Watch the full episode below

In the fifth episode of Prompt & Response, the Inetum podcast about AI in practice, host Piotr Mechliński (Data & GenAI Manager EEMEA, Inetum Polska) talks with Przemysław Saniak (Business Development Director) and Andrzej Gumieniak (IoT Practice Leader) about what changed, what still blocks scale, and where AIoT, Artificial Intelligence of Things, is heading next.

TL;DR. IoT is no longer the buzzword, it is the quiet hero running under the floor of modern enterprises. AI is the brain that finally makes that nervous system pay back, as our expert sums up. The winners in 2026 are the ones who combine edge AI, a Unified Namespace for their device data, and serious data governance, and who start from a business case instead of a shopping list of platforms.

Key takeaways from the episode

• IoT is mature: over 20 billion devices were connected by the end of 2025, a number expected to double by 2030.
• Predictive maintenance can cut unplanned stops by up to 70%. Smart buildings can save 20 to 30% on energy and water (U.S. Department of Energy).
• The next step is AIoT: intelligence moves to the edge, the nervous system gets a brain.
• The closed loop is the real shift: not only reading data, but sending decisions back to devices in real time.
• The biggest blockers are legal (Cyber Resilience Act, NIS2), organisational (who owns IoT inside the company) and technical (20 to 30 year old PLCs, vendor lock-in).
• Unified Namespace is becoming the default architecture to unify device data and feed AI agents with context.
• Data governance is not optional anymore. Without a catalogued context, LLMs and agents cannot act reliably on operational data.

From IoT to AIoT: why the name changed

The term AIoT, Artificial Intelligence of Things, is not marketing polish. It captures a real shift: intelligence no longer lives only in the cloud.

Three forces push in the same direction:

1. Smaller devices, bigger compute. Miniaturisation lets companies deploy far more sensors and actuators without expanding footprint.
2. Edge inference. Predictive models are now small and efficient enough to run directly on the device. Less round-tripping to the cloud, faster action, lower cost per decision.
3. Enterprise context for AI. Real-time sensor streams combine with vector databases, retrieval augmented systems and company documentation so that AI agents can reason over operations, not just transactions.

If IoT is the nervous system of the enterprise, AI is the brain. You just combine them to get value out of the whole body.

Andrzej Gumieniak, IoT Practice Leader, Inetum Polska

Where AIoT actually creates value

There is no shortage of sensors. There is a shortage of good answers to one question: what changes in cost, uptime, revenue or speed because of this deployment?

Przemysław groups the real payback into three streams.

Cost optimisation

• Predictive maintenance. Reduction of unplanned stops by up to 70%, with the matching cut in downtime cost.
• Smart buildings. According to the U.S. Department of Energy, smart buildings can save up to 30% on energy and 20 to 30% on water consumption.
• Smart cities. Harder to quantify, but quality of living and operational efficiency improve.

New business models

Connected products make “as a service” models practical: equipment as a service, usage based contracts, proactive SLA management. The trade-off is real: vendors take on more operational responsibility, but they gain deeper, longer customer relationships.

The closed loop

Andrzej adds the stream that separates mature players from beginners. The data flow is no longer one way, bottom-up. Once you trust the models, you let the system send commands back to the devices and steer production in near real time. That is the step where IoT stops being a dashboard and starts being an operating system.

The value is not in knowing faster. It is in reacting faster, and that is where AI earns its keep.

Przemysław Saniak, Business Development Director, Inetum Polska

Edge AI and custom models: why manufacturing is different

Manufacturing rarely fits a generic AI template. Every plant is different, every process has its quirks, every shop floor has its own mix of old and new equipment. That is why AIoT in factories leans toward custom models and edge AI.

Two practical consequences:

• Not all data is gold. Edge devices filter and summarise at the source, so only relevant signals reach the central systems.
• End-to-end streams. Decisions happen where they belong: at the edge for safety and latency, at the core for cross-plant optimisation and learning.

The integration challenges nobody warns you about

Scaling AIoT is worth it, but the path is rarely clean. Inetum IoT experts name three categories of friction.

Legal and regulatory

The EU keeps raising the bar: Cyber Resilience Act, NIS2, sectoral rules on top. If connected devices touch critical infrastructure, compliance is not a backlog item, it is a design constraint.

Organisational

IoT often lands between departments. Is it IT? Is it production engineering? Who is the product owner? Scaling without a clear owner guarantees pilot limbo. A defined operating model is as important as any platform choice.

Technical and legacy

Factories run on equipment that is 20 or even 30 years old. Old PLCs are already wired into SCADA systems, so connectivity exists, but it is brittle and locked to specific vendors. Inetum has seen sites where assets were linked only by SMS, because that was the only usable channel. That is enough to start, but it is not an architecture.

Unified Namespace: the foundation for agentic IoT

The emerging answer on the shop floor is the Unified Namespace (UNS): a single, standardised place where every device publishes its state in real time, in a format everyone in the organisation can understand.

Two ideas underpin it:

Pub-sub over polling

Instead of legacy systems constantly querying PLC registers, devices publish events to a central message bus. Any consumer, SCADA, MES, analytics, AI agents, picks up what it needs.

Shared naming and semantics

Every signal is named and described consistently, so meaning does not get lost as data propagates.

UNS is not only a networking pattern. It is a data governance decision applied to operational technology. That is exactly what agentic AI needs in order to act safely on the factory floor.

Data governance is the new battleground

LLMs and agents cannot reason well on unlabelled streams. For AIoT to deliver, companies need more than telemetry. They need context:

• catalogued data with clear definitions,
• historical records linked to outcomes,
• documentation, runbooks and maintenance history,
• the tribal knowledge that usually lives in senior technicians’ heads.

Feed that into the “brain” of the enterprise and the agent can make informed decisions in seconds, not after a postmortem. Skip it, and you get confident-sounding suggestions that miss the point.

Data governance, in other words, is now a prerequisite for operational AI, not a sideshow for the data team.

What comes next: agentic IoT and Industry 5.0

Looking a step ahead, the guests see two overlapping horizons.

Fully agentic IoT

Once UNS and governance are in place, AI agents can plan maintenance routes, reschedule production, adjust set points and trigger safety interventions, all while keeping a human in the loop where it matters. Inetum is already seeing early versions of this pattern in manufacturing and utilities.

Industry 5.0 and human safety

The collaboration between IoT, robotics, AI and people is getting tighter. Edge AI is a critical enabler for worker safety in harsh environments: detecting risks and stopping machines before a person can be harmed.

Przemysław references Stanley Kubrick: the HAL 9000-style, AI-enhanced operational control centre is no longer science fiction. With mature data foundations and strong cybersecurity, these centres already exist in advanced industrial sites, taking routine load off operators and letting them focus on exceptions.

Where to start if you are beginning today

If you are just starting your IoT (or AIoT) journey, both guests converge on the same playbook.

1. Start with the business case, not the platform. Name the outcome: less downtime, lower energy cost, new service revenue, safer workers. Then work backwards to technology.
2. Make the value measurable from day one. Run short iterations, define what “worth keeping” looks like, and be ready to kill a use case that does not deliver.
3. Think governance in parallel, not afterwards. Data governance, AI governance and a clear operating model need to grow alongside the first MVPs, not after the third one.
4. Do not try to climb Everest in one jump. Connect devices first. Learn to react on their data. Then add AI on top. Big picture in mind, small moves on the ground.
5. Build vs buy is not the first question. With 600+ IoT platforms by 2021 (a number that has since consolidated, including vendors like Google stepping out of the space), and with GenAI now accelerating custom builds, the choice is easier once you have proven value. Not before.

Start with the business case. Otherwise you will build a compressed solution that connects your freezer to your TV, and not much else.

Przemysław Saniak, Business Development Director, Inetum Polska

FAQ

Ready to take the next step?

Your AI journey starts with a conversation. Share your challenge, idea, or question, and our experts will respond with insights and practical guidance. Talk to us

About the speakers

• Piotr Mechliński, Data & GenAI Manager EEMEA, Inetum Polska, host of Prompt & Response.
• Przemysław Saniak, Business Development Director, Inetum Polska.
• Andrzej Gumieniak, IoT Practice Leader, Inetum Polska.

Previous episodes of Prompt & Response

• All episodes
• AI compliance marketing agent
• AI strategy and use case discovery
• Voice of the customer in the age of GenAI
• AI in manufacturing

The key difference does not start with algorithms or platforms. It starts with the nature of data itself.

This episode is especially relevant for leaders and architects who want to move beyond Industry 4.0 dashboards and understand how modern data platforms and generative AI can support real manufacturing processes, not just proofs of concept.

Watch the full episode below

Prompt & Response Webcast #4 – Transcript & Key Insights

Data does not start digital in manufacturing industry

In digital-native industries, data is born digital. Transactions, clicks, and interactions are structured from the very beginning. Manufacturing works in the opposite direction.

Here, data originates in the physical world. Pressure, vibration, temperature, sound, movement – all of these signals must first be measured and translated into digital form. Only then can analytics or AI be applied.

This is why working with manufacturing data is never just about dashboards. It requires engineering understanding and domain knowledge. You need to know how machines behave, how materials age, and how small physical deviations can signal much larger problems.

The challenge is amplified by legacy environments. Many factories still rely on machines and PLC controllers designed decades ago. AI systems must integrate with technology that was never meant to be connected, scalable, or data-driven.

In manufacturing, data doesn’t come from clicks or transactions. It comes from physics. From pressure, vibration, temperature.
If you don’t understand the physical process behind the data, no AI model will save you.

Marek Czachorowski
(AI & Data Practice Leader, Inetum)

Repeatable processes create real analytical potential for manufacturing companies

Despite these constraints, manufacturing has a powerful advantage: repeatability. Production lines, assembly steps, quality checks, and logistics flows run every day, often thousands of times.

Where repetition exists, patterns emerge. Deviations become visible. This is why manufacturing, when approached correctly, is such a strong candidate for analytics and AI.

The value goes beyond the production line itself. Supporting processes like supply chain management or logistics generate equally important signals. When analyzed together, they reveal how a factory truly operates – not just how it was designed to operate.

Products continue to generate data after leaving the factory

Modern manufacturing no longer ends at production. Products themselves become continuous sources of data once they are in use.

Predictive maintenance is often called the holy grail of manufacturing AI. Not because it’s technically impressive, but because even small improvements can save millions when downtime is critical.

Sebastian Stefanowski
(Chief Architect)

Sensors embedded in machines, vehicles, or equipment allow manufacturers to understand how products behave in real conditions. This enables a shift from reactive maintenance to predictive maintenance – identifying issues before failures occur.

In highly complex environments, such as aviation, even small improvements have a massive impact. Unplanned maintenance is expensive, downtime is critical, and better prediction translates directly into safety improvements and measurable cost savings.

What distinguishes successful cases is rarely the sophistication of the model. It is the maturity of the data foundations behind it.

Quality control moves from inspection to prevention

AI also changes how quality is managed inside the factory. Instead of discovering defects at the end of production, machine learning models can identify early signals that something is going wrong.

Subtle anomalies appear long before a defect becomes visible. This allows production to stop earlier, avoid unnecessary cost, and focus resources where they actually matter.

The result is not just better quality, but higher throughput and lower waste at the same time – a combination that traditional approaches struggle to deliver.

Use AI to your advantage: from collecting data to understanding it

For many organizations, Industry 4.0 initiatives focused on connectivity and dashboards. Sensors were installed, data was collected, charts were built. What is changing now is the role of generative AI.

The focus shifts from collecting data to understanding it. GenAI acts as a bridge between humans and complex technical environments, helping engineers navigate documentation, maintenance history, sensor logs, and operational data more efficiently.

This mirrors patterns discussed earlier in Prompt & Response, including voice of the customer in the age of GenAI, where GenAI serves as a translator between raw data and human decision-making.

The impact is especially visible in knowledge retention. Experienced engineers leave, taking years of tacit knowledge with them. AI does not replace expertise, but it makes existing knowledge accessible and usable at scale.

Start with business reality, not with ai

Most AI initiatives in manufacturing don’t fail because of algorithms. They fail because data foundations and business context were never fixed first

Piotr Mechliński
(Host, Head of AI & Data, EEMEA)

One message remains consistent throughout this episode and the entire series. Successful AI in manufacturing does not start with AI.

It starts with business reality. Clear goals, well-understood processes, and reliable data foundations come first. Only then does it make sense to decide whether machine learning, predictive models, or Generative AI are the right tools — and where they will actually deliver value.

This is the natural continuation of the journey discussed in From AI strategy to execution, where strategy only matters if it can survive contact with reality.

Manufacturing is not harder than other industries when it comes to AI. It is simply different. Organizations that respect that difference are the ones that turn AI from hype into measurable results.

Watch the full episode below, then explore the cleaned and structured transcript.

Prompt & Response Webcast #3 – Transcript & Key Insights

Why customer analytics is more important today than ever

Even if the concept feels old, the context is new. Customers share huge amounts of data with banks, telcos, retailers, streaming platforms – so they naturally expect companies to use it. The frustration of being asked for income details by a bank that already sees your monthly transactions is a perfect example of this gap.

Modern expectations are shaped by:

• constant exposure to personalized digital services,
• dynamic customer behaviour,
• and a market where switching providers takes seconds.

That’s why treating customers the same way they were treated three years ago simply doesn’t make sense any more.

What traditional analytics misses

The classic model focuses on CRM, demographics and historical events. Useful – but not enough. It answers what happened, not why.

Voice of the customer fills this gap by analysing unstructured, emotionally rich feedback coming from:

• social media,
• forums,
• call centre transcripts,
• and even audio signals like tone or sentiment.

This enables companies to understand not only behaviour, but motivation.

“Voice of the customer helps you distil emotions and understand the reasoning behind customer actions – something traditional analytics doesn’t provide.”

This blend of quantitative and qualitative signals becomes especially powerful when used continuously, not as a quarterly exercise.

Real business outcomes: retention, acquisition, experience and cost

Modern customer analytics delivers value across the entire lifecycle:

Retention
Predict who is likely to churn — and act early.

Acquisition
Target the customers most likely to buy, instead of “everyone.”

Experience
Move toward the long-discussed segment of one. Younger generations expect personalized communication, offers and support.

Operational efficiency
Communicating to the wrong customers is costly — not because an email is expensive, but because irrelevant outreach damages loyalty.

Meanwhile, VoC adds something unique: real-time feedback about product perception, service quality and process friction, making it the “cheapest operational audit” a company can have.

Read also: Use data to make informed decisions. Understanding the Importance of Data-Driven Decision Making

Use cases from the field

Pharma: designing what customers actually need

A large pharma company digitized a previously analogue reimbursement process. By analysing comments from patients and healthcare professionals on social platforms, two additional modules emerged as essential:

• direct communication between patient and HCP,
• personalized education on the disease.

These insights weren’t identified internally – they came directly from unstructured market feedback. Including them made the solution more valuable and easier to implement with stakeholders like the Ministry of Health.

Read also: Big Data in healthcare: management, analysis and future prospects for healthcare organizations

Banking: from churn and propensity to behavioural insights

Banks actively use predictions such as churn risk, propensity-to-buy or segmentation models. But the real change comes from combining them with behavioural signals: app usage, transaction patterns, and voice-of-customer input.

Machine Learning finds customer segments automatically, but explaining them used to be difficult. GenAI now helps describe segment behaviour in plain language – making analytics accessible to business teams.

What counts as “Voice of the Customer”?

Practically, it’s the continuous analysis of unstructured data at scale, refreshed daily or even hourly.

Sources include:

• comments, reviews, posts,
• call centre transcripts,
• audio-based sentiment,
• any form of “free speech” not influenced by structured surveys or questionnaires.

What matters most is actionability. Reports without operational follow-up add no value. In a mature setup, organizations use real-time dashboards to guide decisions the same day – not weeks later.

How GenAI transforms VoC and customer analytics

1. Summaries at scale

10,000 comments? GenAI condenses them during your morning coffee.

2. Explaining ML-driven segments

Machine Learning segments are powerful, but often challenging to interpret.
GenAI generates human-readable descriptions that product and marketing teams can use immediately.

3. Personalized communication

With customer attributes in place, GenAI can tailor emails, notifications, offers or scripts – enabling “segment of one” communication in practice.

4. Accessibility for non-technical teams

Sales, CX and marketing teams can finally use advanced analytics without writing code.
This shift accelerates adoption and encourages experimentation.

Low-/no-code tools support this trend, allowing small teams to build prototypes quickly – though enterprise deployment still requires proper governance.

Implementation: where organizations really struggle

The greatest obstacle is rarely the model itself – but the operational process around it.

One bank had excellent analytics, yet results were ineffective. Why?

Analysts exported fresh data to Excel, reworked it for days, prepared presentations, waited for meetings… By the time campaigns launched, customer behaviour had already changed.

The lesson is clear: real-time analytics is useless if wrapped in a slow decision cycle.

Other challenges include:

• unclear ownership of the data and process,
• scattered responsibilities,
• insufficient platform readiness,
• PoCs that never evolve into production.

VoC and advanced analytics require both business and technology ownership – not one or the other.

How to start (and avoid getting stuck in POC mode)

Start small

Choose one product, one market or one business line. Prove value quickly, capture results and use them internally to gain traction.

Design for scale from day one

Even a small PoC should:

• have a clear owner,
• include reporting for decision-makers,
• align with a future architecture that supports 10× more traffic.

Build end-to-end

A PoC isn’t just a model. It should include:

• data ingestion,
• analytics,
• insights,
• dashboards,
• decision outputs.

Only then can leadership understand its value and support further investment.

Why a partner can accelerate results

Customer analytics, VoC and GenAI require a combination of strategy, business understanding, data engineering, modelling and governance.

Not every organization can build all these competencies internally — at least not quickly.

Sometimes the right move is to apply experience gathered from dozens of similar projects, avoid common pitfalls, and accelerate adoption with a partner.

As competition intensifies and fintech-like experiences set new standards, companies that fail to modernize customer analytics risk falling behind.

Ready to take the next step?

Your AI journey starts with a conversation. Share your challenge, idea, or question, and our experts will respond with insights and practical guidance. Talk to us

Introduction

In the second episode of Prompt & Response, host Piotr Mechliński altogether with Wiktor Zdzienicki explore a challenge every organization faces in 2025: how to move from the hype around Artificial Intelligence to a truly business-led AI strategy that scales.

The discussion covers:

• why so many AI pilots fail to grow,
• the essential pillars of a sustainable strategy,
• the risks organizations must anticipate,
• how to select use cases that actually create value,
• introduction of a practical five-hour AI discovery workshop format that companies can apply right away.

Why so many pilots never go beyond proof of concept

Eight out of ten AI pilots never progress to full deployment. This high failure rate often comes down to poor selection of proof-of-concepts, projects chosen at random with no real link to a business problem. Many organizations run pilots simply because the competition does, or because management expects it, not because there is a genuine need.

Another recurring issue is the lack of internal ownership. Without a leader who champions the initiative and aligns stakeholders, even technically successful outcomes remain unused. Teams may deliver working prototypes, but they are treated as isolated experiments rather than embedded business solutions.

Companies also tend to opt for small, risk-free pilots. These projects are easy to deliver but bring no measurable business impact. Instead of targeting use cases that could move KPIs or affect financial performance, organizations pick “safe” ideas and then struggle to prove their value when the pilot ends.

The pillars of an AI strategy

A robust AI strategy is not only about tools and technology. It should connect Artificial Intelligence to the company’s mission and ensure adoption across the organization. Five pillars are especially important:

• Business value: Every AI project must align with corporate goals and deliver tangible outcomes such as revenue growth, customer satisfaction, or efficiency.
• Data: High-quality, accessible, and timely data is the foundation of AI success.
• Culture and organization: People matter most. There must be leaders, champions, and adoption mechanisms to make solutions stick.
• Technology: Models, platforms, and MLOps practices are important, but they should come only after business needs are clear.
• Governance and ethics: In Europe especially, compliance, monitoring, and ethical frameworks cannot be ignored.

Many companies discover that culture and processes, rather than technology, are the main blockers. Tools do not transform businesses, people do.

Where to start: choosing the right use cases

Organizations often wonder how to begin. The answer is not in massive transformation programs or trivial pilots. The best place to start is with quick wins that bring real value.

A practical method is the impact–feasibility matrix, which helps teams identify cases that are both realistic and meaningful. Workshops with business areas such as sales, finance, or operations often generate ideas ranging from small process improvements to ambitious AI-driven innovations.

At the same time, leaders should avoid blindly following hype. Just because chatbots are popular does not mean your company should build one. Sometimes, the real priority is more fundamental, such as fixing reporting or data warehousing. In many projects, “AI” initiatives eventually turn into analytics or BI improvements, because that is where the true value lies.

Also read:

• The rise of AI agents – your new digital coworkers
• AI marketing agents are reshaping how campaigns are built, approved, and delivered
• Revolutionizing coding with the AI coding agent

Common risks to be aware of

Several risks regularly derail AI initiatives:

• Data quality and access: If inputs are incomplete or outdated, the solution quickly becomes irrelevant and may be abandoned within weeks.
• Adoption by employees: Staff who have done their work for years without AI often fear change. Without careful communication and a clear demonstration of benefits, adoption will stall.
• Analysis paralysis: In large or highly regulated companies, months or even years are sometimes spent creating 300-page strategy decks. By the time they are complete, they are already outdated, because technology changes in quarters, not decades.

What AI strategies will look like in the near future

In the next two years, AI strategies will likely become shorter, more iterative, and closer to the business. They will have to account for stricter regulations while constantly refreshing focus, since what is advanced today may be obsolete tomorrow.

Instead of lengthy documents, organizations will move toward lean approaches: iterative roadmaps, actionable guidelines, and strategies that evolve with each new technology wave.

Imagination, leadership, and courage

AI is not just about tools. It is a chance to rethink business models. This requires time for reflection, honest discussion, and leaders who have the courage to break routines. The most dangerous sentence in business remains: “We’ve always done it this way.”

Organizations need leaders who combine imagination with humility. They must be open to testing new paths and willing to accept that ego should never block innovation.

A practical shortcut: the AI discovery workshop

For companies that want to move quickly, a five-hour AI discovery workshop can be the perfect first step. It covers four stages:

1. Inspiration: Industry-specific case studies to show how others apply AI and data.
2. Demystification: A quick overview that clarifies what AI really is, moving beyond the “AI = GPT” simplification.
3. Ideation: Hands-on exploration of a chosen business area, identifying processes and roles where AI can create measurable value.
4. Strategizing: Consolidating insights into the foundation of an AI strategy.

The aim is not to create a final strategy in one session but to ensure organizations leave with more than just a list of use cases. A list is not a strategy. Participants gain a clear sense of what needs attention in the future and practical guidance on how to continue the journey.

Conclusion

The five-hour workshop can be seen as a miniature version of a full AI strategy exercise, or as a demo of how to build one in the long run. It is the first step toward a comprehensive, business-led AI roadmap.

Ultimately, Chat GPT, Google Gemini, Claude Code and other AI tools are only part of the story. The real value comes from combining the capabilities of AI with human creativity, business knowledge, and leadership courage.

If you are considering building an AI strategy, even starting with something as simple as a discovery workshop, reach out to us. We will be glad to explore opportunities and work with you.

Ready to take the next step?

Your AI journey starts with a conversation. Share your challenge, idea, or question, and our experts will respond with insights and practical guidance. Talk to us

Prompt & Response – episode 1 – AI compliance marketing agent

How can AI help marketing teams speed up campaigns while staying compliant with brand rules and regulations?
In the first episode of Prompt & Response Inetum experts discuss the benefits, frustrations, and future of AI agents in marketing.

The reality check

The discussion opens with a striking statistic: according to McKinsey, 78% of companies already use AI in at least one business function. At the same time, MIT notes that 78% do not have AI-ready data. The paradox? Everyone wants AI agents in marketing and customer service, yet many organizations lack the foundations to use them effectively.

Leaders like Unilever are already reaping rewards – reinventing product shoots with AI, accelerating campaigns, and halving costs. But speed introduces new risks. That’s why the webcast explores how to make AI both fast and compliant.

How marketers use AI today

Beata recalls her early years in marketing, when almost every asset, texts, graphics, campaign materials, was created manually. Now, more than 40% of marketers use AI for content creation (HubSpot), and her own team relies on it daily.

AI tools are used to spark campaign ideas, generate quick drafts, and personalize messages. The biggest benefit so far? Time-saving. With tools like Contadu, Beata can create SEO-compliant texts in seconds, spending more time on reviewing and polishing rather than starting from scratch.

The promise of personalization

Asked about the future potential of AI, Beata points to personalization at scale. Marketing teams constantly struggle to reach the right audience with the right content while staying compliant with GDPR. AI marketing agents could enable true “segment of one” campaigns – messages so precise that they feel crafted for each individual.

She also highlights onboarding as a less obvious but valuable use case. New employees often need months to learn brand tone and guidelines. AI assistants could speed this up by helping them create compliant, on-brand assets from day one.

AI in marketing: the frustrations

Despite the promise, Beata admits that AI isn’t perfect. Generated content often feels repetitive or generic. As an experienced marketer, she can instantly spot “AI-written” text. That creates a challenge: ensuring quality and adding the “human touch” to prevent campaigns from sounding artificial.

Wiktor, who is seeing this aspect from technical perspective, agrees, adding that the flood of AI-generated content risks lowering the overall quality of the web. Future models trained on today’s AI outputs may end up amplifying biases or losing natural human voice. “If you use it wisely, it’s powerful,” he says, “but blind automation could backfire.”

What the ideal AI assistant should look like

Asked about her dream AI assistant, Beata imagines one tool combining several key features:

• content creation that doesn’t sound robotic
• personalization based on customer data
• brand alignment across tone and messaging

All merged in one AI system that supports marketers instead of overwhelming them.

Why adoption is accelerating

Why AI is being adopted so rapidly in marketing? First, as Wiktor explains, companies are sitting on mountains of customer data, and expectations for personalization are higher than ever. Second, the rise of large language models after 2022 democratized access to AI. For the first time, non-technical staff could use AI tools directly.

Early outputs were generic, but as marketers learned the importance of context and better prompting, results improved. Multimodality: text, image, audio, video, further expanded possibilities. Creating a full ad campaign with just a few prompts is now a reality.

Finally, both executives and customers are driving adoption. CEOs demand that marketing teams “use AI,” and customers expect smarter, more relevant offers.

Data and compliance: the hard truths

“Garbage in, garbage out,” Wiktor warns. Without clean, unified data, AI models will deliver poor results. Many organizations struggle because customer data is scattered across systems, duplicated, or protected under GDPR.

Compliance is the second key risk. Marketers must align with brand rules, internal guidelines, and external regulations like the EU AI Act. Bias in AI models is another concern – sometimes producing offensive or misleading outputs. For now, human-in-the-loop review remains essential before campaigns go live.

Use cases beyond text generation

AI in marketing is evolving beyond simple copywriting. Here are some examples shared by Wiktor:

• product descriptions at scale – transforming technical specs into customer-friendly narratives
• voice of the customer – analyzing reviews, sentiment, and employee feedback at scale using natural language
• campaign optimization – future agents could select channels, monitor budgets, bid on ads, and even generate performance reports

The live demo: Marketing Compliance Agent

The centerpiece of the episode was a demo of Inetum’s Marketing Compliance Agent.

Here’s how it works:

1. targeting agent filters the right customers from databases
2. marketing agent generates personalized notes from a single product brief
3. compliance agent reviews every message against brand and legal rules, automatically making small fixes
4. messages are approved instantly or routed for human review if needed

The system reduces irrelevant outreach, saves manual review time, and minimizes regulatory risks. Every message carries an audit trail for legal and compliance teams.

Why it matters for marketers

Beata as a marketer believes such tools will become essential. “Marketing is about creativity and speed, but also about compliance and trust. These agents cover all those needs.” Faster execution, better targeting, and built-in safeguards make them a natural fit for future campaigns.

Wiktor adds that the roadmap includes more autonomy (fully automated campaign execution), multimodality (text + images), and adaptability to evolving compliance rules.

Wrap-up

AI marketing agents aren’t just a trend, they’re the future. They won’t replace human marketers but will act as intelligent collaborators, helping teams run campaigns that are faster, more relevant, and always compliant.

Ready to take the next step?

Your AI journey starts with a conversation. Share your challenge, idea, or question, and our experts will respond with insights and practical guidance. Talk to us