Where Is the Value in AI for Marketing in 2026?

AI in marketing is no longer mainly about generating copy.

The relevant opportunity now stretches from research, consumer insights and content production to media optimization, personalization, lead qualification, reporting, experimentation and, increasingly, the automation of complete marketing workflows.

That creates a management problem.

The range of possible applications has expanded much faster than our ability to make reliable claims about their economic value. There is now reasonably strong evidence that generative AI can make individual knowledge-work tasks faster. There is much less independent evidence showing what happens financially when complete marketing processes are redesigned around AI and agents.

That distinction matters when building an AI strategy.

McKinsey estimates that generative AI could create productivity value equivalent to 5–15% of total marketing spending. Importantly, that estimate primarily captures direct productivity effects. Potential indirect gains from better targeting, better insights or improved conversion are largely outside the calculation.

At the same time, actual transformation remains well behind adoption. In McKinsey's May 2026 Global Marketer Survey, almost 60% of marketers reported using AI multiple times per week, while fewer than 10% had started capturing value across end-to-end workflows.

BCG found essentially the same gap. In its 2026 research among 300 global CMOs, 96% said AI was driving an end-to-end transformation of marketing. Yet 42% were still using GenAI mainly to assist people with individual tasks. Just under a third had moved towards agent-led workflows, and only 8% were running campaigns in which multiple agents operated autonomously.

For me, this is the most useful starting point for thinking about AI value in marketing in 2026.

There is substantial value available. But it does not all come from the same place, it cannot all be measured in the same way, and evidence from one level of automation should not automatically be extrapolated to another.

Four very different levels of AI value

I find it useful to distinguish four levels.

This is more than a maturity model.

Each level has a different economic mechanism.

At the copilot level, the primary question is how much employee capacity can be released. At task level, it may be unit cost. At workflow level, handoffs, lead times and external expenditure start to change. At operating-model level, the organization can potentially change the amount, speed and precision of marketing it can perform.

BCG, for example, reports from recent engagements with leading CMOs 20–30% improvements in cost efficiency, a threefold increase in marketing ROI and a tenfold improvement in campaign cycle times. These are important observations, but they are outcomes seen in BCG client engagements with leading organizations, not average effects that a CMO should put into a business case as expected returns.

That distinction between benchmark, observed case result and forecast should run through every serious AI investment case.

The AI value map for marketing

Once those levels are separated, the next question becomes more practical:

Where can a marketing organization actually look for value today?

I would divide the landscape into a number of distinct value pools.

1. Individual marketer productivity

This remains the most defensible place to start.

Marketers spend substantial amounts of time researching, reading documents, preparing briefs, drafting material, processing email, consolidating meeting information, analyzing data and building presentations.

These are exactly the types of general knowledge-work activities for which we now have relatively robust experimental evidence.

Microsoft Research conducted a six-month randomized field experiment involving more than 6,000 employees across 56 organizations. Workers using Microsoft 365 Copilot completed documents 12% faster and spent about half an hour less per week reading email. Almost 40% of employees given access used Copilot regularly during the study.

The relevant implication for marketing is not that every marketing activity becomes 12% faster.

It is that use cases such as desk research, document synthesis, first drafts of briefs, meeting processing, content adaptation and basic analysis now have credible evidence behind the basic productivity hypothesis.

I would therefore avoid claims such as:

“AI makes research 50% faster.”

The more defensible business case is:

Generative AI has produced measurable time savings on common knowledge-work tasks in randomized research. The actual saving within a particular marketing organization should be established through task-level measurement.

This distinction sounds conservative. It actually makes the investment case stronger because it replaces generic AI claims with measurable operational hypotheses.

2. Research, consumer insights and voice of customer

The technical fit between AI and consumer insight is particularly strong.

Marketing organizations increasingly have access to vast amounts of unstructured customer information:

reviews, open survey responses, social conversations, sales calls, service interactions, CRM notes and research documents.

Historically, only a fraction could be analyzed deeply because human synthesis does not scale easily.

Generative AI changes that constraint. It can help classify and synthesize large volumes of unstructured data, identify themes, surface recurring customer problems and connect information that previously sat across separate research sources. McKinsey explicitly identifies the synthesis of unstructured information and the creation of richer customer insights and segmentation as sources of marketing value.

What we do not have is a credible universal benchmark saying that consumer research becomes, for example, 60% cheaper.

So I would model this value pool in two ways.

First, straightforward productivity:

research volume × current hours per study × new hours per study.

Second, and potentially more important:

how much more customer evidence can the same insight capacity analyze?

That second effect can easily disappear from a conventional cost-saving business case.

If a team can move from manually reading a few hundred comments to systematically analyzing hundreds of thousands of interactions, the value is not simply fewer research hours. It may be a fundamentally different customer-sensing capability.

The implementation complexity is usually moderate. Not because the models themselves are unusually difficult, but because the harder problem is often access to clean, appropriately governed customer data.

3. Content production

Content is probably the most visible AI application in marketing.

AI can now support copy, imagery, video, translation, resizing, localization, product descriptions, repurposing and creative variation.

Adobe's 2025 research found that 86% of marketing leaders expected generative AI to significantly increase content speed and volume. That figure represents executive expectations, not an 86% measured productivity improvement. Adobe's 2026 research does, however, show that organizations increasingly report improvements in content volume and employee productivity from GenAI experimentation.

The economic opportunity is easily misunderstood.

The interesting question is no longer merely how quickly one copywriter can create one piece of copy.

It is what happens to the economics of the entire content supply chain when one central idea can become many approved executions:

one concept → multiple audiences → multiple channels → multiple formats → multiple markets → multiple languages.

That changes the bottleneck.

When production becomes cheap, selection, differentiation, brand consistency, approval and quality control become relatively more important.

The value can therefore appear in several places: lower internal production time, lower agency expenditure, shorter lead times and much greater output volume.

I would still calculate the hard-cost case bottom-up:

number of assets × current cost per asset – future cost per asset

and separately:

external production spend before AI – external production spend after AI.

But there is a larger strategic question behind those equations. If producing the twentieth relevant variant costs almost nothing compared with producing the first one, the optimal level of content variation can change fundamentally.

4. Product content at scale

Retailers and other organizations with very large product catalogues have a particularly clear content use case.

A single SKU may require a title, description, attributes, SEO text, marketplace copy, category information, imagery, translations and occasionally video.

Multiply this across tens or hundreds of thousands of products, multiple channels and several markets, and relatively small improvements in unit economics become significant.

Generative AI can produce or enrich many of these objects using product, PIM and supplier data. McKinsey explicitly identifies personalized product descriptions and improved product discovery as GenAI opportunities.

Again, the right answer is not to assume that AI will remove 80% of product-content costs.

A useful proof-of-value would select one category and measure:

human production time, output cost, error rate, approval rate, SEO performance and conversion before and after AI.

Only after that would I extrapolate.

This is a recurring principle throughout the AI value map: where external evidence is weak, internal experimentation should replace invented precision.

5. Paid media

Paid media requires a slightly different lens because much of the AI involved is not generative AI at all.

Machine-learning-driven bidding, audience optimization and allocation have existed for years. Modern advertising platforms are pushing this much further by combining bidding, creative, audience selection and cross-channel optimization.

Google Performance Max, for example, uses Google AI across bidding, audiences and creative. Google reports that advertisers adopting Performance Max see, on average, 27% more conversions or conversion value at a similar CPA or ROAS, including advertisers already using broad match and Smart Bidding. Google also cites a Nielsen marketing-mix-modeling meta-analysis in which Performance Max delivered an average 8% higher ROAS and 10% higher sales effectiveness than Search-only strategies.

These are much more useful external reference points than a generic claim that AI will improve media efficiency by “3–10%.”

They still require caution.

The 27% figure is a Google-reported platform benchmark. Results depend heavily on conversion data, campaign structure, budget, market conditions and the quality of the advertiser's underlying signals.

For a business case, I would therefore use the external numbers to justify the hypothesis and use controlled internal testing to justify the investment.

This is also a good example of why “AI value” cannot be evaluated only through FTE savings.

The principal objective of media AI is usually not fewer people. It is better allocation of each euro of media spend.

6. CRM, decisioning and personalization

Personalization has another economic mechanism again.

AI can support segmentation, recommendations, next-best actions, churn prediction, offer selection, send-time optimization and personalized content.

The central value equation is not:

How many marketing hours can we remove?

It is:

Can we make more relevant commercial decisions for each customer?

This is potentially one of the largest long-term marketing value pools because a decision engine can affect millions of individual interactions.

Retail already provides visible examples. McKinsey describes applications combining real-time behavioral signals, recommendation models and GenAI assistants to improve experiences such as homepage ranking, outfit curation and size recommendations.

But generic claims such as “next-best-offer creates 5–15% revenue uplift” should be treated with suspicion unless they are tied to a specific study and context.

The most rigorous business case is experimental:

AI treatment versus control → incremental conversion → incremental revenue → incremental margin.

That matters because otherwise ordinary revenue can easily be attributed to AI.

For personalization, incrementality is more important than impressive response-rate dashboards.

7. Marketing analytics and reporting

Reporting is often a less glamorous AI opportunity, but operationally it can be attractive.

Many marketing teams still spend considerable time gathering data, combining files, explaining dashboard movements, drafting commentary and preparing management presentations.

AI can increasingly automate parts of the analytical and synthesis layer.

We should be careful not to turn the Microsoft productivity evidence into a claim that marketing reporting will become 70% faster. There is no universal evidence for that. What we do know is that general document and information work can be accelerated.

Reporting is therefore an ideal local measurement case:

number of reports × hours per report × internal cost

versus:

number of reports × new hours per report × internal cost.

I would combine this with non-financial indicators such as errors, time-to-insight, number of data sources analyzed and number of analyses a team can conduct in a period.

The more interesting end-state goes beyond automated report writing. Once analytical agents can reliably interpret governed business definitions and underlying data, the reporting model itself can move from periodic production towards continuous access to analysis.

8. Experimentation

AI also changes the economics of experimentation.

When producing another headline, image, proposition, segment or landing-page variant becomes substantially cheaper, organizations can afford to explore a much larger solution space.

I would not put a universal “2–5× more experiments” benchmark into a business case. The evidence does not support that level of generic precision.

The underlying mechanism remains compelling.

Lower marginal production cost makes more experimentation economically viable.

The useful operating KPI becomes something like:

experiments per marketer per month

followed by the financial KPI that matters:

incremental gross profit generated by winning experiments.

AI's value in this area is therefore indirect. It increases the organization's capacity to learn.

That can be strategically much more important than the production saving itself.

9. B2B marketing and sales integration

The AI value map shifts again in B2B.

A retailer may have millions of transactions and relatively little information per customer. A B2B organization may have far fewer accounts, but much higher economic value and much richer information per account.

That makes a different set of activities attractive:

account research, ICP matching, buying-group identification, intent detection, ABM content, lead qualification, nurture, sales handoff and proposal support.

McKinsey estimates that generative AI could create productivity value equivalent to approximately 3–5% of current global sales expenditure, including through applications such as lead identification, prioritization, customer profiling, follow-up and lead nurturing. This is not a pure marketing benchmark, but it is relevant wherever marketing, SDR and sales activities overlap.

The underlying value equation might therefore combine:

less research time + faster qualification + additional selling capacity + incremental conversion.

The first three can often be modeled reasonably well before implementation.

Conversion should still be proven experimentally.

That distinction matters because AI business cases become fragile when uncertain commercial upside is treated as guaranteed cash flow.

Retail: one of the clearest AI value maps

Retail is useful because relatively concrete sector evidence is now emerging.

McKinsey's June 2026 European retail research reports that GenAI copilots are being used for campaign planning, content generation and targeting, with its analysis indicating roughly 15% reductions in agency spend and, in some applications, conversion improvements of up to 40%.

Those two numbers should not be interpreted in the same way.

The approximately 15% agency-spend figure is a sector-level analytical reference point.

“Up to 40% conversion improvement” is an observed upper-end result across applications. It is not a reasonable base-case forecast for an arbitrary retailer.

The right use of such evidence is to identify where experimentation is economically justified.

For an omnichannel retailer, I would build the AI portfolio around four separate commercial cases.

Content economics

Product descriptions, campaign assets, localization and creative variation can reduce cost per asset and external production expenditure.

Scale economics

The same marketing capacity can support more products, audiences, channels and markets.

Media economics

Platform AI can improve allocation across bids, audiences, channels and creative.

Customer-decision economics

Recommendations, targeting and personalization can improve the relevance of customer interactions and potentially conversion.

These are four different mechanisms. Combining them into a single claim such as “AI can automate 30% of marketing” obscures more than it clarifies.

Consider a typical national promotion.

The traditional workflow might look roughly like:

commercial planning → marketing brief → agency → creative → CRM → performance marketing → e-commerce → stores → reporting.

An AI-enabled version could increasingly look like:

commercial data → campaign copilot → content variants → automated brand and compliance review → CRM decisioning → media activation → measurement and learning.

The interesting change is not the addition of an AI tool.

It is that several boundaries between planning, production, decisioning and activation start to disappear.

That is where workflow redesign becomes more valuable than isolated productivity.

A very different value map: the organization with 100 local offices

Now take a national organization operating through roughly one hundred local branches or offices.

The economics are almost the opposite of retail.

The central challenge may not be optimizing millions of transactions. It may be making centrally developed marketing material locally relevant one hundred times over.

The workflow might be:

central campaign → local adaptation → local event → local mailing → local social content → local landing page → local lead → local sales → central reporting.

Here, central-to-local marketing automation becomes a major value pool.

An AI-enabled platform could combine centrally controlled information such as brand standards, propositions, approved claims, product information, templates and campaigns with local information about the office, employees, events, customers and market.

A local team could then generate relevant material without rebuilding the campaign from scratch.

The economic value comes from repetition.

It can be modeled without claiming an unsupported 70% efficiency gain:

100 offices × number of recurring marketing activities × current minutes per activity

compared with:

100 offices × the same activities × minutes with AI support.

That is already enough to create a credible business case.

When such an organization also serves business customers, a second value pool appears.

An AI system can combine CRM information, prior interactions, proposition data, internal documents, website activity and public account information into an account briefing. It can then support nurture, qualification and preparation for the sales handoff.

In that environment, the strongest AI opportunities may therefore be:

local marketing automation + account intelligence + qualification + sales enablement + reporting.

That is a completely different transformation portfolio from a high-volume retailer.

And that is precisely why I find generic AI maturity scores for marketing increasingly unhelpful.

Context determines the strongest value pool

Compare those two organizations.

The implication is simple but important.

There is no single “AI opportunity in marketing.”

The economics of the business determine where intelligence, automation and scale are worth most.

From use cases to workflow economics

This brings us back to the gap visible in the 2026 research.

Almost 60% of marketers may use AI several times per week, but fewer than 10% are capturing value across end-to-end workflows. BCG likewise finds that a large part of the market is still at task-assistance level.

McKinsey's more recent work makes the same point from another angle: only 28% of surveyed organizations were pursuing fundamental rewiring of teams and workflows.

This suggests that the next productivity frontier will not come primarily from giving every marketer a better chatbot.

The larger opportunity sits in redesigning how work moves through marketing.

Take content production.

A copywriter using AI may produce a first draft faster.

But a redesigned content workflow could change briefing, research, creation, image generation, localization, brand control, legal review, approval, activation and performance feedback.

The second intervention affects far more of the economics than the first.

The same logic applies to insights, campaign management, CRM and B2B qualification.

This is also why early agentic results can be substantially larger than individual productivity gains. BCG reports 20–30% cost-efficiency improvements, 3× marketing ROI and 10× faster campaign cycle times in recent client engagements with leading CMOs. Its earlier work with leading global brands reported 15–20% cost efficiencies across internal and agency spending and 5–10% incremental top-line growth.

These are not universal benchmarks.

They are evidence that the economic ceiling may rise considerably once organizations redesign workflows instead of optimizing isolated tasks.

The business case needs different standards of evidence

A useful AI portfolio should therefore distinguish three classes of evidence.

External benchmark available. Examples include general knowledge-work productivity, some retail applications and platform-driven advertising optimization.

Case evidence exists, but is not safely generalizable. Examples include agentic marketing transformations, specific personalization cases and automated lead qualification.

No credible external benchmark exists. In this category the organization should run its own controlled proof-of-value rather than borrow a percentage from another context.

This simple classification prevents false precision.

It also changes the role of external research.

Research does not need to tell us exactly what our return will be. It needs to help determine which hypotheses are plausible enough to test.

Build the business case bottom-up

The strongest overall economic reference point I would currently use at board level remains McKinsey's estimate that GenAI could create productivity value equivalent to 5–15% of total marketing expenditure.

I would use that as a sense check, not as the financial model itself.

The financial model should start with the actual marketing operation.

For each workflow:

volume × current effort × current cost.

Then run a controlled AI intervention and measure:

new effort, output volume, errors, approval rate and commercial performance.

From that, I would separate five value buckets.

There is one financial distinction I would make explicit in every board-level business case:

capacity release is not the same as cashable savings.

If ten marketers save two hours per week, the organization has created useful capacity.

It has not automatically reduced payroll.

That capacity can produce real economic value if it is reinvested into more output, faster delivery, more experimentation, higher-quality work or avoided hiring. But calling it headcount savings before any cost has disappeared makes the case less credible, not more.

The operating model becomes part of the value case

Once AI moves beyond individual assistance, technology is no longer the only thing that changes.

Content provides a simple example.

If AI allows a marketing organization to produce ten times as many creative variants, somebody still needs to determine which variants should exist, what constitutes acceptable quality, which claims are allowed, how the brand is represented and when human review is required.

The bottleneck moves from creation to orchestration and judgment.

The same pattern appears across marketing.

Insights teams move from manually synthesizing every data point towards designing how evidence is gathered, interpreted and challenged.

CRM teams move from scheduling campaigns towards defining decision rules, objectives, constraints and experiments.

Performance teams increasingly supervise machine-driven optimization rather than manually controlling every bid.

Marketing operations teams start designing workflows that combine people, agents and systems.

The gain is therefore not simply “fewer tasks.”

The operating model itself changes.

BCG's 2026 research makes this particularly visible. Its leading marketing organizations are investing not only in technology, but in data foundations, brand-intelligence layers, orchestration and new talent.

That is an important warning for companies trying to scale AI through licenses alone.

AI adoption and marketing transformation are not the same thing.

Tooling should come after the workflow

This is also why I would resist starting an AI marketing strategy with a long shopping list of tools.

The eventual architecture will normally combine several layers:

general-purpose AI for knowledge work; CRM or CDP capabilities for customer and account context; CMS, DAM and PIM systems for controlled content and product information; creative AI for production; advertising-platform AI for media; analytical infrastructure for measurement; and automation or agent orchestration for connecting activities into workflows.

Microsoft 365 Copilot is an obvious example at the general knowledge-work layer. Adobe is building AI-enabled content workflows around content generation, brand control and activation. Google Performance Max represents a highly developed form of AI-enabled media optimization.

But the sequence matters.

Workflow first. Architecture second. Product selection third.

Otherwise the AI portfolio easily becomes whatever the current technology stack happens to offer.

Governance is now part of the European marketing operating model

For European organizations, AI governance has also moved from future consideration to present operating requirement.

Article 50 of the EU AI Act has applied since 2 August 2026 and introduces transparency obligations for certain AI systems and types of AI-generated or manipulated content. The European Commission published final implementation guidelines on 20 July 2026.

The nuance matters.

It would be incorrect to conclude that every piece of AI-generated marketing content must simply carry an “AI generated” label.

The requirements differ for providers and deployers and depend on the system and content involved. Among other things, the rules cover informing people when they directly interact with certain AI systems, machine-readable marking of AI-generated or manipulated content by providers, and disclosure requirements for deployers in areas such as deepfakes and certain public-interest text generated without human editorial control. Specific exceptions apply.

For marketing leaders, the broader implication is more important than the legal detail.

Governance has to be designed into the marketing workflow itself.

That includes brand standards, privacy and data use, human review, intellectual-property considerations, claims and regulatory compliance, transparency and clear responsibility for AI-generated output.

The more autonomous the workflow becomes, the less credible it is to treat these controls as an approval step added at the end.

What a CMO should take from the 2026 evidence

The evidence now supports a more sophisticated position than either extreme in the AI debate.

It is no longer credible to say that AI value in marketing is mostly speculative.

There is meaningful evidence of productivity improvement, increasingly concrete sector evidence, established machine-learning value in media, and early examples of much larger gains from redesigning workflows.

But it is equally difficult to justify a spreadsheet containing twenty marketing activities with confident percentages for time savings, cost reduction and revenue uplift.

The evidence is simply not that mature.

The strongest general reference point remains McKinsey's estimate of 5–15% productivity value relative to total marketing spending. Individual knowledge-work experiments show more modest and measurable effects, such as Microsoft's 12% faster document completion. At the other end of the spectrum, leading organizations in BCG client work have produced much larger gains when complete workflows and operating models were redesigned.

All three observations can be true.

They refer to different levels of transformation.

For a retailer, I would therefore look particularly hard at product-content scale, the creative supply chain, media AI, CRM and personalization, and experimentation.

For a distributed organization with one hundred local offices, I would prioritize central-to-local content, local campaign activation, account intelligence, lead qualification, sales enablement and automated reporting.

Another business model will produce another map.

That, ultimately, is the point.

The relevant management question in 2026 is no longer:

Which AI tools should marketing implement?

Nor is it:

What percentage of marketing can we automate?

A more useful sequence of questions is:

Where does marketing currently spend money, time and management attention? Where does better decision-making create commercial value? Which high-value workflows can now be redesigned because AI changes the economics of information, content or execution? And for each expected benefit, do we have credible external evidence or do we need to prove it ourselves?

Answer those questions first, and AI becomes much easier to evaluate.

It stops being a catalogue of interesting use cases.

It becomes the redesign of a marketing operating model around the places where better intelligence, lower marginal cost, greater scale and faster learning actually create value.