generates copy on the fly. A churn agent that screens behavioral alerts in Databricks and triggers a marketing campaign the second a buyer crosses a danger threshold — not three days later when a batch file runs.
What makes these use circumstances work is the structure: brokers that stay within the information layer and execute by the engagement layer: no CSV exports, no guide handoffs, no lag. The marketing campaign is just as clever as the info feeding it — which is precisely why the info basis comes first.
As Brinker notes within the report,
Essentially the most precious agentic workflows cross system boundaries as a matter in fact. Brokers are ravenous for context–relevant information, directions, and instruments for the precise objective they’re fixing for.
The composable canvas is what offers them that context. The agent is simply the factor that acts on it.
The place to begin: Establish one high-frequency, high-cost guide course of in your marketing campaign operations — QA, segmentation, content material technology, efficiency reporting — and scope a single agent to automate it. Measure the time saved and the change in final result. Use that proof level to construct the case for the following one.
Why the window to construct that is now
The timeline on Brinker’s three-to-five-year imaginative and prescient is compressing quicker than most advertising and marketing leaders anticipated.
Databricks simply launched CustomerLake, an Agentic CDP natively embedded inside the Lakehouse. Core CDP capabilities together with Buyer 360, identification decision, viewers constructing, marketing campaign automation, activation, and personalization can occur the place buyer information, AI fashions, and governance already reside — no middleware required.
Buyer engagement platforms like Braze are connecting on to Databricks by native Cloud Knowledge Ingestion and OpenSharing. New platform capabilities like Lakebase imply full-stack advertising and marketing purposes — entrance finish, transactional logic, and marketing campaign execution — can run on a single platform with a single governance mannequin, eliminating the necessity to sew collectively 4 separate methods.
The sample Brinker identifies is already taking part in out: the platforms on every finish are getting smarter and extra straight related. The combination layer within the center is getting thinner each quarter.
The manufacturers doing this work in the present day see the place the market goes. They’re constructing the info structure now so they are not scrambling to catch up when everybody else figures it out.
The composable canvas describes the vacation spot. The final mile is the work between right here and there.
Obtain Scott Brinker’s analysis report: The New Martech “Stack” for the AI Age
Databricks and Sew are partnering to assist advertising and marketing organizations shut the hole between fashionable information infrastructure and actual buyer outcomes. In case your group is sitting on a Databricks funding that it hasn’t totally activated for advertising and marketing, discuss to Sew →
