How AI Can Make a Company More Creative, and More Profitable
Companies are producing more content than ever, yet many still struggle to generate ideas that customers notice, remember or act upon. Generative AI appears to offer an answer: it can draft copy, create images, summarise research, personalise campaigns and turn one approved concept into dozens of market-specific versions. The commercial opportunity is considerable, but it is also easily misunderstood.
AI does not create growth simply by helping a marketing department publish more material. It creates value when it removes a genuine bottleneck: reducing the cost of product visualisation, shortening campaign development, making localisation economically viable or allowing a team to test more ideas before committing its budget.
McKinsey estimates that generative AI could create productivity value equivalent to between 5 and 15 percent of total marketing expenditure. Gartner found that, among marketing organisations already using generative AI, 77 percent had adopted it for creative-development work. Yet adoption is not the same as return. Gartner has also found that companies do not necessarily achieve significant business improvements merely by giving employees access to generative tools.
The distinction is important. Providing a creative team with an image generator or writing assistant may save individual employees some time. Redesigning the process through which the company develops, tests, approves and distributes creative work can change the economics of the entire function.
Where AI Creates Measurable Value
The strongest early applications of creative AI tend to solve one of four problems: slow production, expensive adaptation, fragmented information or an inability to test enough ideas before committing budget.
When campaign development takes too long, AI can help teams generate and visualise several early directions before choosing which one deserves full production. Traditionally, an agency or internal department might spend days preparing three creative routes for a campaign presentation. AI can help the team explore a much wider range of possibilities before designers refine the strongest ideas.
The relevant measure is not the number of concepts created. It is the reduction in time from the initial brief to an approved creative direction, together with any improvement in the quality of the decision. Producing 50 mediocre concepts is not more valuable than producing five credible ones.
When localisation is the bottleneck, AI can translate and adapt approved material for different countries, customer groups and channels. A global company may begin with one campaign but require different languages, product references, legal wording, formats and cultural details in each market. Generative tools can help create these variations without asking designers to rebuild every asset manually.
The company should compare the cost and turnaround time per market with the previous process. It should also record how much human correction is required. A translation generated in seconds is not efficient if local teams spend hours repairing inappropriate language or factual mistakes.
When product imagery is too expensive to produce at the required scale, generative tools can create controlled backgrounds, settings and early mock-ups. A furniture company might show the same chair in several interiors. A fashion retailer could visualise alternative campaign settings before arranging a full photoshoot. A property developer might create early representations of different design directions before commissioning final architectural images.
The commercial value should be assessed through the cost per approved asset, the speed at which products reach the market and, where appropriate, the effect on customer conversion. The relevant figure is not the cost of generating the first image. It is the complete cost of producing an image that is accurate, legally usable and suitable for publication.
When customer communication is too generic, AI can produce different versions based on genuine differences in customer needs. A bank might explain the same investment service differently to an experienced investor and a first-time client. A hotel could emphasise reliable internet access and early breakfast service to a business traveller, while highlighting connecting rooms and flexible meals to a family.
The measure of success is whether the adapted communication improves response, conversion, retention or customer satisfaction. Changing a photograph or inserting someone’s first name does not constitute meaningful personalisation if the underlying proposition remains unchanged.
When designers spend too much time resizing, reformatting or adapting existing work, AI can automate routine production while keeping approved brand elements in place. This is often less glamorous than asking a model to invent an advertising campaign, but it may produce a more reliable financial return.
A marketing department might already have a strong campaign but need versions for social media, email, websites, digital advertising, retail displays and several international markets. Automating part of this production can release designers from repetitive work.
The company should calculate the number of production hours saved and establish how those hours will be used. If employees are simply expected to produce more material, the organisation has increased output. If they can redirect their time towards customer research, stronger concepts or neglected commercial problems, it may have improved creative capacity.
When customer research is difficult to use because it is spread across interviews, reviews, surveys and sales notes, AI can help organise the material and identify recurring themes. A team could use it to group customer complaints, summarise interview transcripts or compare reactions to different product concepts.
The benefit is a shorter path from raw information to a usable creative brief. The findings still need to be checked against the original material. A model can overlook context, exaggerate patterns or present an isolated opinion as if it represented a broad customer group.
When sales teams repeatedly create their own presentations and proposals, AI can help produce drafts from approved content, templates and product information. This can be particularly useful in professional services, financial services, technology and industrial companies, where sales material must combine standard information with details relevant to an individual client.
The relevant measures are preparation time, factual accuracy, brand consistency and compliance with internal rules. The system should help employees assemble reliable material, not allow them to invent unsupported claims more quickly.
Finally, when ideas are too costly to test, AI can create inexpensive prototypes before the company commissions full production. A furniture business might visualise several finishes before manufacturing physical samples. A beauty company could test different campaign directions before arranging photography. A hospitality group might explore new room concepts before appointing designers and contractors.
The value lies in the number of credible concepts tested and the reduction in money spent on weak ideas. AI is especially useful when it allows the company to abandon an unpromising direction before significant resources have been committed.
The principle is consistent across all these uses: the company should measure an improvement in cost, speed, quality or customer response. Producing a larger quantity of material is not, by itself, evidence of value.
Start With the Bottleneck, Not the Technology
Many AI programmes begin with the wrong question: “What can we do with generative AI?” This encourages teams to search for impressive demonstrations rather than commercially relevant problems.
A better starting point is to identify where creative work is currently slow, expensive or inconsistent. The company might discover that its principal problem is not generating ideas but obtaining approval from five internal departments. It may already have effective campaigns but lack the capacity to adapt them for smaller international markets. Its designers may be overwhelmed by routine requests from sales teams. Its product-development department may commission expensive prototypes before customer demand has been tested.
Management should document the existing process before introducing AI. For a marketing campaign, this could include the number of days from the initial brief to the first creative presentation, the employee and agency hours involved, external production costs, the number of approval rounds and the performance of the final work.
AI can then be introduced into one defined part of that process. If a team previously needed ten working days to develop three early campaign routes, a pilot might test whether it can develop and evaluate ten credible routes within five days.
The purpose is not to prove that the technology is impressive. It is to determine whether the new process results in a better decision at a lower cost.
Use AI to Increase the Number of Ideas — Not Lower the Standard
One of generative AI’s clearest advantages is that it makes variation inexpensive. A team can ask for ten headlines instead of three, explore several visual worlds or test different ways of explaining the same product.
This does not mean that every variation should reach the customer. AI is most valuable during divergence, when the company wants to expand the range of possibilities. Human judgement becomes more important during convergence, when the team decides which ideas are distinctive, credible and appropriate for the brand.
A financial-services company, for example, could use AI to develop several explanations of a complex investment product: a concise version for experienced investors, a plain-language version for first-time clients, a visual explanation for social media and a longer version for financial advisers.
A compliance specialist must still verify every material claim. An editor must ensure that simplification has not made the communication misleading. AI accelerates the creation of alternatives; it does not assume responsibility for the final communication.
The same principle applies to product design. A furniture company can rapidly visualise a chair in different materials, proportions and domestic settings. Designers can use these images to discuss direction with customers or management before producing expensive technical prototypes. The generated image is not the finished product. It is a decision-support tool.
Creativity Begins Before the Prompt
The most important creative decisions usually occur before anyone opens an AI tool. The team must identify the commercial problem, understand the customer and determine which response it wants to create.
A prompt cannot compensate for an indistinct brand or a superficial brief. When a company asks for “a premium campaign aimed at younger consumers”, the model has little choice but to draw on familiar visual and verbal conventions. The result may look polished while resembling the output of every competitor making the same request.
Effective AI-assisted work depends on proprietary inputs: customer research, product knowledge, brand history, cultural understanding, performance data and the judgement of people who know which conventions to follow and which to reject.
The model contributes speed and variation. The company must supply the point of view.
This is why experienced creatives may obtain more value from generative systems than organisations hoping to bypass creative expertise. A skilled art director can identify the promising detail in an imperfect image. An editor can recognise which draft contains an original argument. A product designer understands whether a visually impressive concept can actually be manufactured.
AI reduces the cost of generating possibilities, but human expertise remains responsible for selection. In creative work, selection is often where much of the commercial value lies.
Personalisation Must Be Based on a Real Difference
Generative AI makes it possible to create material for increasingly small audience groups, but personalisation is useful only when customers’ needs genuinely differ.
A hotel should not simply show a family and a business traveller different background photographs. It should address their different decisions. The family may need information about connecting rooms, meal flexibility and children’s activities. The business traveller may care about reliable internet access, an early breakfast and the journey to the airport.
AI can help produce the relevant versions once the distinctions have been identified. It cannot identify those distinctions reliably if the company’s customer data is poor or its segmentation is superficial.
A useful personalisation process begins by identifying the customer’s decision. What are they trying to choose, understand or accomplish? The company should then determine which needs or constraints materially distinguish one audience from another.
The proposition can then be adapted by changing the evidence, benefit or service being emphasised. Finally, the company should compare the result with a control group by measuring conversion, engagement, repeat purchase or another relevant outcome.
This matters because AI personalisation can become intrusive. A customer may welcome a relevant product recommendation while objecting to communication that reveals how closely the company has inferred their private circumstances. The technical capability to personalise does not automatically create permission to do so.
Real Examples Show Where the Value Sits
The practical uses of creative AI can already be seen in large companies, although vendor case studies should be treated as reported results rather than independent proof that every company will achieve the same outcome.
Adobe says that Gatorade used its Firefly technology to support a digital experience through which customers could personalise squeeze bottles. The system produced hundreds of thousands of controlled, on-brand designs. This is not simply an example of replacing a designer. AI made it commercially possible to offer a large number of individualised designs within parameters established by the brand.
The same technology can be used for less visible production work. Enterprise systems can automate the creation of asset variations for different audiences, channels, formats and geographical markets. The value lies in reducing the manual work required to transform one approved idea into a complete campaign system.
Other companies use generative AI much earlier in the creative process. A retailer might visualise a proposed campaign before organising photography. A consumer-goods company could compare packaging directions before commissioning physical prototypes. An agency might prepare a more complete representation of an idea before presenting it to the client.
These examples represent three different commercial models.
The first is operational efficiency: producing routine brand material more quickly and consistently. The second is creative decision support: visualising and evaluating ideas before significant money is committed. The third is customer participation: allowing customers to create personalised outputs within a controlled brand environment.
Companies should be precise about which model they are pursuing. A tool introduced to reduce resizing work should not be judged according to whether it produces breakthrough campaign ideas. A customer-personalisation service should not be justified solely by the number of images it can generate.
Calculate the Full Cost, Not Just the Subscription
Generative AI is frequently presented as an inexpensive replacement for production expenditure. The cost of the model, however, may be only a small part of the complete implementation.
A realistic business case should include software licences or model charges, integration with existing systems, employee training, the creation of templates and brand controls, legal review, human quality assurance, security work and ongoing maintenance.
A company may save 1,000 production hours but require 600 additional hours of checking, correcting and administering the new system. A saving may still exist, but it is smaller than the headline claim.
The relevant calculation is not the cost per generated asset. It is the cost per approved, accurate and commercially usable asset.
Consider a company that needs 100 localised campaign assets. Under its existing process, the average production cost is €200 per asset, producing a total cost of €20,000 and a completion time of six weeks.
Under an AI-assisted process, the company might spend €2,000 on software and processing, €4,000 on integration and templates and €6,000 on human review. The total cost would be €12,000, with delivery completed in three weeks.
In this hypothetical example, the company saves €8,000, or 40 percent, while halving the delivery time. That is a defensible business case. Saying that the AI generated the first versions “in seconds” is not, because it excludes the work required to make the material usable.
A Practical 90-Day Pilot
A company does not need an enterprise-wide transformation programme to begin. It needs a contained use case, a responsible owner and a reliable baseline.
During the first 15 days, the company should define the problem. It should choose one repeatable, moderately low-risk activity, such as adapting approved social assets, summarising customer interviews, producing early product visualisations or drafting internal sales material.
The present cost, production time, error rate and approval requirements should be recorded. The company should also determine which information the AI system may and may not receive.
Between days 16 and 30, the team should compare two or three tools using genuine company tasks rather than relying on vendor demonstrations. The assessment should cover output quality, speed, security, licensing, data retention, integration and the amount of correction required.
A representative test set might contain between 20 and 50 real examples. It should include difficult and unusual cases, not only the material most likely to produce an attractive result.
During days 31 to 60, a small, trained team can use the selected system within a controlled workflow. Human review should remain in place, and corrections should be documented. The company should compare the AI-assisted process directly with the existing process.
It is important to measure the time required to reach final approval, not merely the time required to produce a first draft.
Between days 61 and 75, management should examine the total time saved, total cost, quality, error rate, employee adoption, customer performance and any legal or security concerns. The pilot should be redesigned or stopped if the system creates more checking and correction work than it removes.
During the final 15 days, the company can decide whether to scale the process. Expansion is justified when the pilot demonstrates a repeatable improvement and the necessary controls are in place. The next stage might cover another market or product category rather than immediately extending the technology throughout the organisation.
What Should Remain Human-Led
A credible AI strategy also defines where the technology should not take the lead.
Final brand positioning should remain accountable to people who understand the company and its market. Sensitive communications involving redundancies, crises, serious customer complaints or significant reputational issues require human judgement. Medical, legal and financial claims need qualified review. Original creative direction should not be reduced to repeatedly imitating material that has already performed well.
The level of human control should rise with the consequence of an error.
Internal ideation, rough mock-ups, formatting and meeting summaries are generally lower-risk applications. Light review may be sufficient, provided confidential information is protected.
Public marketing copy, product imagery, localisation and personalised customer communication carry moderate risk. They normally require formal brand, factual and legal review.
Regulated claims, sensitive customer information, public-policy communication, crisis material and content depicting real people present higher risks. These applications require specialist approval, documented controls and clear accountability.
This graduated approach prevents every use of AI from becoming trapped in bureaucracy while ensuring that consequential material is reviewed properly.
Copyright and Commercial Safety Cannot Be Added Later
Generative AI introduces difficult questions about ownership, training data and resemblance to protected work. Companies should not assume that an output is commercially safe merely because a platform allows it to be downloaded.
Contracts with AI providers need to be examined carefully. Management should understand whether customer inputs are retained, whether they may be used to train the provider’s systems, which commercial protections are offered and who bears responsibility if an output creates a dispute.
Creative teams also require practical rules. Asking a model to reproduce the recognisable style of a living artist may create ethical and reputational concerns even where the precise legal position is disputed. Generated people, voices and events can mislead an audience unless their artificial nature is sufficiently clear.
Brand safety extends beyond copyright. A system may generate inaccurate product features, culturally inappropriate imagery or scenes that conflict with the company’s values. Human review must therefore examine meaning and context, not merely whether the output looks technically convincing.
These controls may appear to slow implementation. They are still less expensive than withdrawing a campaign, facing litigation or explaining why confidential material was uploaded to an unsuitable service.
Creative Roles Will Change Before They Disappear
AI is likely to reduce the time required for some forms of routine production. Basic adaptations, first drafts, background removal, stock-style imagery and standard product descriptions can already be completed with less manual work.
That does not make creative departments unnecessary. Their contribution moves towards defining problems, directing systems, evaluating alternatives and connecting creative decisions with commercial strategy. The premium shifts from execution alone towards judgement.
This creates a less obvious management problem. Junior employees have traditionally developed judgement by performing production work: researching, drafting, correcting mistakes and watching experienced colleagues improve their output. If AI absorbs too much of this work, companies may reduce costs now while weakening the development of future expertise.
Training must therefore extend beyond teaching employees how to write prompts. People need to understand how models fail, how to verify outputs, how to preserve brand distinction and when using AI is inappropriate. They must retain enough subject knowledge to recognise a plausible but defective result.
The future creative professional is not simply a person who works faster with software. It is someone capable of directing a wider production system without surrendering responsibility to it.
What Management Should Ask Before Approving Investment
Before approving a creative AI project, senior managers should be able to answer several basic questions.
Which existing business problem does the system solve? What does the current process cost, and how well does it perform? What will become better for the customer? Which financial or operational measure will demonstrate value? What data, copyright and security risks are introduced? Where must a qualified person retain approval? How easily can the company change providers or return to the previous process?
A proposal that cannot answer these questions is probably still a technology experiment rather than a business initiative.
Management should also ask what the organisation plans to do with the time it saves. If AI reduces routine production work, will employees conduct more customer research, develop stronger concepts or support additional markets? Without a deliberate answer, the apparent productivity gain may disappear into a larger volume of low-value activity.
The Customer May Not Welcome More AI Content
Greater production capacity does not guarantee greater customer appetite. As synthetic images, automated articles and generated advertising become more common, audiences may become more sensitive to communication that feels generic, misleading or unnecessarily artificial.
A customer may value faster translation, clearer product explanations or a service that helps visualise a purchase. The same customer may respond negatively to artificial testimonials, undisclosed synthetic people or generic articles published primarily to occupy search results.
The service test is straightforward: does the customer receive something more useful, relevant or accessible, or is the company simply lowering its own production cost?
The strongest applications achieve both. They reduce internal friction while improving the customer’s ability to understand, compare, customise or use the product.
The Competitive Advantage Is the System Around the Model
Most companies will eventually have access to broadly comparable AI capabilities. The durable advantage will not be the ability to generate a competent image, paragraph or presentation. It will come from proprietary customer insight, distinctive brand judgement, well-organised content, efficient approval systems and employees capable of recognising a weak result.
AI can make a creative organisation faster, but speed magnifies whatever system already exists. A company with clear positioning and disciplined decision-making can test more ideas and reach customers more effectively. A company with an indistinct brand and confused processes can produce greater quantities of indistinct work.
The commercial case for creative AI is therefore neither “replace the creative department” nor “give everyone a chatbot”. It is to identify the parts of creative work where cost and delay prevent the company from learning, adapting or serving customers properly, then redesign those parts around measurable results.
The businesses that benefit most will not be those publishing the largest volume of synthetic content. They will be those using AI to make better decisions before expensive commitments are made, give customers genuinely more relevant service and return human attention to the work where judgement creates the greatest value.
