Kholoud Hussein 

For years, the artificial intelligence race was largely conducted in English. The world’s most powerful foundation models could write, code, summarize, and reason across a growing range of tasks, but their performance often weakened when they encountered the realities of Arabic: its grammatical complexity, regional dialects, cultural references, code-switching, and the enormous gap between formal written Arabic and the language people actually speak.

Saudi Arabia increasingly sees that gap not simply as a linguistic problem, but as a technology and investment opportunity.

The Kingdom is now emerging as one of the most ambitious markets for Arabic artificial intelligence, with government-backed companies, startups, global technology groups and large enterprises building different layers of an Arabic AI ecosystem. At the center of that effort is the development of proprietary language models that can understand Arabic on its own terms rather than treating it as a translation layer on top of English-centric systems.

The scale of the opportunity is reflected in the capital flowing into the wider Saudi AI ecosystem. Saudi Arabia’s National Strategy for Data and AI targets $20 billion in local and foreign investment and at least 300 active AI startups by 2030. Meanwhile, the country’s broader AI market is estimated at $2.14 billion in 2025 and projected to reach $16.9 billion by 2032, according to MarketsandMarkets.

Yet the more important question is not how much Saudi Arabia spends on AI.

It is whether the Kingdom can turn that capital into proprietary intellectual property, globally competitive companies and models that become infrastructure for the Arabic-speaking digital economy.

The Arabic gap is bigger than translation

Arabic presents a distinctive challenge for AI developers.

The language exists simultaneously in Modern Standard Arabic, classical forms, and dozens of spoken dialects. A Saudi user, for example, may switch between formal Arabic, Saudi dialect, English terminology, and industry-specific language within the same conversation.

For AI models trained predominantly on high-quality English data, this creates a structural disadvantage.

Research cited by Arab News estimates that only around 15% of Arabic text available online is clean enough for large-language-model training, compared with more than 50% for English. Developers therefore face not only a shortage of data but also a shortage of high-quality, correctly labelled and culturally representative data.

That creates an opening for companies willing to build the data layer themselves.

This is where Saudi startups and emerging technology companies are becoming important. Their advantage is not necessarily the ability to spend hundreds of millions of dollars training a general-purpose model. It is their proximity to Arabic users, enterprise data, dialects and specific commercial problems.

The resulting market is therefore developing on several levels.

At one end are foundation models such as HUMAIN’s ALLAM, developed in Saudi Arabia for Arabic-first use cases. At another are companies such as Riyadh-headquartered Intella, which has built proprietary speech technologies and small language models designed around Arabic dialects and enterprise applications. Between them sits a growing ecosystem of startups developing voice agents, vertical models, translation systems, enterprise copilots and domain-specific AI applications.

Together, they are attacking the Arabic AI problem from different directions.

From consuming AI to owning the model

The launch of HUMAIN in May 2025 marked a major change in Saudi Arabia’s approach.

The PIF-owned company was established to operate across the entire AI value chain, from data centers and cloud infrastructure to advanced models and applications. Its flagship ALLAM model is positioned as one of the world’s most powerful multimodal Arabic large language models.

HUMAIN later launched HUMAIN Chat, powered by ALLAM 34B. The company said the model was trained on more than 500 billion Arabic tokens and refined using hundreds of domain experts and evaluators. The system was designed to support Arabic and English while incorporating regional and cultural context.

The significance extends beyond having another chatbot.

A proprietary foundation model gives Saudi Arabia control over an important part of the technology stack: the data, model weights, training processes, deployment environment, and intellectual property.

That matters particularly for government, financial services, healthcare, energy and other regulated industries where data residency, security and customization can be as important as raw model performance.

It also changes the economics of the market.

Instead of paying indefinitely for access to foreign foundation models, Saudi companies can increasingly build products on locally developed models or adapt them to specific requirements.

Startups are attacking the problem from the bottom up

The startup opportunity is not necessarily to compete head-on with the largest global AI laboratories.

It is to solve the problems those laboratories have historically struggled to solve.

Intella is one example.

The Saudi-headquartered company, founded in 2021, focuses on Arabic speech intelligence rather than trying to become another general-purpose ChatGPT competitor. Its technology covers speech-to-text, text-to-speech, analytics and industry-specific small language models, with support for more than 25 Arabic dialects. The company says its proprietary speech-to-text technology has reached 95.73% accuracy.

Its commercial proposition illustrates where startups can create value.

A bank does not necessarily need the world’s largest LLM. It needs an AI system that understands how its customers actually speak, recognizes local expressions, complies with data requirements, and can connect those conversations to a banking workflow.

The same logic applies to telecom operators, government agencies, insurers and retailers.

In September 2025, Intella raised $12.5 million in a Series A led by Prosus, taking its total funding to $16.9 million. The round included Saudi investors such as Wa’ed Ventures and Hala Ventures and was intended to support R&D, product development and regional expansion.

The figure is modest compared with infrastructure investments in Saudi AI, but strategically important.

It demonstrates that private capital is beginning to finance the specialist layers that make Arabic AI commercially useful.

The corporate-startup model is becoming more important

Saudi Arabia’s Arabic AI ecosystem is also developing through an unusual combination of startups, sovereign capital and global technology companies.

Google Cloud and PIF announced a $10 billion partnership to advance an AI hub in Saudi Arabia, with the initiative involving HUMAIN. The partnership includes research into Arabic-language models and Saudi-specific AI applications, including work to enhance the Arabic capabilities of Google’s Gemini models using additional Arabic datasets.

The model is significant because it illustrates how international technology companies can provide capabilities that startups and local companies may struggle to build independently: compute, cloud infrastructure, specialized chips, model-development platforms, and global distribution.

In return, Saudi Arabia offers something equally valuable: access to a rapidly digitizing market, large enterprise customers, government use cases, capital and a concentrated pool of Arabic data and talent.

The relationship is increasingly moving beyond conventional technology procurement toward co-development.

In August 2026, Microsoft and HUMAIN announced a long-term strategic collaboration under which ALLAM models are planned to become available through Microsoft Foundry and the Microsoft 365 Copilot ecosystem. The partnership also brings HUMAIN AI specialists together with Microsoft’s Forward Deployed Engineers to develop and deploy Arabic-language AI solutions for organizations.

This is strategically important for Saudi startups as well.

A local model becomes significantly more valuable when it can be distributed through a global enterprise platform.

The same principle is visible in AWS’s expanding relationship with HUMAIN. At LEAP 2026, AWS announced plans to make ALLAM available through Amazon Bedrock, while expanding infrastructure capacity for AI workloads in the Kingdom.

The emerging architecture is therefore not simply “Saudi versus Silicon Valley.”

It is increasingly a partnership model in which Saudi companies own local intelligence and context while international technology companies provide global infrastructure, platforms and distribution.

Aramco adds another dimension

Saudi Arabia’s AI ambitions are also being accelerated by its largest corporate institution: Aramco.

The company has developed its own industrial large language model, trained on decades of proprietary Aramco data. The model is designed for applications ranging from analyzing drilling and geological information to forecasting refined-product markets. Aramco has said its decision to develop its own generative AI capability reflects the need to capture the benefits of AI while managing technology and data risks.

That approach could become particularly influential in the next phase of Arabic AI.

The most valuable models may not be the largest models.

They may be the models that understand a particular industry better than a general-purpose system does.

Energy, banking, government, healthcare and legal services all contain large amounts of proprietary information that cannot simply be uploaded to a public AI platform.

This creates a market for smaller, highly specialized models trained or fine-tuned on proprietary datasets.

For startups, that is a much more realistic opportunity than attempting to reproduce the enormous capital expenditure of frontier-model developers.

The investment story is much larger than LLM funding

One of the biggest challenges in measuring the Arabic LLM opportunity is the lack of a separate investment category.

Saudi Arabia does not publish a single figure showing how much has been invested specifically in Arabic foundation models. Much of the disclosed capital is instead bundled into broader AI infrastructure, cloud computing, data centers, chips, venture capital, and AI applications.

The numbers nevertheless show the scale of the ecosystem being constructed.

Saudi Arabia announced nearly $15 billion of investments and agreements at LEAP 2026 across AI infrastructure, data centers, cloud computing, technology manufacturing and venture capital. AWS alone announced a planned investment of more than $5.3 billion in its Saudi cloud infrastructure region, while other agreements covered major data-center and AI projects.

Aramco has also committed approximately SAR9.9 billion ($2.6 billion) in cash contributions to HUMAIN in 2026, alongside the transfer of AI assets, according to Aramco.

Earlier commitments included the $10 billion PIF-Google Cloud AI hub and Saudi Arabia’s $1.5 billion commitment linked to AI-chip company Groq.

These figures should not be added mechanically: some represent partnerships or multi-year commitments, while others span infrastructure and AI rather than Arabic models specifically.

But collectively they illustrate the capital intensity of the market Saudi Arabia is attempting to build.

At the startup level, the numbers are smaller but equally revealing.

Wa’ed Ventures has a $500 million technology-focused fund and had deployed approximately $270 million across more than 75 companies by 2024. Its mandate includes AI and other deep technologies, while requiring certain international technology investments to localize operations in Saudi Arabia.

The Kingdom is therefore developing both sides of the capital equation: large strategic investment for infrastructure and smaller venture capital for experimentation and commercialization.

How much more money could follow?

The next investment cycle is likely to move from infrastructure toward monetization.

Saudi Arabia’s National Strategy for Data and AI targets $20 billion in local and foreign investment and at least 300 active AI startups by 2030. Meanwhile, the country’s AI market is forecast to grow from $2.14 billion in 2025 to $16.9 billion by 2032.

The enterprise AI segment alone is forecast to rise from $810.6 million in 2024 to more than $5.3 billion by 2030, representing a 37.8% compound annual growth rate.

These forecasts suggest that future capital will increasingly follow commercially proven applications.

That could benefit Arabic-model startups because the value of an LLM is ultimately determined by what sits on top of it.

A model that understands Saudi dialects becomes more valuable when it powers a bank’s customer service. A government-specific model becomes more valuable when it automates document processing. An industrial model becomes valuable when it improves maintenance, engineering or energy efficiency.

The transition is therefore likely to be from model building to model commercialization.

The next frontier: Arabic AI agents

The most important development over the next few years may not be larger Arabic LLMs, but more specialized AI agents.

Global AI development is already moving from systems that generate answers to systems that can perform tasks. In Saudi Arabia, Arabic-first agents could combine language models with enterprise databases, government systems, CRM platforms, and workflow tools.

That creates a much larger commercial opportunity.

An Arabic AI agent for a bank could understand a customer’s dialect, verify information, retrieve account data, and complete a transaction.

A government agent could interpret Arabic documents, identify regulatory requirements, and route applications.

An industrial agent could combine technical manuals, sensor data and historical operational information to assist engineers.

The underlying foundation model is only one component.

Data, security, retrieval systems, workflow integration, and domain expertise increasingly determine whether the technology produces economic value.

This is precisely where startups can complement the large capital providers.

The real competition will be over data and talent

The biggest constraint on Saudi Arabic AI may eventually cease to be funding.

It could be data and people.

Developing an Arabic model requires enormous quantities of high-quality training data, but collecting that data raises questions about copyright, privacy, ownership, dialect representation and governance.

Saudi Arabia has an advantage in that it can combine government datasets, corporate information, Arabic digital content and local linguistic expertise. But access to data does not automatically make it usable for training.

The country therefore needs a broader data economy alongside its AI economy.

Talent will be equally critical.

HUMAIN said its ALLaM team included more than 120 AI specialists, including 35 PhD holders, while Aramco has committed to training more than 6,000 AI developers through collaborations involving institutions such as Imperial College, Caltech and KAUST.

For startups, the competition for this talent could become intense.

The next generation of Arabic AI companies will require machine-learning researchers, computational linguists, data engineers, Arabic-language experts, cybersecurity specialists and enterprise software developers.

From linguistic gap to economic infrastructure

Saudi Arabia’s Arabic AI push is ultimately about more than language.

It is an attempt to establish ownership over a layer of digital infrastructure that could sit underneath the region’s future economy.

The opportunity is significant because Arabic is spoken by hundreds of millions of people, while businesses and governments across the region are accelerating digital transformation.

But building a competitive Arabic LLM does not automatically create a successful technology business.

The coming years will test whether Saudi companies can turn models into recurring revenue, whether startups can scale beyond government contracts, whether proprietary data can become a defensible advantage, and whether international partnerships create technology transfer rather than simple dependence on foreign infrastructure.

The strongest companies are likely to occupy the space between these worlds.

They will understand Arabic deeply enough to solve problems global models struggle with, but build products sophisticated enough to compete internationally.

That is where Saudi startups have a potentially decisive role.

The Kingdom does not need to build every component of the global AI stack itself. It needs to identify the layers where local knowledge creates an enduring advantage — Arabic language, regional data, industry expertise, sovereign deployment and culturally relevant applications — and build globally competitive businesses around them.

The first phase of Saudi Arabia’s AI strategy was about attracting infrastructure and capital.

The next phase is about turning that infrastructure into intellectual property.

And the ultimate test will be whether Arabic AI becomes something Saudi Arabia merely helped develop — or an industry in which Saudi companies own the models, data, applications, and businesses that serve the next generation of the Arabic-speaking digital economy.

Ghada Ismail

When a company wants to grow, buying another business can sometimes be easier than building something from scratch. Instead of spending years developing a new product, entering a new market, or hiring a specialized team, a company can acquire a smaller business that already has what it needs.

This is the idea behind a bolt-on acquisition.

A bolt-on acquisition is when an established company buys a smaller business and adds it to its existing operations. The acquired company usually brings something specific to the table, such as new technology, customers, talent, products, or access to a particular market.

The focus is not necessarily on changing the entire business. It is about adding another useful piece to what is already there.

How does a bolt-on acquisition work?

It usually starts with a company identifying an area where it wants to grow.

Take a software company that has a large customer base but does not offer cybersecurity services. Rather than spending years developing those services internally, it could acquire a smaller cybersecurity company that already has the technology, employees, and customers.

The buyer can then add those capabilities to its existing business.

The acquired company may keep its own name and management team, or it may be fully integrated into the larger company. That depends on the businesses involved and what the buyer believes will work best.

What matters is that the acquisition fills a specific gap or creates an opportunity for further growth.

Why do companies choose bolt-on acquisitions?

Speed is one of the biggest reasons. Building a new product or entering a new market takes time. Companies need to hire people, develop products, find customers, and build relationships. Buying an established business can shorten that process considerably.

Bolt-ons can also give companies access to new markets. A business looking to expand into another country, for example, could acquire a local company that already understands the market and has an established customer base.

Technology and talent are another major attraction. In areas such as artificial intelligence, fintech, and software, smaller companies often develop highly specialized products or expertise that larger businesses may want to bring in quickly.

There can also be financial benefits. Once the businesses are combined, the buyer may be able to share infrastructure, eliminate overlapping costs, and introduce the acquired company's products to a much larger customer base.

How is it different from a major acquisition?

Not every acquisition is a bolt-on.

A large or transformational acquisition can significantly change the direction of a company. It could involve buying a major competitor, entering a completely new industry, or acquiring a business that becomes a central part of the company's future.

A bolt-on is usually more focused.

The buyer already has an established business and is looking for smaller companies that can strengthen it in specific areas. In simple terms, it is less about rebuilding the business and more about adding to it.

That can make bolt-ons easier to manage than very large deals, although integration still requires careful planning.

What is the challenging part here?

Smaller acquisitions are not automatically easy acquisitions.

One of the biggest challenges is making the two businesses work together. Different company cultures, technology systems, and ways of working can create problems if they are not handled properly.

There is also the question of price. A company may look like a perfect fit, but if the buyer pays too much, the deal may not generate the expected returns.

Then there are the promised synergies. Buyers often expect an acquisition to increase sales or reduce costs, but those benefits do not happen automatically. They need to be planned and executed.

To Wrap Things Up…

For companies with ambitious growth plans, bolt-on acquisitions can offer a practical way to expand without making one huge bet. Instead of spending a large amount on a single transformational deal, a company can make several smaller acquisitions over time. Each one can add something different, whether that is technology, customers, talent or geographic reach.

This approach is particularly common among private equity-backed companies. An investor may acquire a larger “platform” business and then use a series of bolt-on acquisitions to expand it.

Ultimately, a successful bolt-on acquisition comes down to one simple question: Does the smaller company add something the buyer genuinely needs?

If the answer is yes, and the two businesses can work well together, a bolt-on can be a relatively straightforward way to accelerate growth without starting from zero.

Noha Gad

Offering a strong salary is no longer enough to attract and retain top talent in today’s competitive job market, as employees increasingly look beyond base pay to evaluate the full value of a job offer, and that is where fringe benefits come in.

Fringe benefits are forms of non-wage compensation provided to employees in addition to their regular salary, including cash equivalents, property, services, or other privileges, such as health insurance, retirement contributions, company cars, tuition assistance, or paid time off.

Although they are viewed as extras, fringe benefits play a pivotal role in modern compensation packages for both employers and employees. For employers, they serve as powerful tools to enhance employer branding, boost employee morale and productivity, and gain tax advantages when structured correctly. For employees, they can significantly increase the real value of their compensation while improving financial security, health, and work-life balance.

What are fringe benefits?

Fringe benefits are additional remuneration that employees receive from their employers. They are designed to enhance the overall employee experience and provide added value beyond monetary compensation, serving as incentives that attract top talent and boost employee morale and satisfaction. By offering these extras, companies aim to create a positive work environment where employees feel valued and motivated.

Fringe benefits encompass a wide range of non-wage compensation that add another layer of appeal to any employment package, while creating a supportive workplace culture where employees feel appreciated for their hard work and dedication without only relying on financial remuneration.

Examples of fringe benefits

There are various types of fringe benefits that companies can offer to their employees, including:

• Health insurance: Many employers offer comprehensive health insurance plans, covering medical, dental, and vision expenses for employees and their dependents.
• Retirement plans: Companies may contribute to retirement savings accounts or offer pension schemes to ensure financial security for employees after they retire.
• Paid time off: In addition to statutory holidays, companies often provide vacation leave, sick leave, personal days off, or paid parental leave to support employee well-being and family needs.
• Employee Assistance Programs (EAP): These programs offer confidential counseling services for employees dealing with personal issues such as stress management or substance abuse problems.
• Education reimbursement: Some organizations support continuous learning through tuition reimbursement programs or scholarships for further education or professional development courses.
• Wellness programs: These initiatives promote employee health through gym membership discounts, wellness challenges, on-site fitness classes, or access to mental health resources.

Why do companies offer fringe benefits?

Offering fringe benefits gives companies a competitive edge in the job market, helping them to attract and retain top talent. Some advantages of providing fringe benefits include:

• Increasing employee satisfaction. These benefits make employees feel valued and appreciated, leading to higher job satisfaction and making them more likely to be loyal and committed to their work.
• Improving morale and motivation. Through fringe benefits, employers show they prioritize employees’ well-being, thereby boosting their morale and motivation.
• Attracting top talent: A comprehensive package that includes attractive fringe benefits can be a major draw for highly skilled professionals.
• Enhancing productivity: Offering fringe benefits helps create a positive work environment where individuals are motivated to excel. 
• Reducing turnover: Investing in fringe benefits can help reduce employee turnover rates as individuals are less likely to leave an organization that provides valuable perks beyond salary alone.
• Saving costs for employees: Some fringe benefits, like health insurance or retirement plans, may come with cost savings for employees compared to purchasing these services individually.

To sum up, fringe benefits have evolved from optional extras into a core component of strategic compensation, enabling employers to differentiate their offers, strengthen retention, and build a culture where employees feel genuinely supported.

These non-wage benefits can materially raise the real value of employees’ compensation while improving health, financial security, and work-life balance. For employers, a well-designed mix, aligned to workforce needs and local tax rules, can drive morale, productivity, and long-term cost efficiency.

Fringe benefits become a genuine investment in employees and a real advantage when it comes to winning and keeping great talent. For employers, all what they need to do is to choose benefits that truly fit their team and their goals, understand the full cost and tax picture, explain them in plain language, and revisit them often to see how they stack up.

Kholoud Hussein 

Artificial intelligence is moving beyond consumer applications such as chatbots, image generators, and personal productivity tools. As businesses shift from experimenting with AI to integrating it into core operations, a new category is gaining prominence: Enterprise AI.

At its simplest, Enterprise AI refers to the use of artificial intelligence within organizations to automate processes, analyze data, support decision-making, and improve operational efficiency. Unlike consumer AI, which is designed primarily for individual users, Enterprise AI addresses the more complex requirements of businesses, including data security, governance, integration, scalability, compliance, and measurable returns on investment.

What is Enterprise AI?

Enterprise AI encompasses AI-powered technologies deployed across functions such as finance, human resources, sales, marketing, customer service, cybersecurity, supply chains, and operations.

A bank, for example, may use AI to detect suspicious transactions, assess credit risks, automate customer support, and analyze financial data. A retailer could use AI to forecast demand, optimize inventory, and personalize customer recommendations, while a manufacturer could deploy it to predict equipment failures and reduce downtime.

The key distinction is that Enterprise AI is not simply about introducing an AI model into a company. It involves integrating AI into existing business systems and workflows to generate measurable business outcomes.

This makes integration one of the defining characteristics of Enterprise AI. Even a sophisticated AI model has limited business value if it cannot securely access relevant company data or interact with systems such as enterprise resource planning, customer relationship management, accounting, and supply-chain platforms.

From experimentation to infrastructure

The rapid development of generative AI has changed how companies approach the technology. Many businesses initially experimented with publicly available AI tools to generate content, summarize documents, or improve employee productivity.

The next stage is more complex: moving AI from an individual productivity tool to an integrated component of business infrastructure.

This transition is creating demand for technologies that connect AI models with proprietary company data and existing business applications. It is also increasing the importance of cybersecurity, data privacy, regulatory compliance, and human oversight.

As a result, companies are increasingly looking beyond the AI model itself and considering the infrastructure required to deploy AI securely and effectively at scale.

Where startups fit in

This shift creates a significant opportunity for startups.

Large technology companies may provide foundational AI models and cloud infrastructure, but startups can build specialized applications on top of these technologies to address specific enterprise problems.

Businesses often do not need a general-purpose AI system. They need a solution that understands a particular industry, workflow, or operational challenge.

A startup could, for example, develop an AI platform for insurance claims, legal document analysis, financial compliance, procurement, or logistics. By focusing on a specific problem, it can develop specialized workflows, integrate with existing enterprise systems, and potentially demonstrate a clearer return on investment.

This has contributed to the emergence of vertical AI startups—companies applying AI to specific industries rather than attempting to serve every type of customer.

Why Enterprise AI can be attractive to startups

Enterprise customers may be willing to pay more for technology that can reduce costs, increase productivity, accelerate revenue, or mitigate risk. This creates an opportunity for startups to build business-to-business AI products with higher contract values than many consumer applications.

However, selling to enterprises also raises the barriers to entry. Startups may need to pass security assessments, demonstrate regulatory compliance, integrate with existing systems, and convince multiple decision-makers before securing a contract.

Technical capability alone is therefore not enough. Successful Enterprise AI startups need to combine AI expertise with enterprise sales, cybersecurity, data governance, product integration, and a strong understanding of customer workflows.

The importance of proprietary data

Data is another critical component of Enterprise AI.

Companies hold large volumes of proprietary information that can make AI applications more relevant to their specific environments. Customer records, internal documents, transaction histories, operational data, and industry-specific knowledge can all support more specialized AI solutions.

This creates an opportunity for startups to build products around enterprise-specific data and workflows, rather than competing solely on the performance of an underlying AI model.

At the same time, enterprises increasingly expect clear controls over data access, storage, model training, and privacy, making responsible data management a central part of the Enterprise AI proposition.

The next opportunity for startups

The Enterprise AI opportunity extends well beyond building another chatbot. Startups can create value across the AI ecosystem, from data management and security to specialized applications, workflow automation, and AI agents.

AI agents are particularly significant because they can move beyond generating responses to performing sequences of tasks. An enterprise agent could retrieve information, analyze it, update a business system, and trigger a workflow with limited human intervention.

For startups, the central question is therefore not simply "Where can we use AI?" but "Which expensive, repetitive, or complex business process can AI fundamentally improve?"

That distinction captures the essence of Enterprise AI. Its value lies in transforming artificial intelligence from a standalone technology into a practical business capability that can be integrated into workflows, measured through business outcomes, and scaled across organizations.

For startups, this represents a growing opportunity—but also a higher bar for execution. Winning in Enterprise AI will increasingly depend not only on developing powerful AI technology, but on understanding a business problem deeply enough to turn that technology into a reliable, secure, and economically valuable solution.

Ghada Ismail

In this second part, Abu Zannad turns to the resource startups actually have plenty of: creativity. He explains why “out-noticing” the competition matters more than out-spending them, and why so many founders confuse visibility, reputation, and meaning when they talk about “building a brand.”

How can startups use creativity as a competitive advantage when they cannot compete with larger companies on advertising budgets, resources, or brand recognition?

I think we first need to stop treating creativity as incidental, as this magical thing that occasionally happens when a talented person walks into a room. Creativity is becoming a much more important competitive capability precisely because AI is making so many other capabilities abundant.

Today, almost everyone can produce more content, more variations, more designs, more headlines and more analysis, faster and cheaper than ever before. So producing more is becoming less interesting. The competitive advantage increasingly lies in seeing something other people did not see.

I often describe it as the difference between information and insight. Two companies can have access to exactly the same data and come to completely different conclusions. Same data. Different eyes. That difference is human judgement.

And I don’t think insight has to be left to luck. There are conditions that make it more likely. Experience gives you patterns. Curiosity makes you notice what does not fit. Scepticism stops you accepting the first explanation. Contradictions reveal where reality is behaving differently from the category’s assumptions. Connections allow two things that normally live separately to collide.

Sometimes even constraint helps. I call that creative desperation: when you genuinely cannot solve the problem in the conventional way, you are forced to find another path. That is why startups may actually have an advantage. A large incumbent can often buy another media plan. A startup cannot. It has to notice something the incumbent has stopped noticing.

Look at the extraordinary group of younger businesses emerging around us:

Dollar Shave Club did not beat the shaving establishment by producing a more expensive shaving commercial. It understood internet humour and attacked the seriousness of the category.

Liquid Death looked at bottled water and asked why water had to behave like bottled water at all. It borrowed from punk, heavy metal and entertainment culture.

PRIME understood that creator communities themselves could become an extraordinary distribution system.

Crumbl turned cookies into something closer to sneaker drops; weekly anticipation, scarcity, reviewing and participation.

Sleep or Die looked at the soft, calming visual language of the sleep category and contradicted it completely.

And Dubai Chocolate may be one of the most fascinating cases of all. Someone created an unusually sensory product: “the crack of the chocolate, the colour of the pistachio, the texture of knafeh and a platform discovered that people could not stop watching it”. The algorithm accelerated the phenomenon; it did not originate the human fascination.

I think we should stop treating cases like these as amusing stories about things that “went viral.” They are evidence. We are watching something close to a new applied science of cultural creativity develop in front of us.

Every platform is producing an enormous live laboratory of human behaviour. Every unexpected breakout gives us something to study. What was the human tension? What cultural code did the brand recognize? What category convention did it violate? What community carried the idea? What made somebody want to participate rather than merely watch? What behaviour did the platform reward? What made the idea travel from one subculture into another?

Those are not questions only for advertising people anymore. They are questions for founders, anthropologists, behavioural scientists, strategists and technologists. And over time, we can begin building frameworks around them; not formulas for producing virality, because culture will never be that obedient, but better places to look for the unexpected.

That distinction matters. Creativity is not a formula. But neither is it magic. We can study it. We can develop our intuition. We can accumulate cases. We can recognize patterns. We can learn the grammar of a platform, a category, a culture or a subculture; and then have the courage to break that grammar when the human insight tells us to.

This, to me, is where AI becomes enormously useful. Let the machine search wider. Let it retrieve more cases, make more connections, generate more possibilities and accelerate experimentation.

But the human still has to ask: Which one matters? Which contradiction is interesting? Which observation is merely strange… and which one reveals something genuinely human? Which idea deserves to exist?

Because AI can increasingly generate ten thousand possibilities. The scarce capability is knowing which possibility is worth pursuing.

So my advice to startups would be: don’t try to out-produce the large companies. You probably can’t. And increasingly, there is little advantage in doing so anyway. Out-notice them. Out-understand them. And then use creativity to turn what you noticed into something the culture cannot ignore.

What do you think startups misunderstand most about building a brand: is it about visibility, reputation, or creating an identity people want to associate with?

I think what startups misunderstand most is the word brand itself.

They often think the sequence is: Build the product. Acquire customers. Grow. And when we become big enough, we will “do the brand.” Usually that means a new logo, a brand book, perhaps a large campaign.

But the uncomfortable truth is that you are building the brand from the first day whether you intend to or not. The first product experience builds it. The first customer complaint builds it. The way your founder speaks builds it. The price builds it. The packaging builds it. The people who choose you build it. The things you repeatedly say, and the things you repeatedly do, build it. So visibility, reputation and identity are not really three competing answers. They are three different layers.

Visibility means: I know you exist.

You can buy visibility. You can hack it. You can go viral and acquire enormous visibility almost overnight. But visibility is not a brand. We are surrounded today by things that became very visible and disappeared six months later.

Reputation means: I have learned what to expect from you.

You deliver. The product works. You keep your promises. There is consistency between what you say and what actually happens. Reputation takes longer because it has to survive contact with reality. And then there is something more interesting.

Meaning.

At some point, the strongest brands begin to signify something beyond the immediate utility of the product. Choosing the brand says something. Sometimes it says something to other people. Sometimes, more importantly, it says something to ourselves.

That is very close to the argument I make in AdEntity. Modern advertising became powerful because it taught objects to carry meaning. A watch stopped being only an instrument for telling time. A car was not only transportation. A pair of shoes was not only protection for the feet. Commercial objects became signals through which ambition, taste, rebellion, belonging, care or achievement could become socially legible.

And AdEntity does not argue that brands invented those desires. It argues that the surrounding system; the brand, product, image, celebrity and media environment… helped teach people how those desires could be recognized.

That is why I would hesitate to tell a founder, “Create an identity people want to associate with.” It is almost right. But it can lead to another mistake: inventing a beautiful brand personality with no relationship to the actual business.

Meaning has to be earned through product truth.

If Liquid Death behaved like a rebellious entertainment brand but the product, packaging and every interaction reverted to conventional bottled-water behaviour, the mythology would eventually collapse.

If Apple talks about creativity but produces experiences that feel careless, the symbolism weakens.

A brand cannot indefinitely advertise a meaning that the business itself does not substantiate. And this is where I think startups face a particularly modern trap. Startups live inside dashboards: ‘CAC. ROAS. Conversion. Cost per click. Retention. Downloads. Funnels’.

These numbers matter enormously. I would never advise a founder to ignore them. But because they are visible every morning on a dashboard, they begin to acquire psychological authority. What we can measure immediately starts to look more important than what is accumulating slowly.

And brand accumulates slowly. Memory accumulates. Familiarity accumulates. Trust accumulates. Distinctive assets accumulate. Meaning accumulates. This is why performance marketing is so seductive. You spend today and something happens tomorrow.

Brand building is more like compound interest. For a while, it can look as though very little is happening. And then one day people search for your name instead of the category. They recommend you without being paid. They recognize you before they see the logo. They forgive you a small mistake because there is accumulated trust. They consider you before the performance ad arrives. They may even pay slightly more because the alternative does not feel equivalent.

That is an economic asset, not a communications indulgence.

Airbnb gave us a fascinating demonstration of this. When the company dramatically reduced marketing during the pandemic, traffic recovered to roughly 95% of its 2019 level before marketing expenditure fully resumed. By the fourth quarter of 2020, more than 90% of traffic was direct or unpaid. Brian Chesky’s conclusion was essentially that Airbnb had become culturally established enough that the brand itself was generating demand.

That is what founders should aspire to. Not necessarily becoming a verb. But getting to the point where every customer does not have to be rented again from an advertising platform. Because if every sale requires another paid impression, another promotion and another retargeting message, you may have built an efficient acquisition machine. You have not necessarily built a brand.

There is another problem that optimization culture creates for startups: they change too much. New headline. New proposition. New design. New tone. New campaign. New audience. New creative every week because something performed 4% better. Experimentation is essential for discovering what works. But once you discover something valuable, brand building requires the opposite capability: the discipline to repeat it.

Memory needs consistency. And let’s not confuse consistency with repetition.

The Ehrenberg-Bass work on distinctive assets is useful here. Colours, sounds, shapes, characters, packaging and other recognizable cues only become assets when people repeatedly learn to associate them with one brand. They are built and protected over time; they do not become distinctive because somebody declared them distinctive in a brand guideline.

So perhaps I would give founders a very simple architecture: Be visible enough to enter the mind. Be good enough to earn a reputation. Be consistent enough to become remembered. Be meaningful enough to stand for something.

And make sure the product continuously earns the story you are telling.

Because a brand, in the end, is not the campaign. It is not the logo. It is not the number of followers. It is not even what the founder says the company stands for. A brand is the memory and meaning that remain when the advertising disappears. That is what startups should start building from day one.