How to express AI Productivity in marketing

One of the most important misconceptions in AI business cases is that a percentage productivity gain can automatically be translated into the same percentage reduction in employees.

If AI allows a marketer to perform an activity faster, the immediate result is additional capacity. It does not automatically create a payroll saving.

Research increasingly confirms substantial productivity effects.

In a randomized MIT experiment involving 453 college-educated professionals, selected writing tasks performed with ChatGPT took 40% less time, while output quality was rated 18% higher.

In a field experiment involving 758 BCG consultants, participants using GPT-4 on suitable tasks worked more than 25% faster, completed more than 12% more tasks, and produced significantly higher-rated work. The same study showed that performance could deteriorate on tasks outside the model's capabilities.

In a Microsoft experiment involving more than 6,000 knowledge workers across 56 organizations, documents created with Copilot were completed on average 12% faster.

These are real productivity effects.

But they do not mean that an organization achieving a certain percentage of time savings can automatically remove the same percentage of its workforce.

Those are two different economic measures.

From task productivity to business value

The more accurate logic is:

AI saves time → capacity becomes available → management decides how to use that capacity → only then does a financial effect emerge.

That capacity can be used for:

more output;
higher quality;
shorter lead times;
more experimentation and innovation;
lower external spending or avoided hiring;
actual reduction of internal capacity.

The International Labour Organization concluded in its 2025 global analysis that approximately one in four workers worldwide is employed in an occupation with some exposure to generative AI, but that because many human tasks remain necessary, job transformation is more likely than complete replacement.

Its June 2026 review of empirical research reached a similar conclusion: GenAI productivity gains are real but uneven, while large-scale labor displacement has so far remained limited. Employee-reported time savings have also not automatically translated into proportional increases in measured output, income or employment changes.

That is the difference between a technical AI business case and an economic AI business case.

Why “40% faster” does not mean “0.4 FTE less”

Suppose a copywriter completes part of the writing process significantly faster.

We still do not know:

how much of the total job consists of that activity;
whether AI works equally well across all writing tasks;
how much human review remains necessary;
whether additional demand for content emerges;
whether there is already a backlog;
whether time shifts to other activities;
whether external agencies or future vacancies can be reduced instead.

Most jobs are bundles of different tasks.

Acemoglu and Restrepo's research on automation makes a similar distinction between a displacement effect, where technology replaces human labor in existing tasks, and a reinstatement effect, where technology creates new human tasks and activities.

AI therefore tends to automate tasks within jobs, not entire jobs in a simple one-to-one relationship.

That matters in marketing.

A content marketer may also perform customer research, briefing, stakeholder management, concept development, quality control, brand management, distribution, performance analysis and optimization.

If AI accelerates first-draft production, the rest of that value chain does not disappear. Some downstream activities may even increase because more content can now be produced.

More output with the same people

For many marketing organizations, the first source of AI value will simply be doing more with the existing team.

Marketing demand is rarely fully satisfied. There are almost always campaigns, segments, landing pages, customer journeys, analyses or experiments that remain undone because of capacity constraints.

The relevant measure may therefore be less about cost per marketer and more about marketing output per employee.

McKinsey estimates that generative AI could create productivity value in marketing equal to approximately 5–15% of total marketing spend. Importantly, its analysis does not fully account for additional benefits from better insights, new campaign ideas or improved targeting.

More experimentation can create more value than cost reduction

Marketing is not a purely administrative function. Additional capacity can create commercial value.

If AI makes it easier to develop propositions, vary creatives, test audiences, personalize journeys or create landing-page variants, the same team can test more hypotheses.

That increases learning velocity: the speed at which the organization discovers what works.

Quality can be part of the return

Traditional productivity calculations often focus only on time.

But the research also shows quality effects.

In the MIT study, output quality increased by 18%. In the BCG experiment, quality also improved significantly for tasks within GPT-4's capabilities.

In the study by Brynjolfsson, Li and Raymond involving 5,179 customer-support employees, productivity increased by an average of 14%, with the strongest gains among less experienced employees. The researchers also found improvements in customer sentiment and employee retention, alongside evidence that employees learned from the system.

AI may therefore also help distribute the knowledge and practices of stronger employees more widely.

For marketing, that could improve briefs, brand consistency, reuse of campaign knowledge, access to customer insights, analysis and on-the-job coaching.

Part of the value may come from reducing quality differences within the team rather than reducing the team itself.

Speed is also a business case

A campaign going live earlier can create value without eliminating a single role.

The same applies to faster lead follow-up, quicker product-content creation, earlier trend detection or shorter approval cycles.

The value of an hour saved is easy to calculate. The value of being one week earlier to market is harder, but may be commercially much greater.

An AI business case should therefore measure not only labor hours, but also:

campaign lead time;
briefing-to-live time;
concept-to-approval time;
time-to-insight;
lead-response time;
product-content lead time.

The business case expands from cost per output to speed of output.

Released capacity can reduce external costs

There is an important difference between:

making an internal employee redundant

and:

buying less external work.

Higher internal productivity may reduce dependence on agencies, freelancers, translators, production partners or temporary capacity.

McKinsey explicitly notes that generative AI may allow companies to shift resources toward higher-quality owned channels and reduce spending on external channels and agencies.

These savings can often become cashable without reducing the internal workforce.

A marketing organization can therefore retain its people, increase their productivity and reduce marginal external production spending.

AI can also prevent future hiring

Another important category is avoided hiring.

A growing organization may otherwise need more marketers to support additional customers, products, campaigns, markets or channels.

With AI, the existing team may be able to absorb part of that growth.

The financial benefit then comes from personnel costs growing more slowly, rather than existing employees leaving.

The relevant management question becomes:

How much additional business volume can our current organization support before we need additional capacity?

PwC's 2026 AI Jobs Barometer, based on more than one billion job advertisements and company data, found that headcount in companies in the most AI-exposed categories had grown faster since the chosen base period than in the least exposed categories.

The evidence is correlational and does not prove that AI caused that job growth. But it does challenge the assumption that higher AI productivity must automatically lead to lower headcount.

Work changes, and so do the skills that matter

When producing a first draft becomes cheaper, other capabilities become relatively more valuable:

defining the problem;
judging quality;
understanding customers;
setting priorities;
making creative choices;
designing experiments;
interpreting data;
influencing stakeholders;
assessing brand and reputational risk;
reviewing AI output.

PwC reports in its 2026 AI Jobs Barometer that skills in highly AI-exposed occupations are changing faster and that some AI-exposed junior roles increasingly require capabilities traditionally associated with seniority, such as judgement and leadership.

Again, this is observational rather than causal evidence. But it supports the idea of job transformation, not only job elimination.

For marketing, the implication is clear:

AI may reduce the economic value of some production tasks while increasing the value of judgement, creativity, commercial decision-making and orchestration.

AI transformation therefore also requires job redesign.

Productivity gains are not evenly distributed

A business case should also avoid assuming that every employee benefits equally.

In the Brynjolfsson, Li and Raymond study, average productivity increased by 14%, but by 34% for less experienced and lower-performing employees. Effects for the most experienced employees were limited.

The BCG research demonstrated another issue: AI improved performance for tasks within its technological frontier but could reduce performance on tasks outside it.

A CFO therefore cannot responsibly assume:

“AI makes every marketer X percent more productive, so we can remove X percent of the workforce.”

Effects vary by task, employee, process, AI system, data quality and maturity of use.

AI business cases should therefore be built around processes and tasks, not a generic reduction factor applied to payroll.

Capacity release is not the same as cashable savings

For discussions with executives and finance teams, I would distinguish explicitly between two concepts.

Capacity release

Time that is no longer required for existing activities because of AI.

That capacity can be reused.

Cashable savings

Costs that actually disappear from the P&L.

For example:

supplier contracts are reduced;
external production disappears;
planned vacancies are not filled;
overtime falls;
temporary capacity is reduced;
in specific cases, internal staffing is adjusted.

Not every capacity release becomes a cashable saving.

That means the business case should never simply be:

hours saved × average salary = cost saving.

That calculation initially represents the economic value of released capacity.

Only when the organization decides not to redeploy that capacity and can actually remove the cost should it be recognized as a hard saving.

For marketing, reinvestment may be rational

Marketing is to a significant extent a growth function.

When AI releases capacity, management has a choice:

A. generate the same commercial output with lower marketing costs

or:

B. maintain the marketing cost base and try to generate more commercial value.

The second option is often underestimated when AI is positioned primarily as an efficiency program.

McKinsey's estimate of 5–15% productivity value relative to marketing spend explicitly excludes some potential benefits from better insights, campaign ideas and targeting.

A pure cost-cutting business case can therefore miss part of the potential return by design.

A marketing director can reasonably say:

Our objective is not to monetize every hour saved by AI through fewer people. We want to reinvest part of that capacity in customer interaction, experimentation, better content, personalization, analysis and innovation where that can create additional commercial value.

The same logic applies outside marketing

The principle is broader.

In customer service, AI productivity can be used for shorter waiting times, more complex cases, better service or greater customer volume.

In finance, released capacity can move toward forecasting, scenario analysis, business partnering or risk analysis.

In HR, it can move from documentation and administration toward talent development, workforce planning or employee support.

In IT, faster software development can mean a larger backlog processed, less technical debt, better security or faster product development.

The general rule is:

AI reduces the human input required for certain tasks. What that does to headcount depends on the amount of work the organization wants to perform and the value of additional output.

AI can still reduce headcount

A credible argument should not deny this.

The World Economic Forum reported in its Future of Jobs Report 2025 that 41% of surveyed employers expected to reduce staff where AI automates certain activities. At the same time, 77% intended to upskill or reskill employees for AI, while almost half expected to redeploy employees from AI-affected roles elsewhere in the organization.

The right conclusion is therefore not:

“AI will never cost jobs.”

It is:

“It is economically incorrect to treat every AI productivity gain in advance as a headcount reduction.”

Structural workforce reduction is more plausible when:

work is highly repetitive;
volume is stable;
little human judgement is required;
quality can be checked automatically;
no additional demand for the output exists;
released capacity cannot create sufficient value elsewhere.

In those conditions, headcount reduction can be a real outcome.

But it should be an outcome of process analysis, not a generic assumption imposed on every AI investment.

Why a no-layoff objective can support transformation

There is also an organizational-change argument.

If employees believe that every efficiency improvement they help create will ultimately be used to remove their own role, the organization creates the wrong incentive.

Why would employees document processes, share expertise, train AI systems, identify inefficiencies or actively experiment if success could make their own position less secure?

OECD research into AI in the workplace provides indications that training and worker consultation are associated with better outcomes. Workers at organizations that involved employees or employee representatives in technology implementation more often reported positive effects on performance and working conditions. The OECD emphasizes that these findings are correlational.

A marketing director could therefore establish a clear principle:

“The objective of this phase is not to automate people out of the organization. It is to automate work that adds little distinctive value, so that our people can spend more time on work that creates greater value for customers and the business.”

That gives employees a reason to become designers of the transformation rather than passive subjects of it.

What management should promise instead

A more credible AI business case has four value objectives:

Productivity: reduce time spent on routine production and administration.

Growth and throughput: execute more relevant campaigns, interactions, analyses and experiments with the same people.

Quality and speed: improve output and move faster from insight to activation.

Cost control: reduce external capacity and support growth with a cost base that grows more slowly.

Only when capacity is structurally redundant, cannot be redeployed productively and can actually be removed from the cost base should it be classified as a cashable personnel saving.

A better AI scorecard

Management should measure multiple forms of value.

Productivity

hours per activity;
output per employee;
manual processing;
AI usage and adoption.

Growth and commercial output

campaigns;
experiments;
segments served;
personalized communication;
incremental revenue or margin where causally measurable.

Speed

time-to-market;
campaign cycle time;
time-to-insight;
lead-response time.

Quality

approval rates;
errors;
brand compliance;
customer response;
objective quality measures where available.

Cost

agency spend;
freelance and temporary spend;
overtime;
avoided hires;
AI and technology costs.

People

skill development;
internal mobility;
employee satisfaction;
retention of critical expertise.

This creates a much clearer picture of what AI is actually doing to the organization.

A better question for the AI business case

“How much more economic value can the same organization produce with the same amount of human capacity?”

And only then:

“What part of that value do we want to realize through growth, quality, speed, avoided external costs, avoided future hires or actual personnel savings?”

The empirical evidence supports this broader perspective. The ILO currently sees mainly job transformation and still-limited large-scale displacement. The OECD emphasizes that AI can create new tasks while automating existing ones. Economic research on automation explains why displacement and the creation of new human activities can happen simultaneously.

A marketing director can therefore consistently say:

“We expect substantial productivity improvements from AI.”