Marketing needs to learn – not just automate

Christian Schmitz: Because usage does not yet constitute an operational model. The Lünendonk Survey 2025 highlights this gap very clearly: whilst 95 per cent of the service providers surveyed already offer AI agents, only 31 per cent of user companies had implemented corresponding solutions.

At the same time, 65 per cent cited the integration of AI into marketing processes as a key challenge. The problem is rarely the model alone. What is lacking are prioritised use cases, reliable data access, clear decisionmaking authority and a measurement framework that extends all the way to the bottom line. Successful companies therefore do not start with the longest possible list of use cases. They select a few highly relevant value streams, appoint a responsible owner and es-tablish a baseline in advance: for example, lead time, quality rate, incremental revenue or costs avoided. Only once bene-fits, quality and risk have been substantiated is the initiative scaled up.

This is what distinguishes industrial transformation from ‘pilot theatrics’. Speed is important – but only when ba-cked by evidence. Otherwise, you simply automate existing inefficiencies and, in the worst case, produce more mediocrity at a faster rate.

Christian Schmitz: The brand does not become any less important, but rather more demanding. When generic pro-duction is virtually free and available in unlimited quantities, the value of what cannot be reproduced at will increases: a clear stance, distinctive codes, cultural relevance and a strong idea.

That is why we speak of ‘Meaning for Machines’. Brands must continue to connect with people on an emotional level, whilst at the same time being clearly readable by search sys-tems, recommendation algorithms and AI assistants. These systems are increasingly acting as intermediaries between supply and demand. Those who optimise only for traditional search engines are falling short. Companies need a consistent architecture of meaning: structured product and knowledge data, verifiable statements, recognisable language, and assets that fit together across channels.

This must not result in technocratic optimisation. AI can generate variations and predict reactions; however, it cannot make the business decision as to what a brand wants to stand for. The greater the efficiency of machines, the more import-ant human ambition, taste and the courage to be different will become.

Christian Schmitz: The industry has long confused per-sonalisation with creating the most detailed personal profiles possible. Yet relevance is not always determined by who someone is in the long term, but by what that person wishes to achieve at a specific moment. Intent can be deduced from context, behaviour and situation – often using less data and proving more useful than a comprehensive identity pro-file.

A learning marketing system should therefore interpret signals, form hypotheses and test them at the next point of contact, rather than hastily categorising people into rigid segments. This not only improves the customer experience, but also strengthens trust and resilience in a world where less third-party data is available. Clear rules are essential: which signals may be used for which purpose? How long do they remain relevant? When is human approval required? And how can a decision be traced? Responsible AI is not an after-the-fact compliance filter, but an integral part of the design. Truly good personalisation does not feel like surveillance. It reduces effort, increases situational benefit and respects customers’ autonomy in decision-making.

Christian Schmitz: We need to move away from ‘Metrics to Money’. Reach, clicks or engagement remain useful for dia-gnostic purposes, but they are not a sufficient answer to the question of what contribution marketing makes to business success. This requires an end-to-end chain stretching from the activity itself, through observable behavioural changes, to incremental revenue, margin, customer value or risk reduction. No single method can map this chain on its own. Marketing mix modelling, experiments, attribution and business data must be combined as appropriate for each decision. It is equally important to define the decision before the analysis: which budget, which offer or which touchpoint is to be changed? This is how measurement becomes actionable.

Plan.Net deliberately positions itself here not as a provider of yet another isolated platform, but as an orchestrator between marketing, finance, data and technology. Our aim is to establish a framework that remains firmly embedded within the organisation: with transparent assumptions, repeatable tests and clear learning cycles. This builds trust with the CFO and prevents optimisation from focusing solely on effects that are easily measurable in the short term but strategically weak.

Christian Schmitz: Many market players are currently, and quite rightly, investing in their own AI platforms, content engines and agents. What sets us apart is the combination of three elements: integrated implementation expertise, a consistent learning architecture and ‘client-owned intelligence’. We do not want to set up a new ‘black box’ alongside the organisation, nor do we want to create a dependency on a proprietary ‘agent zoo’. Together with our clients, we structure data access, decisionmaking rules, prompts, evaluation procedures and knowledge bases in such a way that they become part of their own assets.

We then openly orchestrate the appropriate models and technologies around these. At the same time, we link this technical layer with brand, crea-tive, media, customer experience and commerce. After all, faster content production alone does not equate to better marketing. Only when a strong idea takes effect across the entire experience, systems learn from results and the economic impact is proven, does a sustainable advantage emer-ge. In practical terms, for the next twelve months this means: selecting a relevant value stream, defining governance and a baseline, appointing an interdisciplinary team and learning productively in short cycles. It is not the organisation with the largest collection of tools that wins, but the one with the highest responsible rate of learning.