What Polytechnique's digital business course does not yet tell small organisations (2026 update)
The Digital business course that École polytechnique offers on Coursera explains platforms well, but its readings stop around 2016. Since then, across Europe, the visitor often gets an answer without coming to your website, and machines read your pages in the visitor's place. For a tourist office, a local authority or a service SME, what depends on you needs no budget: count bookings instead of visits, keep each piece of information in one place, and keep your data in your own hands.
What does the course say, and what still holds in 2026?
École polytechnique offers a course in English on Coursera, "Digital business: grow on digital world": three modules, about ten hours, free to enrol. It explains platforms, network effects, the disruption Airbnb caused and the arm-wrestling with regulators. Its readings date from 2013 to 2016. They are still right on the principles.
Four rules have not moved. A platform does not sell a product: it sells one group access to another, and it subsidises one side to charge the other. It only takes off if it removes a real nuisance, otherwise it is a fad. You launch one high-value connection before you chase volume. And a price advantage that vanishes as soon as the rule applies to everyone is not efficiency, it is a way round the rule.
For a small organisation these rules read in reverse. A tourist office is itself a platform: it connects visitors and local providers, and it is funded from one side. But the gateway does not belong to it: the regional or national tourism database, the search engine, the large booking sites. The course teaches you to read that position. It says nothing of what has changed since, and that is what matters to you in 2026.
Why does your website get fewer visits than before?
Because the visitor gets an answer before reaching you. From January to April 2026, between 62 % and 70 % of Google searches ended without a single click in the four European countries measured: 69.5 % in the United Kingdom, 65.3 % in France, 63.4 % in Italy, 62.1 % in Germany, according to the Similarweb panel analysed by SparkToro (study of 16 June 2026). About two searches in three stop on Google's own page.
Answers written by AI push this further. In the United States, the agency Seer Interactive followed 3,119 informational questions from June 2024 to September 2025: when Google shows a generated answer, the click rate to websites falls from 1.76 % to 0.61 %, which is 61 % less (summary published by Search Engine Land, 4 November 2025). The same study notes that clicks also fall, less sharply, on questions with no generated answer. For Europe I have no comparable figure: the European Commission records publishers' concern in its staff working document of 28 April 2026 on the Digital Markets Act, without measuring it.
The consequence is simple. The page that used to inform no longer brings the visitor: it is read by the machine that answers for you. Your visit count will fall without your activity necessarily falling. Anyone who watches visits alone will decide on a number that no longer means anything.
Will assistants book in place of your visitors?
Not yet, and it pays to be precise because many are not. Since late 2025, several assistants can complete a purchase without leaving the conversation. The path is not a straight line: ChatGPT launched the feature in September 2025, and Shopify's documentation now says that a buyer coming from ChatGPT finishes the order on the merchant's own shop. Payment inside the conversation exists elsewhere, in Gemini or Copilot for example, and for now it is an American affair: on each channel Shopify opens to its merchants, the shop must sell to customers in the United States, and Copilot's payment is shown only to them.
For your trade, accommodation booking is announced, not open. Google has published a standard for hotel booking by its assistants and invites professionals to join a waiting list. Activities and ticketing are not in it. I found no case in service in Europe.
So this is not an emergency, it is a direction. The technical rules that connect assistants to business software are published as open standards, and several were handed to independent foundations in 2025 and 2026. I infer that access will be open to all. What will make the difference is who holds the up-to-date information and the relationship with the customer.
What can you demand from the large platforms?
Two things, and neither comes on its own. The Digital Markets Act (Regulation (EU) 2022/1925) applies across the European Union and targets the very large platforms the Commission has designated, Google, Apple and Booking among them. It does not target you. It gives you rights against them when you sell or make yourself known through them.
The first: to promote your offers to customers acquired on the platform and to conclude with them outside it, free of charge (Article 5(4)). The Commission has enforced it: on 23 April 2025 it fined Apple 500 million euros for preventing businesses from steering their customers to their own channels. The second: to access, free of charge, the data your activity produces on the platform (Article 6(10)).
These rights are only exercised if someone asks for them, in writing, from the platform. In a small organisation the question is who. What I write here is a reading of the text, not legal advice, and it does not cover the United Kingdom, which has its own rules.
What can you do now, with no budget?
Four moves, and none asks you to build an assistant, a chatbot or a platform. Money is not the brake. In 2025, 17 % of small businesses in the European Union (10 to 49 employees) used AI, against 55 % of large ones, and among those that considered it without adopting it, the first reason given is lack of expertise (Eurostat, data of December 2025).
First move: count bookings, calls and quote requests, not visits. One reading a month is enough.
Second move: keep each piece of information in one place. Opening hours, prices, availability, conditions: one reference record that feeds the website, the Google listing and the booking tool. It is what search engines and assistants read today, and what the purchase channels will read if they open in Europe.
Third move: write in plain words the facts the machine will quote. Hours, prices, capacity, area covered, names of people, current dates. If the generated answer quotes someone, it may as well be you. And decide which robots you let read your website, instead of putting up with them: the services that deliver websites offer that setting, often at no charge.
Fourth move: keep the data and the relationship in your own hands. At each software renewal, ask in writing for an export in an open format. The customer list, the bookings and the invoicing stay in a tool whose access your organisation holds.
What I think
The course is good and I recommend it to a director who wants to understand why the large booking sites take a commission from him without producing anything. But it describes a world in which the visitor clicked. That world is closing, and the temptation will be to buy something in response: an assistant on the website, a redesign, one more subscription. I think the opposite is needed. What machines read is your information. If it is entered in three places and contradicts itself, no tool will correct that, and the machine will quote the wrong version with confidence.
In several cases you need nobody. If your bookings and calls hold from one year to the next, the fall in visits costs you nothing: note it and move on. If your hours and prices are the same on the website, on Google and in the printed brochure, your information is already kept: keep the person who keeps it. If your website comes from a shared platform run by a region or a national body, these questions are for that platform. And the check has its limits: comparing three displays does not say where the gap comes from, only that it exists.
If you are in the situation where the same information is entered twice, in the regional database and on the website, and the screens show something else, the architecture diagnostic is what answers it. It draws on one page the map of your tools, the flows between them and who holds each access, it names the dependencies that expose you, and it leaves you a ranked action plan that your team or your current supplier can carry out. It sells nothing to build. Its price is set out in the written note that follows the first call.