SignalShift October 2026: The New Economics of Marketing Infrastructure

Salesforce-Braze, ChatGPT ad economics, Campaign Agent, CustomerLake, retail media and AI slop sites: six October signals repricing marketing infrastructure.

TL;DR

In October, the money moved toward marketing's plumbing. Salesforce agreed to pay about $5.66 billion for Braze, Hightouch bought a startup whose agents build integrations, Databricks pushed its warehouse-native CDP into beta, and Salesforce made its Campaign Agent generally available. OpenAI took ChatGPT ads to 89 countries and more than 100,000 advertisers. Then the first serious working paper on ChatGPT ad economics found paid visits earning roughly a third of what paid Google Search visits earn. Meanwhile, media kept fragmenting faster than measurement could keep up, and AI-generated made-for-advertising sites showed how quickly cheap content finds its way into paid media. Put together, the message is this: marketing infrastructure is being repriced around two scarce assets, connected customer context and credible proof of outcomes. Capability is getting cheaper every month. Proof is getting more expensive.

Introduction

In September, I argued that our buyers had started to delegate. OpenAI and Meta shipped always-on agents that research, compare and transact, and I made the case that marketing organizations built to capture human attention now have to earn machine delegation. I ended with three things that decide whether a brand survives that handoff: access, evidence and actionability.

That edition was about the buyer's side of delegation. October was about the seller's side.

If agents are going to act on behalf of buyers, and increasingly on behalf of marketers too, someone has to own the systems they operate through. October is the month the industry started paying serious money to be that someone. Salesforce bet billions on real-time engagement. Hightouch bought its way into agent-built integrations. Databricks walked into the CDP category with a beta and a named enterprise customer. Salesforce, HubSpot, Adobe and TikTok all shipped features that move campaign execution away from humans clicking through interfaces and toward agents working against goals.

And then October did something more useful than another product launch. It gave us a reality check. A working paper put hard numbers on ChatGPT advertising for the first time, and they weren't flattering. Measurement vendors rushed to close the gap between media spend and business outcomes, and one of them promptly published results grading its own launch partner. Researchers traced thousands of AI-generated sites monetized through programmatic ads, some carrying ads from household-name brands.

I'll declare a bias up front. I run a B2B email platform, so a $5.66 billion bet on an engagement layer isn't abstract news for me. But I've also spent close to two decades watching marketing technology promise more than it proves, and October had both of those stories running side by side. That's what this edition is about: not whether the new infrastructure is impressive (it is), but whether we can tell when it's actually working.

Executive Dashboard

#	Shift	What changed in October	Why it matters for marketing leaders
01	The prize is customer context	Salesforce agreed to acquire Braze; Hightouch acquired Strata	Martech value is moving to whoever owns the connective tissue between data, decisions and execution
02	Conversational ads face an economics test	ChatGPT ads reached 89 markets while a working paper found paid visits earn far less than paid search	Conversational media needs its own performance model before it earns search-level budgets
03	Execution becomes agentic	Salesforce Campaign Agent went GA; HubSpot began feeding CRM conversions to ChatGPT Ads	Marketing teams move from operating campaigns to designing the systems that run them
04	The CDP moves toward the warehouse	Databricks moved CustomerLake into beta, with HP as an early enterprise example	Customer intelligence becomes an architecture decision, not a tool purchase
05	Media outruns measurement	Lowe's and TikTok extended their networks off-platform while NIQ, Cint and IAB worked on outcome measurement	The advantage shifts from being everywhere to knowing what actually worked
06	The ad supply chain has a trust problem	A long Google spam update, 2,100+ suspicious AI recipe domains and new consumer data on AI disclosure	Content quality and media integrity become governance issues, not brand-safety checkboxes

Signal 01: The Martech Stack Is Consolidating, but the Real Prize Is Customer Context

For most of the last decade, martech vendors competed by owning a slice of the workflow. One platform held customer records, another sent email, another handled mobile engagement, another unified data, and a long tail of tools handled analytics, attribution and orchestration. The result was an operating environment that nobody designed as a whole, and most of us inherited the job of stitching it together.

October brought two deals that point somewhere else.

On October 20, Salesforce signed a definitive agreement to acquire Braze, the real-time customer engagement platform behind brands like Canva, Instacart and Venmo. The deal values Braze at roughly $5.66 billion in enterprise value, pays $47.50 a share in cash, and is expected to close in Salesforce's fiscal year 2028, subject to shareholder and regulatory approval. That's worth reading carefully: this is an agreement, not a completed acquisition, and as PPC Land worked out from the fiscal calendar, nothing changes for Braze's roughly 2,770 customers until February 2027 at the earliest. Braze's CEO has said the company will keep building as Braze.

Eleven days earlier, on October 9, Hightouch announced it had acquired Strata, which it describes as an agent-driven platform for building and maintaining custom API integrations without code. Hightouch was explicit about why: more connections across the marketing stack give its customers' AI agents the context they need to make decisions.

One deal is about real-time execution. The other is about connective tissue. Both are about the same thing: whoever can connect customer data to decisions to action, fastest and most reliably, owns the next decade of martech.

Why we're calling this a signal

Platform signal. Salesforce isn't buying a feature. It's buying an event-streaming architecture built to turn live customer behaviour into engagement across apps, web, email, SMS and messaging, and positioning it alongside Data 360 and Agentforce. That's a statement that real-time context is now core to the CRM, not an add-on.

Product signal. Strata's pitch is that an agent builds and maintains the integration from a plain-language description. Integrations have always been the part of martech we underfunded and resented. When agents start building them, the cost of connecting systems drops, and the strategic question shifts from "can we connect this?" to "should this agent be allowed to?"

Market signal. As PPC Land's deal coverage pointed out, Braze, Salesforce and Adobe were the three Leaders in Gartner's 2025 Magic Quadrant for Multichannel Marketing Hubs. One of those three is now being absorbed by another. Add Adobe's Semrush acquisition, completed in April, and the pattern is hard to ignore. The suites are buying the capabilities agents will need: real-time signals, visibility intelligence and connectivity.

Continuity signal. Last month, I wrote that Dreamforce and HubSpot UNBOUND turned the CRM from a system of record into a system of action. October is the follow-through. Having repositioned the CRM as something agents act through, the vendors are now buying the parts that let it act in real time and across more systems.

What this means

The standalone tool gets harder to defend. A specialized application can still be excellent, but its position gets fragile when a suite can absorb the capability. The defensible products will be those with differentiated data, deep workflow expertise or capabilities that are genuinely hard to rebuild. I went deeper on this in Salesforce Bought Braze. Adobe Bought Semrush. Is Marketing Technology Consolidating Too Much? Speaking as someone who owns a specialist platform, that's a standard I'd rather hold myself to now than discover later.

Integration stops being implementation work. We've always treated integrations as the plumbing IT sorts out after the strategy is decided. In an agentic environment, integrations are the strategy. An agent can't make a useful decision if it can't see the right customer context, and it can't execute one if it can't safely reach the system that needs to act.

Consolidation doesn't have to mean fewer vendors. Most of us will still run specialized systems. What changes is that they'll need to share context through interoperable data models, APIs and permissions. The goal isn't one platform that does everything. It's less friction between the platforms that matter.

In the stack reviews I do with clients, the expensive tools are rarely the real problem. The problem is the gaps between them: the field that doesn't sync, the event that never reaches the email platform, the lead score that lives in one system and the conversation history in another. October's deals are, in effect, vendors pricing those gaps.

Organizational impact

Martech decisions are becoming architecture decisions, and most marketing organizations still make them as procurement decisions. That needs to change. CMOs, CIOs and revenue operations leaders should be evaluating platforms on where customer identity and context are maintained, how events move between systems, whether agents can access what they need securely, whether actions can be executed across applications, and how easily a component can be swapped out without breaking everything around it.

There's also a governance angle that ties back to September's argument about agent identity. Strata-style integrations need credentials to your marketing systems. Every agent-built connection is another set of permissions that can outlive the project that requested it. If marketing ops doesn't own an inventory of what is connected to what, and with whose credentials, nobody does.

What CMOs should do next

• Map your customer context. Document which systems hold identity, behavioural signals, campaign history and engagement data, and where those systems fail to talk to each other.
• Pressure-test pending roadmaps. If you run Braze or Marketing Cloud, ask both vendors what the integration roadmap looks like during the approval period, and don't sign multi-year commitments based on slides about a combined product that doesn't exist yet.
• Inventory integration credentials. List every connection between your marketing systems, who authorized it and what it can read and write. Agents will multiply these quickly.
• Make architecture a standing review. Set up a quarterly martech architecture review with marketing, IT, data and revenue operations in the same room.

Closing thought: The next martech leader may not be the platform with the most features. It will be the one that understands the customer across systems and can turn that understanding into action without a human stitching the pieces together.

Signal 02: ChatGPT Is Becoming an Advertising Platform. Can It Deliver Search-Level Economics?

For two decades, digital advertising has run on a simple principle: find out what people are looking for, put a relevant message in front of them, and measure what happens next. Google built one of the most profitable businesses in history on that idea. Queries became commercial signals, keywords became inventory, and the closer someone was to a decision, the more that moment was worth.

Conversational AI offers something different. People don't type keywords into ChatGPT. They explain problems, describe constraints, compare options, push back on answers and work their way toward a decision over several turns. That's potentially a richer advertising environment than search. It's also a very different one.

OpenAI is now commercializing it at speed. On October 21, the company expanded ChatGPT Ads to 26 more countries, bringing availability to 89, and said more than 100,000 advertisers had run campaigns since ads launched just over eight months earlier. For context, The Information reported that the figure was around 20,000 advertisers in late August. In September I noted the ads business had already hit a $1 billion annualized run rate, and I unpacked the early trajectory in ChatGPT Hit $100M in Ad Revenue. Here Is What That Means for Every Marketer Right Now. Reuters framed the expansion as part of OpenAI building revenue ahead of a potential IPO.

That's no longer an experiment. It's a global media business under real pressure to grow.

Then came the other half of the story. An October 2026 working paper by Maximilian Kaiser of WU Vienna and Christian Schulze of Frankfurt School of Finance & Management, which reached the trade press in the first days of November, tracked 1,898 online shops through OpenAI's country-by-country ad rollout. As Search Engine Land reported on the findings, in US traffic, a visit from a paid ChatGPT ad generated about $0.89 in revenue, against $2.45 for a visit from a paid Google Search ad. Search Engine Journal's breakdown explains the paper's "parity price" logic, which the researchers used to estimate that a ChatGPT ad click would need to cost roughly $0.26 to match search on return. Conversion rates told the same story: 0.37% for ChatGPT ads versus 1.17% for Google Search ads.

The authors' own summary of the findings goes further. At publicly reported ChatGPT ad prices, each advertising dollar returned about $0.52 in revenue from ChatGPT ads against $3.40 from Google Search ads, which is roughly 6.6 times better value for search. They also found that paid ChatGPT ads produced fewer purchases and less revenue than ChatGPT's own unpaid recommendations.

The caveats matter, and SignalShift doesn't skip them. This is an observational working paper, not a peer-reviewed industry benchmark. The sample is ecommerce shops, not B2B companies with long buying cycles. The paid-advertising comparison draws on a smaller subset of the full sample. And revenue per session can't capture purchases that happen later or influence that shows up weeks down the line.

But even with all of that, the contrast is the point. OpenAI has proved that advertisers will spend money inside conversational AI. It hasn't yet proved that the money comes back at the rate advertisers expect from their existing channels.

Why we're calling this a signal

Market signal. Five-fold advertiser growth in roughly two months, 89 markets and a billion-dollar run rate. Whatever the economics turn out to be, the budget is already moving.

Research signal. For the first time, we have independent, multi-site data comparing ChatGPT ad performance to paid search on revenue, not just clicks and impressions. That's the comparison every CFO is going to ask for.

Behavioral signal. The finding that ChatGPT's unpaid recommendations outperform its paid placements says something important about how people use the product. They trust the answer more than the ad. That fits what I wrote last month about Sponsored Agents: a sponsored conversation has to hold up next to an answer the user already trusts, and right now it mostly doesn't.

Platform signal. The paper cites company figures suggesting free ChatGPT access costs around $5 per user per year while projected ad revenue is closer to $1 per user. That gap creates real pressure on OpenAI to improve ad yield, which means formats, pricing and targeting will keep changing quickly. Early benchmarks will age fast in both directions.

What this means

Conversational intent and search intent aren't interchangeable. Someone searching "best enterprise CRM software" is often already building a shortlist. Someone asking ChatGPT how CRM software works may be months from buying. Both are expressing intent, but not the same readiness. Treating a conversational click as if it carries search-level purchase probability is how budgets get misallocated.

The model has to evolve beyond the click. Search advertising optimized around keywords, bids, clicks and conversions for twenty years. A conversational buyer might ask a question, compare three products, challenge a recommendation and only then decide whether to visit a site. The commercial value may sit somewhere in that sequence rather than at the click. That opens room for contextual sponsorships and interactive product discovery, but those formats will need transparent measurement. More engagement doesn't automatically mean more revenue.

Platform success and advertiser success are two different businesses. OpenAI has every reason to build advertising into its model, since ads fund free access and monetize users who don't subscribe. None of that guarantees attractive acquisition costs for the businesses paying. That tension isn't unique to AI, but it matters more when a new platform is asking us to pull budget from channels with decades of performance history.

My take: the advertising opportunity inside AI conversations may eventually be bigger than search, because a conversation reveals needs, context, constraints and preferences that a keyword never could. But that long-term potential shouldn't be confused with present-day performance. I watched the same pattern play out with early social advertising: reach arrived years before reliable returns did, and the brands that came out ahead were the ones that tested with discipline instead of either ignoring the channel or over-committing to it. I'd treat ChatGPT ads the same way, and it's also why the two companies' very different approaches, which I compared in OpenAI and Google Reveal Competing Visions for AI Ads, deserve a close read before you commit budget.

Organizational impact

Conversational advertising needs to be separated from paid search in planning, measurement and budget allocation. Folding ChatGPT into your search team's reporting as if it were another Google campaign type will hide the differences in intent, interaction and conversion behaviour that the research just surfaced.

Performance teams should build separate benchmarks covering cost per qualified visit, conversion rate by intent category, revenue per session, cost per opportunity and acquisition, assisted conversions and pipeline contribution, and incremental revenue against existing search campaigns.

For B2B, this matters even more. A conversation that helps a buyer understand a complex category might influence a deal weeks before anyone fills in a form. Last-click attribution will undervalue that. But crediting every later conversion to an AI interaction would be just as misleading. The job is to establish credible measurement before this channel absorbs meaningful budget, not after.

What CMOs should do next

• Treat conversational ads as an emerging channel, not a replacement for search, and give them a separate experimental budget until the economics repeat.
• Run controlled tests with defined audiences, budgets and conversion objectives, and compare against equivalent search campaigns while accounting for differences in buying stage.
• Calculate your own parity price. Divide revenue per visit from ChatGPT ads by revenue per visit from paid search, multiply by your search CPC, and compare that number to what you're actually paying.
• Measure pipeline, not just clicks. In B2B, the only numbers that should move budget are opportunity creation and downstream revenue.

Closing thought: OpenAI has shown that conversational AI can become an advertising business. The next milestone is proving it can become an effective advertising channel for the businesses paying for it, and those are two very different milestones.

Signal 03: Marketing Execution Is Becoming Agentic. The Campaign Manager Is Turning Into a System Designer.

Marketing automation was built to make repetitive work easier. We built segments, designed workflows, defined triggers, picked channels, scheduled sends and watched the dashboards. The platform executed instructions, but a human coordinated everything around them.

That operating model started changing in a visible way this month.

Salesforce made its Campaign Agent generally available in Marketing Cloud Next. The idea is that a marketer defines a goal, and the agent assembles the audiences, content, journeys and channels to pursue it, using customer signals and brand guidelines, while the marketer reviews and approves before anything launches. The commercial detail is just as telling: Salesforce says limited Campaign Agent capacity comes with Marketing Cloud Next Advanced, and additional agentic campaigns and content generations cost extra. Salesforce also listed as generally available in October an MCP-powered architecture that lets any LLM work on journeys, flows and messaging from any surface.

HubSpot moved the other end of the loop. Its documentation for syncing ad conversion events with ChatGPT, updated in mid-October, shows how lifecycle-stage changes and form submissions in the CRM can be sent back to ChatGPT Ads, which uses those signals to optimize targeting over time. It's a beta, it requires contact consent, and it only counts record updates made after the conversion event is set up. But the architecture is what matters: downstream customer outcomes now flow back into the system that generated the original interaction.

Adobe added another layer. Its CX Enterprise Coworker documentation added new material on October 16 for creating, simulating and analyzing cross-channel journeys, along with journey analysis, audience analysis and attribution skills that work through conversation. And TikTok, while opening its ad network to US advertisers, also began letting advertisers run campaigns through AI tools like Claude and Perplexity.

These aren't interchangeable products, and I'd be wary of anyone who presents them as such. But the direction is consistent. Platforms are moving from helping with individual tasks to coordinating larger pieces of the workflow, and in two cases, from requiring you to log into their interface at all.

Why we're calling this a signal

Product signal. Salesforce, HubSpot and Adobe each shipped capabilities in the same month that move from instruction-following automation to goal-driven execution. That's three of the largest marketing platforms converging on the same model.

Platform signal. Salesforce's MCP architecture and TikTok's AI-tool access both mean campaign execution can now happen outside the vendor's own UI. Last month I wrote about Salesforce bringing its CRM into Claude and Slack. October extends that to marketing execution itself.

Commercial signal. When a vendor prices "additional agentic campaigns and content generations" separately, it's telling you how it expects value to be measured: by agent output, not by seats. That will reshape martech budgets in ways most planning cycles haven't accounted for yet.

Continuity signal. In September, I argued that search management was becoming input management after Google's AI Max migration. HubSpot's ChatGPT integration is the same idea arriving in conversational media. The platform optimizes; we supply the signals. The quality of those signals becomes the job.

What this means

Campaign architecture becomes dynamic. Traditional journeys are built around expected paths. Agentic campaigns can adapt content, timing, channel and treatment as context changes. That reduces the rigidity of static workflows, and it raises the stakes on governance, because the system is now making choices you didn't explicitly draw.

Marketing operations becomes a system-design function. When agents execute more of the work, someone has to define what they're allowed to do, which data they can touch, what outcomes they optimize for, and where a human has to approve. That's the design and management of a decision system, and it's a very different skill set from building journeys in a canvas.

CRM data becomes an active optimization input. The HubSpot integration points to something B2B marketers have wanted for years: ad platforms optimizing toward qualified leads and sales progression, not just clicks and form fills. The flip side is that platforms will optimize toward whatever you send them. Send them sloppy lifecycle stages and they'll find you more sloppy leads, efficiently.

From where I sit running an email platform, the hardest part of a campaign was never pressing send. It was deciding who should get what, when, and why, and then knowing whether it worked. Agents compress the execution. They don't remove the judgment, and they make bad judgment scale faster.

Organizational impact

The campaign manager's role is moving from building every campaign to managing the system that builds and improves them. That requires capabilities most teams haven't formally developed: defining business outcomes and optimization goals, designing reusable campaign logic, managing agent permissions and approval workflows, maintaining brand and compliance guardrails, connecting CRM outcomes to execution, and auditing what agents actually did.

This is also where September's warning about the talent pipeline gets concrete. If agents assemble the segments and journeys that junior marketers used to build, we need to deliberately teach those juniors how to evaluate the agent's choices. Otherwise we'll have a team that can approve a campaign but can't tell you why it's wrong.

What CMOs should do next

• Pick the delegable work. Identify repetitive campaign tasks that agents can take on without putting brand or customer experience at risk, and start there.
• Set explicit approval thresholds for campaign creation, budget changes, audience selection and anything that reaches a customer.
• Fix lifecycle definitions before syncing them. Before sending CRM conversion signals to any ad platform, make sure your lifecycle stages mean what everyone thinks they mean.
• Model agent pricing in 2027 budgets. Ask every vendor how agentic usage is priced and what happens to cost if usage doubles.
• Retrain ops around objectives. Train marketing operations to manage goals, permissions, workflows and performance, not just platform interfaces.

Closing thought: Marketing automation made execution faster. Agentic marketing can make it adaptive. The advantage won't come from letting agents do more. It will come from designing systems that know what they should do, what they shouldn't, and when a human needs to step in.

Signal 04: The CDP Is Being Rebuilt Around the Data Platform

The customer data platform was supposed to fix one of marketing's oldest problems: customer information scattered everywhere. In practice, many organizations ended up with one more system to integrate, sync and govern. Customer data still lived in warehouses, CRMs, commerce platforms, analytics tools and ad systems, and marketing still had to move it around before anything could be activated.

October brought an enterprise example of a different approach.

On October 19, Databricks moved CustomerLake Campaign Agents into Beta and Profile Agents into Private Preview, embedding audience building, activation, personalization and Customer 360 capabilities directly in its data and AI platform. CMSWire noted that Databricks hasn't announced a general availability date, pricing or beta eligibility criteria, so this is early.

The headline example is HP. According to Databricks' HP case study, HP became audience-ready within two hours of starting with CustomerLake, compared with two to three months on a traditional CDP, expanded intent-data ingestion from 75 topics to roughly 20,000, and estimates the approach could cut CDP total cost of ownership by up to 40%. These are company-reported figures and an estimate, not audited benchmarks, and they should be read that way.

But there's a detail in the Databricks announcement that I think matters more than the headline numbers. HP says it's now building audiences directly on its Databricks data and running hundreds of audience syncs every month to Salesforce Marketing Cloud. The warehouse didn't replace the engagement platform. It became the brain, and the engagement platform stayed the hands.

Read that next to Signal 01 and October's martech story gets clearer. Salesforce is paying billions to own the real-time hands. Databricks is making a play to own the brain. Both are betting that the customer context layer is where the value concentrates.

Why we're calling this a signal

Platform signal. A data infrastructure company entering the CDP category with agents, not just storage, changes who competes for customer intelligence. The warehouse is no longer a passive store that marketing tools copy from.

Enterprise signal. HP isn't the only named example. Databricks also reported the Philadelphia Union cutting audience turnaround by 90% and Wine.com launching production emails 23 days after kickoff. Early, vendor-reported, but real organizations going live.

Architecture signal. CustomerLake works on top of the schemas, governance and data models an enterprise already maintains. That's the core of the warehouse-native argument: fewer copies of the customer, fewer sync failures, one governed source.

What this means

Data duplication becomes a strategic cost. Every time customer data is copied into another platform, we add synchronization work, governance overhead and another place for the record to go stale. A more unified foundation can reduce that cost, though it doesn't remove the need for identity resolution, data quality and activation infrastructure.

Profiles get more useful when they carry business context. A customer record on its own isn't enough. Decisions get better when profiles connect to product usage, account health, buying signals, service history and commercial outcomes. In B2B, that matters even more, because the buying group and the account relationship often matter more than any individual contact.

Agents need governed context, not just more data. An agent with access to complete, reliable customer context has a much better foundation for segmentation and personalization. Broader access also means permissions, consent, lineage and auditability have to be designed in from the start, not retrofitted after something goes wrong.

None of this makes standalone CDPs obsolete. They'll stay relevant wherever they offer differentiated identity resolution, real-time processing or activation that an enterprise data platform can't match. The structural shift is that customer intelligence is becoming an architectural capability rather than just another marketing application we buy.

Organizational impact

Customer data architecture can't be a marketing-only decision anymore. The CMO, CIO, chief data officer and revenue operations leader need to agree on the authoritative source for customer and account data, how behavioural and transactional signals are unified, which systems may activate customer data, how consent is enforced, how agents access data without exposing sensitive information, and how customer context reaches sales and service as well as marketing.

The organization also needs an honest answer to a harder question: where does a dedicated CDP still earn its keep, and where are we paying for data movement our existing platform could handle?

What CMOs should do next

• Count your customer copies. Audit how many systems hold duplicate customer profiles and behavioural data, and what it costs to keep them in sync.
• Separate capability from data movement. Identify which CDP functions are genuinely differentiated and which are mostly moving data between places.
• Bring the data team into the evaluation of warehouse-native and composable options, rather than presenting them with a finished vendor decision.
• Set agent access standards for customer and account information before agents go live, not after.
• Measure the architecture by outcomes: audience readiness, activation speed, data quality and downstream commercial impact.

Closing thought: The future of customer intelligence may not belong to the platform that stores the most customer data. It will belong to the architecture that makes trusted context available wherever a decision needs to happen.

Signal 05: Media Is Expanding Faster Than Measurement

The number of places brands can buy media keeps growing. Retailers are building media networks. Social platforms are selling inventory beyond their own apps. Streaming services are opening up. And almost every one of them offers its own measurement system to tell you it worked.

More inventory makes one question harder, not easier: which of these investments are actually creating incremental business value?

October showed both sides of that tension.

On the supply side, Marketing Dive reported that Lowe's Media Network is expanding beyond its owned properties to partners including Roku, DoorDash, Reddit, TikTok and YouTube, and partnering with Ipsos for independent incrementality measurement, bespoke marketing mix modeling and attribution. The same piece cites WARC's prediction that US retail media revenue will reach $100 billion next year, and notes that the Reddit integration matters partly because Reddit has become a top source of AI citations. A retailer's media business is no longer limited to its own site and app.

On October 5, TikTok opened its ad network, formerly Pangle, to US advertisers. According to TikTok's own help center, the network spans nearly 400,000 apps worldwide and reaches more than a billion daily active users through app traffic. Buyers were measured in their enthusiasm. Digiday's agency sources said the network is only valuable if it reaches audiences they can't already reach, and if advertisers get proper controls over which apps their ads appear in.

On the measurement side, three things landed within a few weeks of each other. On September 30, NIQ integrated Google's open-source Meridian MMM into NIQ Cadence, citing its own CMO research showing 74% of CMOs under more scrutiny to prove ROI and only 9% feeling they have adequate access to real-time insights. PPC Land pointed out that NIQ didn't quantify how much faster the integrated workflow actually is. Two days earlier, Cint launched self-service outcomes measurement combining brand lift and sales lift in one study, with Magnite as the first platform to offer it.

Then came the moment that captured the whole problem in miniature. Shortly after, Magnite published an analysis claiming its campaigns' ad recall lift was 136% higher than Cint's benchmarks. As PPC Land noted, brand lift studies are commissioned by advertisers who chose to measure, and the post didn't say whether every study on Magnite inventory was included. That's not an accusation. It's the structural issue: when the platform selling the media also curates the evidence, marketers need an independent way to weigh it.

Which is exactly what the IAB tried to provide. On October 20, it released a Cross-Channel Measurement Decision Guide for public comment, built for marketers who have to weigh marketing mix modeling, incrementality testing and attribution when the methods tell different stories. The comment period ran for just 15 days and closed on November 4.

Why we're calling this a signal

Market signal. Retail media networks and social platforms are both extending off their owned properties at the same time. Inventory is becoming a distribution network rather than a destination.

Research signal. A 65-point gap between CMOs under ROI pressure and CMOs with real-time insight is a measurement problem, not a media problem.

Product signal. Measurement providers are embedding outcome analysis directly into buying platforms and workflows, which moves measurement closer to decisions but also closer to the sellers.

Industry signal. When the IAB's response is a decision guide for interpreting conflicting evidence, rather than another metric definition, it's telling us the industry has accepted that conflicting evidence is the normal state of affairs.

What this means

Audience access matters more than media ownership. A retailer doesn't need to own every surface to run a media network, and a social platform can sell reach beyond its own app. That creates more options for advertisers and more need to understand where ads actually run and whether those audiences are genuinely new.

Measurement has to move closer to investment decisions. Traditional measurement happens after the money is spent. The direction now is to connect measurement, planning and optimization so budgets can shift on evidence rather than waiting for a quarterly read.

Attribution, incrementality and MMM answer different questions. Attribution shows which interactions were associated with conversions. Incrementality asks whether advertising caused outcomes that wouldn't have happened anyway. MMM estimates channel contribution at an aggregate level. None of them is a universal source of truth, and the IAB's decision-first approach is useful precisely because it starts with the decision you need to make, not the method you happen to have.

Retail media has to prove more than proximity to purchase data. First-party shopper data is valuable. But the strategic value depends on demonstrating incremental outcomes and reaching new audiences, with measurement that someone other than the seller can verify. Lowe's bringing in Ipsos is a sign the better networks understand that.

Having managed search budgets for a long time, I've seen what happens when every platform grades its own homework. Each one looks like it's working, the totals add up to more conversions than the business actually recorded, and the CFO stops trusting any of the numbers. More networks without independent measurement takes us straight back there.

Organizational impact

Marketing needs a measurement architecture that works across platforms rather than leaning on each platform's own attribution. That means real coordination between media planning and performance marketing, finance and marketing analytics, retail media and commerce teams, data science and marketing ops, and agencies and independent measurement partners.

The marketing leader's job is increasingly to decide how much evidence is enough for a given budget decision, not to collect more dashboards.

What CMOs should do next

• Build a measurement framework that states which decisions use attribution, which use incrementality and which use MMM, and what happens when they disagree.
• Require incrementality from media partners, not just audience scale, and prefer partners who bring independent measurement to the table.
• Check off-platform placement controls before extending campaigns into networks like TikTok's, including block lists and placement reports.
• Connect measurement to budget cycles so findings change allocation in-quarter, not in next year's plan.
• Validate independently wherever platform-reported performance is the only evidence behind a material spend.

Closing thought: The next media advantage won't come from being present on the most platforms. It will come from knowing which investments created value across all of them.

Signal 06: The Advertising Supply Chain Has a Trust Problem, and AI Content Is Making It Harder to Ignore

The internet has always had a quality problem. Low-value sites, misleading content, artificial traffic and made-for-advertising publishers have been around for years. What's changing is the speed and scale at which generative AI can produce content and fill those environments.

Three developments this month belong together.

First, Google's September 2026 spam update, which began on September 24, completed its rollout on October 8, applying globally and across all languages, according to Google's Search Status Dashboard. Search Engine Journal noted that it took close to the full two-week window, while this year's three previous spam updates each finished in under three days. Google didn't announce a new policy targeting AI-generated content, so anyone calling this an "AI content penalty" is overreaching. But a long, broad spam update is a reminder that enforcement is working hard to keep pace with content volume.

Second, in mid-October, DecryptAds published an investigation into AI recipe slop networks. Starting from around 400 suspicious AI-generated recipe sites, the researchers used technical fingerprints to connect more than 2,100 made-for-advertising domains, and AdExchanger reported that ads from major blue-chip and Fortune 500 brands were appearing on them, alongside fabricated author pages. The precision matters here: as Tech Times noted, the 2,100 figure covers connected domains, not 2,100 individually confirmed AI recipe sites or proven scams. It's a sample, not a census. It still shows how quickly cheap content can become part of the paid supply chain.

Third, on October 20, Cint published research conducted with Advertising Week, based on a survey of about 1,000 US consumers in August. It found 74% believe brands and creators have a duty to disclose AI use in advertising, 44% say undisclosed AI-generated advertising decreases their trust in a brand, and 36% describe AI-created advertising as untrustworthy.

None of this proves that AI-generated content is inherently low quality or that consumers reject AI-assisted advertising. It points to something narrower and more useful: our ability to produce content cheaply is expanding faster than the systems we use to judge its quality, provenance and commercial integrity.

Why we're calling this a signal

Platform signal. Google's longest spam update of the year shows search quality enforcement working against rising volume.

Industry signal. The DecryptAds findings show the made-for-advertising problem scaling with generative AI, and reaching the budgets of brands that almost certainly never chose to be there.

Behavioral signal. Consumers are drawing a clear line around disclosure. The risk isn't using AI. It's being seen to hide it.

Continuity signal. In September, building on Google Doesn't Need to Send You the Click Anymore, I wrote that machines are now mediating content on the way out and on the way in. October shows the dark side of that: the same cheap production that lets us publish more also lets bad actors manufacture inventory, and our ad dollars can end up funding it.

What this means

Content volume stops being an advantage. AI can multiply publishing capacity, but volume doesn't create authority, differentiation or customer value. The more cheap content enters the ecosystem, the more original research, first-hand experience and reliable expertise stand out. That's the same evidence argument I made in September: agents and people both reward credible, corroborated information.

Brand safety becomes supply-chain governance. A brand's reputation is shaped not only by the creative it approves but by the environments its ads appear in. Programmatic buying puts distance between advertiser and publisher, which makes inventory transparency, verification and exclusion controls far more important than a standard brand-safety checkbox.

AI disclosure becomes a trust decision. AI-assisted creative improves production efficiency. When and how to disclose it should depend on the content, audience expectations and applicable regulation. Cint's data suggests concealment, not use, is what damages trust.

Search quality and ad quality are connected. Google's spam update and the DecryptAds investigation are separate developments, one about ranking and one about ad supply. Both reflect pressure from low-quality content environments, and both are a warning against assuming that content which can be produced and monetized automatically deserves distribution.

Organizational impact

Content strategy, SEO, programmatic media, brand safety and legal need a shared governance model, not five separate policies. That includes editorial standards for AI-assisted content, human review for factual accuracy and brand-sensitive claims, domain-level ad controls, independent verification of programmatic inventory, clear rules for AI-generated creative and disclosure, and ongoing monitoring of low-quality environments.

For organizations investing heavily in AI content production, it also means changing what we measure. Publishing output is not a proxy for content value, and a team rewarded for volume will produce volume.

What CMOs should do next

• Review AI content workflows to make sure original expertise, fact verification and editorial accountability are built in, not bolted on.
• Audit programmatic placements and tighten exclusion controls for low-quality and made-for-advertising inventory.
• Push agencies and verification partners for publisher-level transparency, not just category-level reporting.
• Write a disclosure policy that defines when AI use is disclosed in advertising and branded content.
• Retire volume metrics for content and replace them with qualified engagement, trust signals, conversions and business outcomes.

Closing thought: AI has made content production easier. It hasn't made credibility any easier. As supply expands, the brands that protect their standards, their audiences and the environments they fund will have an advantage that gets more valuable every month.

October in 90 Seconds

• Martech is consolidating around customer context. Salesforce's pending $5.66 billion Braze deal and Hightouch's Strata acquisition show that real-time engagement and connectivity are becoming the most valuable layers in the stack.
• Conversational advertising entered its economics test. ChatGPT ads reached 89 markets and 100,000+ advertisers, but an early working paper found paid ChatGPT visits earned $0.89 per session against $2.45 for paid Google Search.
• Campaign execution turned agentic. Salesforce Campaign Agent went GA, HubSpot began syncing CRM conversions to ChatGPT Ads, and Adobe and TikTok moved more execution into conversational and AI-tool interfaces.
• Customer data moved toward the warehouse. Databricks pushed CustomerLake into beta, with HP reporting audience readiness in two hours, while still syncing audiences to its engagement platform.
• Media expanded faster than measurement. Lowe's and TikTok extended their networks off-platform while NIQ, Cint and the IAB worked to connect spend to outcomes, and a platform promptly graded itself against a measurement vendor's benchmarks.
• Ad quality became a governance issue. A long Google spam update, 2,100+ suspicious AI recipe domains carrying major-brand ads, and consumer demand for AI disclosure all point the same way.

The idea to take away: capability is getting cheaper every month, and proof is getting more expensive. The marketing organizations that win the next three years will be the ones that invest in the second as seriously as the first.

The SignalShift Perspective

Each edition this year has added a layer. In May, marketing became a system rather than a function. In June, it entered its intelligence era. In July, intelligence became the interface, and in August, that interface started to disappear. September showed us what replaces it: delegation, with buyers handing research and comparison to agents. Every edition lives in the SignalShift archive if you want the full arc.

October showed us the bill.

When I laid out access, evidence and actionability last month, I was describing what a brand needs for a buyer's agent to trust it. October answered a different question: what does a marketing organization need underneath to deliver those three things at scale? The answer turned out to be infrastructure, and the industry spent the month repricing it.

There's a temptation to read October as six separate technology stories: an acquisition, a new ad channel, a campaign agent, a CDP launch, more retail media inventory, another spam update. I think that reading misses what connects them. Three shifts sit underneath.

1. Marketing infrastructure is becoming a decision architecture. The advantage is moving away from how many applications we own and toward whether our systems can share reliable context, coordinate action and learn from results. A CRM that holds records but can't inform execution is incomplete. A campaign platform that generates content but can't see downstream outcomes is limited. An agent that can recommend an action but can't reach the right data, or act within governance rules, isn't an advantage at all. That's why Salesforce buying Braze, Hightouch buying Strata and Databricks launching CustomerLake belong in the same paragraph. They're three different bets on the same problem: making customer context actionable. HP's setup, with the warehouse as the brain and the engagement platform as the hands, is probably the most honest picture of where most enterprises will land.

2. AI capability is not commercial performance. This is October's most important counterweight. A platform can have enormous reach, attract 100,000 advertisers and generate a billion dollars in run-rate revenue without delivering the economics of an established performance channel. A campaign agent can execute more work without producing better strategy. A unified data platform can build audiences in two hours without automatically understanding customers any better. We have to separate technological progress from commercial evidence, and the more decisions our systems make on their own, the more that separation matters. Automation increases the volume of decisions. Measurement is how we find out which ones were good.

3. Trust is becoming an operating constraint, not a brand principle. AI makes it easier to create content, personalize messages, expand inventory and automate interactions. Scale also makes it easier for poor content to spread, for ads to land in environments we'd never choose, for automated systems to optimize the wrong outcome, and for commercial messages inside conversations to erode the user's confidence in the answer. Trust can't be a brand exercise we run after the technology is deployed. It has to be built into the operating model: data permissions, content standards, media controls, agent approvals, measurement integrity and transparency with customers.

Put the three together and you get the idea I'd most like leaders to take from this edition. For the last two years, the scarce thing in marketing was capability. Could we generate the content, build the audience, run the campaign, reach the channel? That scarcity is ending fast. Every vendor in this edition is making capability cheaper. What's becoming scarce, and expensive, is proof: knowing which decisions created value, which channels were incremental, which content was credible and which environments were safe.

That doesn't mean every business needs an enterprise data lake, an autonomous campaign engine or a consolidated suite. It means every marketing leader needs to understand how customer data, decision-making, execution, media and measurement fit together in their own organization, and where the evidence is weakest.

For years, the defining question for marketing has been some version of "how do we reach more customers?" The question October puts on the table is harder: how do we build an organization that understands customers, makes good decisions, executes intelligently and can prove the value of every important action? That's the infrastructure challenge behind this month's signals, and I suspect it will define the next era of marketing leadership.

Signal Ledger

#	Signal	Structural shift	Time horizon	Primary owner	CMO question to ask this quarter
01	The Martech Stack Is Consolidating	From owning workflow slices to owning customer context	1 to 3 years	CMO, CIO and marketing ops	Where does our customer context live, and who controls the connections between systems?
02	ChatGPT Ads Face an Economics Test	From search-style intent to conversational influence	Now to 2 years	Performance marketing and finance	What is our parity price for conversational ads, and are we paying above it?
03	Execution Becomes Agentic	From operating campaigns to designing decision systems	Now to 2 years	Marketing ops and legal	What can our agents launch without a human, and who approved that line?
04	The CDP Moves Toward the Warehouse	From standalone profiles to governed, architecture-level context	1 to 3 years	CMO, CDO and CIO	How many copies of our customer do we maintain, and what does each one cost?
05	Media Outruns Measurement	From platform-reported results to independent, decision-first measurement	Now to 2 years	Media, analytics and finance	Which of our channels have proven incrementality, and which only have platform reports?
06	The Ad Supply Chain Has a Trust Problem	From brand-safety checkboxes to supply-chain and content governance	Now to 1 year	Brand, media, content and legal	Do we know every environment our ads funded last quarter?

Signals We're Watching

• The first real ChatGPT Ads benchmarks from B2B. The working paper covered ecommerce. Will conversational ads improve their economics as targeting and conversion signals mature, or settle into a discovery and influence channel with a different commercial model from search?
• The Salesforce and Braze integration roadmap. The strategic value depends on how Salesforce combines Braze's real-time architecture with its customer context and Agentforce, and on what happens to Braze's open, composable positioning. The deal remains subject to approvals and won't close before Salesforce's fiscal 2028.
• The economics of agentic marketing. Will campaign agents show measurable gains in productivity, conversion and efficiency, or will organizations discover that the real work still sits in data preparation, governance and review? Watch how agentic usage pricing shows up in renewals.
• Standalone CDPs versus the warehouse. Will enterprise data platforms absorb more activation capability, or will specialist CDPs hold their ground through identity resolution and real-time processing?
• The independence of measurement. As platforms offer more integrated reporting and as sellers publish lift studies on their own inventory, will marketers hold the line on independent standards? The final IAB guide, after its comment period, will be worth reading closely.
• Off-platform networks proving incrementality. Retail media and networks like TikTok's are growing. The question is whether they can show incremental reach and reliable measurement rather than simply more inventory.
• AI disclosure norms. As AI-assisted creative becomes standard, watch for disclosure expectations to harden into platform policies, industry standards or regulation, and for brands that get ahead of it to turn transparency into a trust advantage.

Frequently Asked Questions

Why does the Salesforce Braze acquisition matter for marketers?

Salesforce agreed on October 20, 2026 to acquire Braze for about $5.66 billion in enterprise value. It matters because it shows the largest CRM vendor betting that real-time customer engagement and AI decisioning belong inside the core platform. The deal is still pending and isn't expected to close before Salesforce's fiscal 2028, so Braze customers shouldn't expect changes before February 2027 at the earliest.

Are ChatGPT ads worth it compared to Google Ads?

Early evidence says not yet on pure return. An October 2026 working paper covering 1,898 online shops found a paid ChatGPT ad visit generated about $0.89 in revenue in US traffic, against $2.45 for a paid Google Search visit, with conversion rates of 0.37% versus 1.17%. The study is observational, covers ecommerce rather than B2B, and doesn't capture later purchases, so the right move is controlled testing with a separate budget, not wholesale reallocation.

What is Salesforce Campaign Agent?

Campaign Agent is a generally available capability in Salesforce Marketing Cloud Next that takes a marketer-defined goal and assembles audiences, content, journeys and channels, with the marketer reviewing and approving before launch. Limited usage is included in Marketing Cloud Next Advanced, and additional agentic campaigns and content generations cost extra.

How does HubSpot connect to ChatGPT Ads?

HubSpot can send lifecycle-stage changes and form submissions from its CRM back to ChatGPT Ads as conversion events, which ChatGPT uses to optimize targeting. The feature is in beta, requires contact consent for data sharing, and only counts record updates made after the conversion event is created.

What is Databricks CustomerLake?

CustomerLake is Databricks' agentic customer data platform, built directly on its data and AI platform. On October 19, 2026, its Campaign Agents moved into Beta and Profile Agents into Private Preview. Databricks reports that HP became audience-ready in two hours instead of two to three months, though those figures are company-reported.

What is the difference between attribution, incrementality and marketing mix modeling?

Attribution identifies which interactions were associated with a conversion. Incrementality testing asks whether advertising caused outcomes that wouldn't have happened otherwise. Marketing mix modeling estimates each channel's contribution at an aggregate level. They answer different questions, which is why the IAB's draft Cross-Channel Measurement Decision Guide starts with the business decision and works back to the right method.

What are AI-generated made-for-advertising sites?

They're low-quality sites built mainly to monetize programmatic ad traffic, now often filled with AI-generated content and fabricated authors. In October 2026, DecryptAds connected more than 2,100 such domains to a network that started with around 400 suspicious AI recipe sites, and found ads from major brands running on them.

Should brands disclose AI use in advertising?

Consumer research suggests disclosure protects trust. Cint's October 2026 study with Advertising Week found 74% of US consumers believe brands have a duty to disclose AI use in advertising, and 44% said undisclosed AI-generated advertising decreases their trust. Each brand should set a clear policy based on content type, audience expectations and applicable regulation.

What should a CMO do first after reading this edition?

Map where customer context lives across your stack, calculate a parity price before scaling any conversational ad spend, and identify which of your channels have independent proof of incrementality versus platform-reported results only. Those three steps will show you where your infrastructure is strongest and where your evidence is weakest.

SignalShift is a monthly executive briefing on the structural shifts reshaping marketing. If this edition was useful, share it with one marketing leader who is about to sign a multi-year martech contract or move budget into a new channel, before they do.