London Business Club took place in London on November 19, 2025, as an in-person business talk and discussion. The speaker was Igor Ivitskiy, an entrepreneur and former researcher in mathematical modelling.
The talk, The AI era in marketing and business, followed Igor’s own year-long path: an unsuccessful attempt to automate too much, a simpler way to preserve company context, specialised AI advisers for different business roles, customer models for reviewing ideas, and a new view of search, advertising and brand visibility.
The useful layer is context, not automation theatre
Igor’s first attempt followed the familiar integration route: connect AI, customer systems and automation tools, then let data move through an elegant diagram. In practice, connections failed, maintenance invoices kept arriving and a client-facing bot invented promises. The experience changed the question. Instead of asking how much of the company could be automated, he began asking where AI could improve a decision while a person still owned the outcome.
The answer was a deliberately lighter architecture. Company context, audience knowledge, communication principles and operating constraints live in documents controlled by the business. An AI system processes that context for a defined role, but it is not the only place where the knowledge exists. This matters in marketing because useful output depends on the same discipline as sound keyword research: the quality of the evidence and the frame determines what can be learned from it.
The same logic extends to advertising data. Google describes website visitors and app users as part of an advertiser’s own data segments. Igor’s broader point was strategic: when platforms have less observable context and more automated decision-making, the business must become more deliberate about the information it owns and the outcomes it sends back.
Role-specific advisers examine strategy, finance and marketing from different angles.
Company-owned documents preserve context outside any single model or chat.
Customer evidence is turned into models that challenge internal assumptions.
Marketing results and failed tests become reusable organisational knowledge.
My principle
I keep business context in documents that I control and use AI as a processor of that context. This lets me change models without rebuilding the company’s memory and keeps a person responsible for every client-facing decision.
This is also the approach we teach: understand the business problem, preserve the evidence and evaluate the output before expanding its role. The training goes deeper into the working method without treating AI as a substitute for judgement.
Key insights
- AI reflects the quality of the person and context behind the request. A generic chat tends to return generic advice; accumulated business knowledge changes the level of the conversation.
- Portable context reduces model dependence. Company knowledge stored outside a chat can be attached to another system when a product changes or a conversation disappears.
- Different roles need different frames. Separate strategic, marketing and financial advisers reduce the risk that one recent topic distorts every later answer.
- Humans remain the client-facing filter. AI can prepare analysis and drafts, but the talk’s failed integration showed the cost of letting it make unchecked promises.
- Customer models are a pre-test, not market proof. They can expose blind spots before money and time are committed, but they do not replace real behaviour.
- Automated advertising amplifies the data it receives. Broad reach, including broad-match keywords, needs reliable conversion signals and continued control of irrelevant demand through tools such as negative keywords.
Key moments from the talk
From scepticism to a different question
A year before the talk, Igor had publicly dismissed AI as a distraction. His first serious implementation was more ambitious: outside integrators connected AI tools, customer systems and automated workflows. The diagram looked persuasive, but the operating reality was brittle. Connections broke, support work continued after launch and a bot promised a customer something the company had never offered. The lesson was not that AI had no business value. Visible technical complexity could hide a weak definition of value. Igor reframed the task around decisions, knowledge and responsibility. “AI is, in effect, a mirror.” A vague business gives it vague context; a thoughtful operator can use it to examine assumptions that would otherwise remain implicit.
Three historical comparisons shaped that reframing. Printing removed a monopoly on access to knowledge, photography freed painters from merely copying reality, and the compass reduced uncertainty for navigators. Igor argued that AI combines elements of all three: it can compress access to knowledge, absorb routine cognitive work and help an entrepreneur inspect an uncertain market. That does not mean delegating the company to a machine. His practical alternative was a Pareto-style route, seeking much of the useful effect with ordinary AI products and documents rather than an enterprise platform. A lighter system demanded less attention, preserved room for strategic work and was easier to stop when maintenance outweighed insight.
Memory that belongs to the business
The turning point came after months of building context inside one long conversation. The chat disappeared. That failure exposed a structural risk: if the only useful memory lives inside one provider’s interface, the business does not truly own it. “I do not want to be tied to one chat,” Igor recalled deciding. He began keeping stable knowledge in external documents: descriptions of the audience, the company’s voice, strategic constraints and other recurring context. A model could then receive those materials for a task, while the source remained available for another model later. He compared the difference to moving from a game that restarts after every failure to one that can save progress. The value lies less in a clever prompt than in not having to explain the same company from zero every morning.
One general adviser was still not enough. Long conversations pull a model’s attention towards the topic discussed most recently, so Igor separated strategic, marketing and financial viewpoints. Each adviser looks at the same company from a narrower professional frame, while a strategic role compares their conclusions. He extended that logic to employees by giving them focused assistants for recurring tasks. Yet the failed bot remained a boundary condition. “A human communication filter is indispensable.” In his system, a person reviews the draft, checks it against reality and remains accountable for the message sent to a client. AI reduces the amount of specialised preparation required for a task, but it does not inherit the company’s authority to commit, promise or decide.
What changed in clients, visibility and advertising
For customer work, Igor assembled evidence from questions, reviews, complaints and the language people use in different markets. From that material he created several synthetic customer perspectives and used them to challenge a landing page, an advertisement or a product idea before committing months of work. This countered the founder’s habit of assuming that customers understand a product exactly as its creators do. The same evidence-first idea informed his work on generative search visibility. He built broader third-party presence, published technically deep material and strengthened a central source of identity. Months passed without an obvious result. Then a US prospect said an AI assistant had recommended him, accepted a $900 hourly consultation, and similar inbound enquiries began arriving at roughly two per week.
The final part of the talk moved from brand discovery to advertising systems. Igor sees platforms becoming simpler on the surface while their algorithms make more decisions underneath. That raises the value of first-party audience information and of accurate feedback about which leads became useful business outcomes. Google’s own guidance says advertisers define what counts as a conversion and can attach values to focus on higher-value conversions. The risk described on stage was the inverse: if job seekers, bots or irrelevant enquiries are recorded as success, an automated system can search for more people who resemble them. Search behaviour is changing at the same time, and answers increasingly appear before a site visit. Ads therefore have to carry a clear idea, while campaign history, audience evidence and past tests need to remain accessible as organisational memory.
The full recording of this talk is included in the bonus pack for students.
Key takeaways
Useful AI adoption began when company context moved outside individual chats and every role received a clear frame.
In marketing, customer evidence and trustworthy outcome data mattered more than technical complexity.
The reported commercial signal was recurring inbound demand from people who arrived after an AI recommendation.