Latest Thinking

The context: from 4 to 10 November, a sales promotion brings together products from the Disney Store and more than 20 other brands on TikTok Shop. Creator livestreams accompany the promotion. Two new products will not appear on DisneyStore.com until 11 November.

My hypothesis: the promotion could generate purchases for Disney that would not have happened without the contact on TikTok. Someone discovers a product in a livestream and buys it without having planned to visit the Disney shop beforehand. Additional sales and marketing costs could therefore pay off if they generate profitable incremental purchases. Another opportunity could arise after the first purchase: if the shopping experience and the product leave a positive impression, the customer might later order directly from Disney.

Where there is light, there is also shadow: a positive experience on TikTok could just as easily encourage the next purchase to take place only on TikTok — or existing fans might simply bring forward a purchase they had already planned to make at Disney, or shift it away from the Disney shop entirely because of the promotion.

For digital sales, the decisive question would therefore be how much incremental contribution margin is generated across both channels (Disney and TikTok). I think this is a very interesting trial and hope that we will also hear about the results.

Source: Disney Parks Blog, 29 October 2026

A recent Skift interview with Adam Schwab, CEO of Luxury Escapes, the internationally active Australian travel platform founded in 2013 (something of a Secret Escapes from Down Under), offers an interesting perspective: through the paid travel club LuxPlus+, members receive special prices and additional benefits on selected offers. Schwab sees exclusive terms as increasingly important in light of AI booking agents.

For me, it is worth distinguishing between two services an intermediary provides: it helps customers find suitable offers. And it negotiates with hotels over the prices and services it can offer its customers.

If a personal AI assistant takes on more search and comparison tasks, the first part could become easier to replace. Attractive terms, however, remain a reason to consider a provider when booking, provided customers cannot obtain equivalent terms elsewhere.

With a paid membership, that advantage also has to justify the fee. Whether Luxury Escapes offers lasting advantages that are difficult to copy cannot be judged from the interview.

My conclusion for digital sales: if AI makes it easier to compare offers, an intermediary’s value could depend more on the special terms it secures for its customers. Luxury Escapes could be an illustrative example of this strategic bet.

Sources: Skift and Luxury Escapes

Booking Holdings introduced Vayo on 14 October and outlined a potential investment of US$100 million, depending on product progress and market conditions. The service is launching as a private beta that people can apply to join.

The idea: a traveller connects their email inbox. Vayo brings the travel confirmations it contains together in a single overview, aims to identify gaps in the planning and suggests suitable additions. It also takes account of bookings made through other platforms and services booked directly with hotels and airlines.
For example: a flight booked directly with the airline, a hotel booked directly with the hotel and a restaurant booked through OpenTable. Vayo is intended to hold these elements together. If the flight changes, the service can, according to the announcement, flag implications for the hotel arrival and restaurant reservation. Direct flight and hotel bookings within Vayo have been announced as a later addition.

I find it strategically interesting that Vayo CEO Mike Fridgen named general-purpose personal AI agents as the main competitors in an interview with GeekWire. Clearly, if customers ask their AI assistant to plan their trips in future, it could also determine which providers and booking platforms they consider in the first place.

For established travel platforms, this creates a risk of losing direct access to customers and increasingly becoming interchangeable suppliers in the background.
For me, Vayo is a response to this risk. Booking Holdings would rather build the agent itself that customers are meant to trust with their travel planning. Integrating bookings made elsewhere fits this approach: even a trip that starts elsewhere can become the entry point to a customer relationship of its own. Whether customers will use an additional specialised travel assistant for this or prefer to organise their trips with their general-purpose AI assistant remains an open question.

In my view, that is precisely the strategic bet: Booking Holdings wants to remain among the providers where travel decisions begin. It has recognised that the agentic game could disrupt its own business. That makes this strategic insight and the corresponding approach with Vayo all the more worthy of recognition, I think.

Sources: Vayo/Booking Holdings and GeekWire

When customers ask an AI about prices, services or suitable providers, it can access company websites. What happens if it cannot find enough information there?

A scientific study by Era Labs (ora.ai), a widely discussed AI infrastructure startup from Tel Aviv, provides interesting findings. The company develops solutions that make websites usable for AI agents. The authors call this “Agent Experience”, or AX: how well can an agent find, understand and use a website? I find AX an intriguing term that further humanises AI.

The study examined almost 38,000 automated research tasks covering 1,056 companies using four AI agent systems. The questions concerned pricing, product features and setup. I find this observation particularly relevant: in around 99% of research tasks in which agents reached a dead end on a website, they still answered the question. To do so, they used information they found elsewhere.
For more accessible websites, an average of 78% of the answer text came from the companies’ own pages. For less accessible websites, the figure was 58%.
Clear recommendations were also more frequent: around 20% compared with 11%.
Note: the study has not yet undergone independent scientific peer review. Era Labs uses its own evaluation framework; differences in the depth of content could help explain the results. Revenue and conversion were not examined.

For me, this points to a concrete task for Digital Sales: our own website must also function as a reliable source of information for AI agents. If important details are missing during automated retrieval, external sources could play a greater role in determining how our offering is described. That gives us less influence over whether customers receive up-to-date and complete information on prices, services and limitations.

I would therefore test typical customer questions: what information does the AI use? What is missing? Where does it turn to other sources?
In future, the quality of a sales website will also depend on how reliably it provides AI agents with the basis for an informed customer decision.

Source: Era Labs/ora.ai, 6 October 2026

That is precisely why a statement by CTO Sejal Amin caught my attention: In a recently published interview, she describes the bigger hurdle in adopting AI. Individual product teams at Priceline can develop very good applications for their own areas, but creating a really good end-to-end user flow and user experience across teams and functions is a major challenge. Sejal Amin appears to have managed it well, though.

This is certainly not a new insight. Even before AI assistants, you could almost read a company's organisational chart from some websites, because the user experience along the journey often was, or still is, inconsistent.

So anyone responsible for digital sales should actively demand a seamless journey from their product organisation and other supporting stakeholders, even when changes to the tech stack, for example, mean things may take longer and everyone understands that. It is about the ambition.

The best way to support the call for more teamwork across product teams is always to use data. These are verifiable facts that, not least, take the emotion out of the discussion. I would test the booking flow with real users, from their first travel wish through to a service issue, and identify and jointly discuss the friction within and between the steps - always with a clear metric in mind. Your own customer service team is also a great source of UX improvements.
Because one thing is clear: Every day with a suboptimal user experience costs business. With or without AI.

Sources: Priceline and PhocusWire. Screenshot below: AI assistant "Penny" on priceline.com

For context: Meta launched Muse in the US on 8 September 2026. The AI agent can visit websites, search for products and, with the user's approval, make purchases, book trips and more. Just ten days later, the app ranked first among free iPhone apps in the US App Store, ahead of ChatGPT. (Muse is not yet available in Germany.)

Shopify gives Muse access to products from eligible merchants and supports purchases through checkout flows on Meta. This fits its strategy: Shopify has been integrating its merchants into AI-powered shopping journeys since 2025, starting with OpenAI's Instant Checkout.

Amazon is now doing precisely the opposite: On 20 September 2026, Muse was blocked from Amazon's websites. According to Amazon, Meta had neither announced nor coordinated its access and does not identify itself as an AI agent when browsing. Amazon also points to risks to security and customer data. This seems consistent with the persistent criticism Meta already faces over various privacy controversies and its approach to content moderation.

While Shopify opens access to its merchants for the AI agent Muse, Amazon shuts it out. My thesis: Setting privacy and related concerns aside, the two decisions reflect different ways of generating profit. Shopify makes money from commerce infrastructure, while Amazon also monetises attention on its marketplace through advertising. Alongside AWS, it is advertising that generates the substantial profits, rather than the traditional e-commerce business.

Shopify wants to become the infrastructure behind as many AI interfaces as possible. Amazon is protecting its marketplace, where attention is sold alongside products.

Which strategy makes sense depends on where a company earns its money. To some extent, the future of AI shopping assistants depends on this too. After all, an assistant that is not allowed to enter a shop is useless.

Sources: Meta, Amazon, Shopify, Axios, GeekWire and Wall Street Journal.

OpenAI is now testing its next advertising format: so-called “Sponsored Agents”. What are Sponsored Agents?

After clicking an ad in ChatGPT, users enter a separate conversation with the advertiser's AI agent, clearly labelled as sponsored. Users can describe their needs in more detail and ask follow-up questions. The agent provides answers and presumably also tries to sell (my assumption), bringing “Persuasive AI” into play. Users could then be directed to the advertiser's website to make a purchase. OpenAI is currently testing this format with selected advertisers in the US; no results are available yet.

What I find so interesting about the idea is that a paid placement can become a kind of sales conversation, very early in the discovery phase. From a sales perspective, that is quite valuable.

For context: the speed at which OpenAI is developing advertising formats shows how significant this revenue stream is intended to become in its business model. According to the Wall Street Journal, OpenAI is at around one billion dollars in annualised advertising revenue for 2026. That is an enormous distance from the announced ambition of reaching 100 billion dollars in annual advertising revenue by 2030. Sam Altman needs this growth story for an IPO that will take place no earlier than 2027. I remain sceptical.

My professional assessment: Sponsored Agents will not become a mass-market format, but they could be exceptionally valuable in situations that require detailed explanations and involve clear purchase intent.

Sources: OpenAI product demo, 16 September 2026; Wall Street Journal; Associated Press; Axios.

Specifically, Google is expanding Data Manager to connect advertisers’ own data sources. Google Ads is adding a metric intended to show the additional contribution of a first-party data setup to conversion measurement. With Meridian and GeoX, Google is also expanding the options for examining marketing impact through modelling and geographic experiments.

The direction is clear: for years, Google has been automating more of what used to be the operational craft of performance marketing – bids, audiences, delivery, campaign management and increasingly creative assets. It is becoming a black box.

What role does that leave for performance marketers? They need to answer questions at an increasingly higher level: Which data helps guide decisions? Which sales would have happened without our advertising? And how much additional revenue does our marketing actually generate?

Google will continue to automate optimisation within its systems. But setting business objectives, ensuring the quality of the data foundation and assessing the actual business impact remain the company’s responsibility.

In performance marketing, the ability to understand which customers are valuable, which data helps guide decisions and where marketing truly generates incremental business results is therefore becoming increasingly important.

This is what good performance marketers should be measured by – and the expectations placed on them are rising enormously.

Source: Google Ads & Commerce, 10 September 2026.

It becomes particularly interesting when we look more closely at how loyalty is changing in the travel market. The American Express Platinum Card is a good example. Depending on the card market, cardholders receive direct access to status tiers in several global hotel programmes. In Germany, these include Hilton Honors Gold, Marriott Bonvoy Gold Elite and ALL Accor Gold. The usual qualification requirements are waived. A guest can therefore hold an elevated status with several hotel groups without first earning it through loyalty to the respective brand.

For me, this raises a fundamental CRM question: who is this customer actually loyal to? The hotel brand? Or rather the wider ecosystem that gives them access to different hotel brands, status benefits and other travel benefits – American Express in this example?

For hotels, the quality of the actual customer relationship therefore becomes more important. Status alone says little about customer lifetime value. What matters is whether a company knows the guest, uses their preferences meaningfully, creates recognisable added value and turns this into repeat bookings and a long-term direct relationship.

Perhaps the more important metric behind Gold status is therefore not the number of status customers, but how many of them are genuinely loyal and valuable customers – and, above all, active ones.

Sources: Skift, 4 September 2026; American Express.

This raises a new measurement question for digital sales: how do we capture the commercial influence of AI discovery? What does this mean for marketing attribution?

A customer might discover a product in ChatGPT, for example, but not click directly on the website link, instead searching for it on Google later. GA4 records the visit to the website, but ChatGPT’s earlier influence does not appear as a traffic source.

Specialist platforms such as Peec.ai measure how brands and products are visible, positioned and recommended in AI answers, and can connect direct AI traffic with analytics data. Attribution remains more difficult when AI influences discovery but the purchase takes place through another channel.

This is precisely the gap NIQ and Similarweb aim to address with an “Agentic Commerce Measurement” solution announced on 2 September 2026. Similarweb contributes data on digital usage and traffic behaviour, while NIQ contributes product, consumer and purchase data. Together, they aim to connect AI discovery with traffic, conversion and purchasing behaviour.

This would also be relevant for pure-play e-commerce companies: they know their conversions, but may not know the full journey leading up to them.

Whether NIQ and Similarweb can establish this connection reliably remains an open question. The solution is still in development and has been announced for Q4 2026. The methodology for attributing a later purchase to earlier AI discovery has not yet been explained in detail either. Caution is therefore warranted.

Yet this is exactly what will determine the value of the approach. If ChatGPT and similar systems take on a growing share of discovery and consideration, their influence will also need to become visible in attribution. Alongside touchpoints such as Google, Meta, direct traffic and email, AI discovery increasingly belongs in the same analysis.

Only then can we better understand which platforms actually contribute to conversions and revenue, and how digital sales budgets should be allocated accordingly.

Sources: Lab42, AI Shopping Report 2026; NIQ/Similarweb, 2 September 2026.

In Germany alone, 28.1 million people currently play mobile games, with an average age of 38.8.

Beyond gaming itself, I am interested in why mobile games manage to make people return voluntarily time and again.

This is precisely where many loyalty programmes struggle. They work well around a transaction: make a purchase, collect points, redeem benefits. Between two purchases, however, there are often few reasons to engage with the programme or its app.

Mobile games work differently. Progress, challenges, status and rewards repeatedly create reasons to interact.

A 2024 academic study by a predominantly German research team, published in the renowned Journal of Marketing Research, shows that such mechanics can be transferred to other apps. The researchers analysed the daily use of a gamified app by almost 19,000 users. Game rewards increased engagement in addition to conventional value rewards and generated higher business value. The study also identifies a limit: too much engagement with the game can distract users from the activities that actually create value.

At Gamescom, I saw how even Google uses these mechanics for loyalty. Google Play Points combines challenges with rewards, leaderboards and exclusive benefits for status customers.

Providers such as the German adtech company adjoe transfer the benefits of mobile gaming directly into loyalty apps. Users play mobile games and receive points in the relevant loyalty programme. Among other results, adjoe reports higher app-opening rates and an increase in customer lifetime value.

What do I take from this? Particularly in business models with long booking or purchase cycles, it may not be enough to think about loyalty solely in relation to the next transaction. In travel, for example, weeks or months can easily pass between two hotel stays or trips.

So what reason do we give customers to return before their next purchase? Perhaps this is exactly what CRM leaders can learn from mobile games: customer loyalty is also created during the time when nothing is being sold.

Sources: Photo: author’s own, Gamescom 2026; Newzoo: Global Games Market 2025; game.de: Mobile Games in Germany, 6 July 2026; Paschmann et al., 2024: “Driving Mobile App User Engagement Through Gamification”, Journal of Marketing Research, Vol. 62 (2); Google: Play Points at Gamescom, 25 August 2026; adjoe.io: Fetch Case Study.

Booking already uses AI review summaries that condense numerous reviews into a small number of statements. Google now also provides such summaries for websites and apps through its Places API. They are currently available in four languages, although German is not yet one of them.

This is convenient for customers: instead of reading numerous customer reviews, they receive a few sentences to support their decision.

These summaries clearly have an impact. According to Phocuswright, summarised reviews lead 33% of surveyed US travellers to take action based on an AI recommendation. Only price comparisons are more effective, at 44%.

At the same time, a recent study in the academic journal Tourism Management highlights a potential problem. The researchers examined the introduction of AI-generated review summaries at Ctrip, a major Asian OTA. According to the study, overly positive AI summaries can raise expectations before a trip to such an extent that subsequent hotel ratings in the analysed Ctrip dataset were effectively lower or more extreme.

What I find particularly interesting is that AI-generated review summaries create a new layer of interpretation before the purchase decision — one that can represent both an opportunity and a risk. AI systems decide which patterns across many reviews appear relevant, which aspects are highlighted and which may disappear.

This creates a new task for tourism providers. In addition to ratings and individual reviews, they should monitor the overall picture that platforms generate from their customer feedback using AI. Does the summary accurately reflect what guests are writing? Are individual negative points being given too much weight? Are relevant strengths disappearing?

Companies have little direct control over this interpretation. They can, however, ensure a broad and current base of authentic reviews, analyse the underlying topics and address demonstrably incorrect AI summaries with the platform concerned.

In my experience with customer journeys, customer expectations significantly shape the subsequent perception of the experience itself. If AI shapes these expectations in advance, its interpretation of customer feedback also becomes economically relevant — both positively and negatively.

Sources: Phocuswright, “The AI Surge 2026”; Tourism Management, Wang et al. 2026; Google Maps Platform; Booking.com. Graphic: Google.

The reasoning is clear: conversations involving human agents are comparatively expensive for companies. Although the technology has improved in recent years, mediocre speech recognition, lengthy issue qualification through cumbersome decision trees, waiting queues and human first-level agents with limited authority still cause too much frustration for customers.

From a senior-management perspective, customer orientation and service have always been described as highly important. In practice, however, service teams were often treated as efficiency-driven administrators of scarcity and remained largely invisible.

AI voice agents can change this. IONOS, for example, has offered a commercial AI telephone assistant since autumn last year. It interacts with callers around the clock using natural language. Through API integrations, it can retrieve CRM data, create support tickets or book callback appointments directly during the conversation.

Zendesk is pursuing a similar direction with its Voice AI Agents for larger service organisations.

The fact is that if there is one stable and proven enterprise use case for AI, it is customer service.

I have never viewed customer service primarily as a cost-saving case. I see it as critical to CRM, an investment in customer lifetime value and a driver of growth. When an AI voice agent recognises the customer, takes their history into account and resolves the issue, it becomes a critical part of the customer journey.

This naturally requires a reliable data foundation. Customer history, transactions, service cases and product information must be current and available in a controlled manner.

Customers, meanwhile, expect their issue to be resolved or progressed immediately and the existing context to be preserved. All of this is now possible. Second-level support can then be provided by even better-trained people who are assisted by AI when needed.

In the longer term, the opportunity extends beyond inbound customer service. AI voice agents are also relevant for sales. Support-to-sales, meaning inbound upselling, as well as outbound activation such as re-engaging abandoned bookings or shopping baskets, are entirely conceivable.

Perhaps one of the oldest contact channels will gain new importance through an exemplary collaboration between people and AI.

Sources: IONOS AI telephone assistant; Zendesk Voice AI Agents.

Consider the OTAs. Expedia is expanding its long-established B2B business into a broader travel stack. Depending on their focus, partners will be able to use this infrastructure to integrate hotels, flights, rental cars, activities, insurance, payments and other services into their own offerings.

Booking Holdings is moving in the same direction and, according to Skift, is currently planning to consolidate the B2B activities of Booking, Agoda and Priceline even further.

This means that an airline, bank, hotel group or cruise company can sell travel services to customers on its own website or app while the supply and parts of the technology are provided by Expedia or Booking in the background.

So what does “direct” still mean? Technically, the distinction is becoming increasingly blurred.

I would therefore define direct sales more broadly. What matters is who controls customer access, owns the relevant first-party data, manages pricing and availability, shapes the customer experience and can turn a single transaction into a long-term customer relationship. For me, that customer relationship is the strategic value and the objective of “direct”.

Using a company’s own channel as the booking route alone does not create this value. For hotels and other travel providers, the more important question is therefore which parts of the customer journey they need to control themselves and which they can source from platforms — not least with gross margins in mind.

Expedia and Booking demonstrate how powerful the infrastructure in the background has become. This raises expectations for what must happen in the foreground of direct sales, alongside the commercial challenge and economic necessity of getting users onto a company’s own channels in the first place.

Sources: Expedia Group B2B/Explore 2026; Skift on Booking Holdings’ new B2B unit.

Layla is an AI travel agent that creates personalised day-by-day itineraries, displays real-time prices and combines inspiration with human experts, among other capabilities.

Why is this relevant? Digital direct-sales strategies aim to make a company’s own channels — its website and app — the preferred place for conversion.

Yet the planning phase no longer begins on the provider’s website. It increasingly starts on social media, Google or ChatGPT, while organic direct traffic tends to decline. Every guest who reaches a company’s own channels therefore matters, including those who do not yet have a specific booking intention. Ignoring them sends users back from the company website to OTAs such as Expedia. With Layla, Expedia is acquiring precisely the transformation of this discovery phase, in which travellers have only a rough idea of their trip.

For direct-sales leaders, the practical implication is clear: bring the planning phase into your own channels — even without Layla. A simple AI concierge offering conversational search on the website or app can often be enough. It can capture preferences, travel dates and interests before a specific booking request emerges. Such a tool, however, is only as good as the proprietary product and CRM data behind it. That is exactly where a company’s advantage over third-party channels such as OTAs lies.

Analyse how many website visits over the past three months took place during this research phase without progressing into the booking process. Conduct user interviews to understand why, and bring the discovery phase onto the website without distracting from the direct-booking option, which should remain visible at all times.

Source: Expedia Group, “Expedia Group acquires Layla, accelerating its AI-powered trip planning and booking strategy”, 31 July 2026.

For many years, CRM was primarily a system of record. It documented customer data, activities, opportunities and transactions. It should still fulfil this role today as the leading system for customer data.

Around 15 years ago, the term system of engagement, coined by Geoffrey Moore, became established in enterprise IT. CRM systems evolved into platforms through which companies also interact with customers across digital channels.

AI now marks the next stage of this evolution. Salesforce, Microsoft, HubSpot and Oracle are ultimately pursuing the same direction: through the use of AI agents, CRM is developing into a system of action. It does more than document information; it prioritises leads, recommends the next action, automates processes and ideally supports operational decisions.

For CRM leaders and teams, this means that customer data quality becomes even more important because AI prepares decisions on this basis. CRM processes also gain significance because companies must define which tasks AI may perform and which remain under human responsibility. The ability to guide AI effectively, establish guardrails and make decisions understandable therefore becomes more important.

The value of an AI agent in CRM depends not least on whether the CRM professional understands which recommendations the agent makes, why it makes them, which data they are based on and under which conditions they should be implemented. This is precisely where I see the next stage of CRM evolution.

Sources: Geoffrey Moore, “Systems of Engagement and the Future of Enterprise IT” (2011); Salesforce Agentforce; Microsoft Dynamics 365 AI & Copilot; HubSpot Breeze; Oracle AI for Fusion Applications.

A recently published scientific preprint offers an interesting perspective. The two authors, researchers from Canada and France, compare two identical e-commerce websites in an experiment, including the same product catalogue, prices, inventory and checkout.

The only difference was that one website had been specifically optimised for AI agents using a framework with clearly structured information, unambiguous actions and machine-readable product data. Browser agents based on GPT-4.1, Gemini 2.5 Flash and Grok 4 Fast then had to complete typical shopping tasks such as product searches and comparisons.

A total of 300 test runs were evaluated. The agent-optimised website increased the browser agents’ success rate from 49.3% to 89.3%, an improvement of 40 percentage points. It should be noted that this was a proof of concept and the results cannot be transferred directly to real-world websites.

If AI agents take over research, comparison or parts of the buying process in the future, a website must provide information in a way that agents can reliably understand and use.

SEO and GEO help companies get found. Usability, accessibility and conversion rate optimisation help turn visitors into customers. Agent readiness will determine whether AI agents can successfully interact with a website. I consider this scenario particularly realistic for low-involvement products and standardised purchases.

Websites are only the beginning. Native apps will face the same question: how easily can AI agents retrieve information and execute transactions directly without having to operate the user interface like a human?

Source: “Designing Agent-Ready Websites for AI Web Agents: A Framework for Machine Readability, Actionability, and Decision Reliability” (Said Elnaffar, Farzad Rashidi), preprint published on arXiv on 13 July 2026.

Chris Nassetta, Hilton’s iconic CEO who has held the role since 2007, described two practical AI use cases for hotels this week on “The Angle”, a podcast by asset manager T. Rowe Price. First, AI should identify service issues during a stay, for example by analysing guest messages or social media posts, and help employees resolve them while the guest is still at the hotel. Second, Hilton wants to use booking history, search behaviour and preferences to present guests with significantly more relevant offers. Nothing revolutionary at first glance.

These may initially sound like separate projects, but both use cases are based on the same foundation: high-quality first-party data.

AI models are becoming more capable and accessible to everyone, while access to high-quality customer data remains company-specific. For digital sales and CRM leaders, the implication should be clear: companies need to build and prepare the data foundation across the entire customer journey that enables AI to deliver better service, more relevant offers and stronger customer loyalty, whether in hospitality or other industries.

First-party data is not everything, but without it many AI applications cannot realise their true value for customers. First-party data is the strategic competitive advantage.

Source: “The Angle” podcast (T. Rowe Price), interview with Hilton CEO Chris Nassetta, 15 July 2026

Kollective, an Athens-based agency specialising in hospitality, analysed almost 14,000 responses from ChatGPT, Copilot, Gemini, Google AI Overviews and Google AI Mode about hotels in 100 destinations for its AI Visibility Index. A fixed set of search queries was used, such as “Recommend a boutique hotel in Lisbon”. According to the study’s preliminary findings, the AI systems agreed on the same “top hotel” in only around 4% of cases for an identical query, meaning the hotel mentioned first or ranked highest in each system’s response. In an intraday test in which the same prompt was submitted twice on the same day, one hour apart, the top hotel recommendation changed in an astonishing 45% of cases.

This fundamental observation is consistent with what we know about AI systems: AI responses are not deterministic. They depend, among other things, on the model, context and timing. Many hotels therefore try to appear directly in the AI response for “the one” typical traveller query, which is hardly feasible. This also transfers the logic of conventional keyword rankings to a system that simply works differently.

My experience confirms this: properties with current, structured data and a presence across many channels are also more visible in AI systems. Recurring presence across multiple systems matters more than a single top position on one particular day.

In practice, a hotel website that is already strong in conventional off-page and on-page SEO, technically well structured and equipped with machine-readable data, current prices and availability will be more consistently discoverable across multiple AI systems. That is the work that actually makes a difference.

Source: Kollective, Boutique Hotel AI Visibility Index, preliminary findings reported by Hospitality Net, 8 July 2026.

At the end of June, Salesforce released its largest agentic commerce update to date. Its centrepiece is the Shopper Agent, which supports customers on the merchant’s website from product discovery through to checkout.

Salesforce supports the value proposition of its Shopper Agent with data from the 2025 holiday season in November and December. Companies with their own Shopper Agent, including Pandora and SharkNinja, grew by 6.2%, compared with 3.9% for companies without agents – an increase of 59%. This is, of course, a Salesforce analysis, but the figures are a useful signal. In this model, the merchant’s own website remains the place where the purchasing decision is made. The customer relationship and the final step before purchase remain with the merchant.

Google is taking a different approach. Through the new Universal Commerce Protocol (UCP), Google plans to enable checkout directly within Google AI Mode and Gemini, without customers leaving the Google interface. The positive aspect is that the merchant remains the merchant of record. Although the customer purchases in Gemini, the merchant receives the full data for its CRM system and subsequent marketing activities – but it still needs to use that data effectively.

On the one hand, I see a new platform risk emerging, comparable to what merchants know from Amazon Marketplace or travel providers from Booking and Check24: less website traffic, fewer cross-selling and upselling opportunities, less first-party data and weaker brand loyalty.

Instead of avoiding AI channels, however, companies should shape their customer relationships so that customers buy directly from them next time. Loyalty programmes, personalised added value and exclusive benefits can deliberately guide customers back towards direct purchasing or booking.

Initial reports about UCP from the United States still focus primarily on technical integration and the growing number of AI search queries. It will therefore be interesting to see when the first reliable data becomes available on how merchant revenues develop through UCP. I will continue to watch this closely.

Sources: Salesforce press release, “Salesforce Introduces Agentforce Commerce” (25 June 2026); Salesforce Shopping Index, “2025 Holiday Shopping Data”.

Alongside China Speed and Creator Commerce, agentic commerce was the dominant topic at K5 – Future Retail 2026 in Berlin. It is not limited to product-based e-commerce. It is equally relevant to travel and ticketing, subscriptions and digital services, although these areas receive comparatively little attention.

For the travel industry, this represents a differentiated opportunity. Not every travel booking is the same. Planning a dream holiday in the B2C segment can be an emotional experience. In this situation, users are less likely to want an agent to make the decision for them. Commoditised bookings, such as business travel, are different: price, availability, location and speed are the primary criteria. This is the realistic entry point for agentic commerce in travel distribution. In every case, success depends on data quality, API accessibility and machine-readable offer structures. Yet providers will not want to become mere product feeds in the long term.

Based on my many years of successfully managing profitable direct bookings against third-party channels, I can say that the counterstrategy includes strong brand management, relevant CRM and quality on property that generates repeat bookings and recommendations. Brands that remain top of mind are more likely to be booked directly: commission-free, with control over pricing and customer data – and therefore less often through agents.

Agentic commerce is still at an early stage. OpenAI’s Instant Checkout failed in March 2026 because of insufficient real-time data synchronisation, fraud prevention, tax infrastructure and merchant adoption. In the EU, regulatory questions involving liability, data protection and payment authorisation remain unresolved. Broad user adoption is also still to come.

This makes it even more important to monitor developments closely and move into implementation as a fast follower as soon as reliable evidence emerges about what works.

Sources: IBM Think 2026; K5 conference, June 2026; Google Cloud, January 2026; Forrester Research 2026.

Les Binet and Peter Field, the two godfathers of marketing effectiveness research associated with the UK advertising industry body IPA, established the concepts of brand building and brand activation. Their study “The Long and the Short of It” (2013) covers almost 1,000 campaigns across 30 years. Its rule of thumb for maximising total returns is to allocate 60% of the budget to brand building and 40% to brand activation. Brand building fills the funnel from the top; activation empties it at the bottom.

In 2026, a new argument for brand building has emerged. According to Seer Interactive, organic click-through rates have fallen by 61% since the arrival of Google AI Overviews, which are still shown somewhat less frequently for purely commercial searches, while paid click-through rates have fallen by as much as 68%. Brands cited in AI responses, by contrast, see 35% more organic clicks and 91% more paid clicks.

A brand is not cited solely because of its awareness. What is becoming visible, however, is that brands with high authority and a strong presence across the digital ecosystem are used as references disproportionately often. Brand building influences the signals that matter to Google in this context.

My many years of experience in digital sales lead me to one clear conclusion: performance campaigns work better when brand awareness and trust have prepared the ground.

Activation determines today’s revenue; brand building determines tomorrow’s revenue. Companies that pursue only one optimise at the expense of the other.

Sources: Les Binet, IPA Effectiveness Conference, October 2025; Binet & Field, “The Long and the Short of It”, IPA 2013/2018; Seer Interactive AI Overviews study, September 2025.

Industry data shows where most companies stand today: 61% use their CRM primarily for email marketing and automated communication. Only 34% use its analytical capabilities to support decisions. From my own experience, customer data is generally available in sufficient quantities. The greater challenge is analysing it properly, deriving better decisions from it and consistently translating those decisions into operational processes.

Salesforce, HubSpot and Microsoft are systematically developing their CRM platforms towards AI-supported decision-making. Analytical CRM – including scoring, segmentation, churn prediction, customer lifetime value optimisation and next best action – has traditionally been the domain of professionally organised teams with dedicated specialists.

AI-powered modules in the major CRM suites are now making these capabilities more accessible and lightweight for regular business teams. Nevertheless, an understanding of the core concepts of analytical CRM remains essential. Judgement and a human in the loop still matter.

Companies that continue to use CRM primarily as a campaign tool leave a great deal of potential untapped. The future of CRM will be shaped more by the quality of decisions prepared by AI and validated by people than by the number of campaigns deployed.

Sources: HubSpot Breeze AI Guide 2026; DesignRush/Salesforce CRM Statistics, 2026.