Technology Strategy Patterns — Episode 2

The guest is Andrey Ivanov, who missed the previous session because of illness. His company does intelligent analysis of text data: a large customer can accumulate a million reviews from many channels, and without machine learning and NLP there is no way to tell what clients write about and which problems keep recurring. Earlier Andrey and Alexander built a microservice CMS for the Tinkoff site and its advertising pages on a JavaScript and React stack, and before that he was a team lead at a flight search engine. He praises part one of the book: MECE lists and logic trees are exactly what fills his Obsidian, and hunting for insights and building hypotheses is his daily work in a large team. He particularly liked that the architect here is not set against the business: instead of "selling your ideas", the book asks you to understand the business first and only then propose how to help it.

Chapter three is the world context, and PESTEL comes first. Andrey describes it as a tool for reasonably long-term trends and immediately adds a pinch of salt: in February 2020 his company analysed the situation, thought about trends and wrote a plan — in March covid arrived and the plan was binned on the spot. Alexander recalls that the framework is older than its current shape: some fifty years ago the first four letters were written — political, economic, social, technological — while the ecological and legal agendas came later. In a large organisation somebody has usually done this analysis already; it bites hardest for a startup with no financial cushion. Andrey gives his own examples: a separate stream of security audits and compliance work, because every market has its own regulation; the option of picking GCP regions with lower CO2 emissions; and a labour market where remote work and globalisation level salaries out and cancel the strategy of hiring where it is cheaper.

Alexander adds the technology layer: waves such as decentralised data warehouses and "data as a product" arrive through reports and other people's success stories, though he has seen few implementations that actually work. Andrey names another one — the democratisation of AI, where research models become a service on a cloud platform and can kill somebody else's product. What is valuable, in Alexander's view, is not even the final recommendation but the work itself: the information gathered is the input for scenarios, and PESTEL is also a list of "what if" questions, down to what happens to user data if a social network demands passport-based registration. Scenario planning starts exactly there — with the default strategy of changing nothing: we are big, there is plenty of money, it will only get better. The counter-example is the turkey fed for three hundred days and beheaded on Thanksgiving; financial organisations spent some time seriously fearing big tech, and Apple with its cards is now trying to gain a foothold in a new market through access to a huge audience.

The mechanics, per Andrey: PESTEL supplies the factors, logic trees give first- and second-order consequences, and the branches assemble into scenarios. Alexander frankly calls probability estimates a finger in the air and recommends coarse categories instead of false precision like "25 and a half per cent": in his example of a friend asking for a loan, a mortgage and a startup turn out to be different branches with different expected values, while in real life a floor-finger-ceiling rule of thumb usually wins. The futures funnel separates possible, plausible, probable and preferable outcomes: if the preferable one falls outside the outer funnel, it is a fantasy not worth focusing on. Backcasting moves from the future back to the present, and Andrey relays the author's advice: under high uncertainty pick less ambitious goals that sit more within your control, fewer bugs and faster support replies.

His example: during the covid mini-crisis sales fell, while some clients were flooded with home food orders; the team decided that selling at the old level would not work but the product could be improved a lot, and launched a new vertical with an end point and intermediate steps such as access to client data — call-centre recordings, chats, support tickets. They did not call it backcasting at the time. Alexander translates the idea into architecture: Conway's law says a system repeats the communication structure, while the inverse Conway manoeuvre goes from the properties the system needs to the team that should build it; you cannot influence your shape directly, but you can influence how much you eat and how much you train. Chapter four is the industry context. Andrey calls SWOT the most classic pattern for assessing a business concept: it works for a department or a team too. The author suggests filling it in by asking people from different areas what we are strong at, what is weak and what they would change, then sorting the answers along the helpful-harmful and internal-external axes.

The author advises starting with weaknesses — tidying the house is easier than conquering new horizons — while Alexander adds that strengths deserve investment too: Tinkoff grew its coverage area and delivery flexibility while everyone else invited you to queue. Andrey reads SWOT in pairs: expertise across industries neutralises the threat of AI democratisation, if you explain to clients why a packaged product beats an assembly of cloud services. There are everyday uses too: a new joiner runs a SWOT to work out what the company actually does, and the owner of a legacy system discovers its roadmap does not fit the company's plans. Porter's five forces bring the threat of new entrants — Uber in taxis, cheap consumables against expensive copiers, cloud AutoML that slashes the cost of entry — and the question of how easily a client can switch: similar interfaces make it easy, an ecosystem of watch, headphones and tablet makes it expensive. Then come substitutes, supplier power with its coupling to a cloud's unique services, buyer power and the market for lemons with its information asymmetry. The Ansoff matrix closes the conversation with four squares: penetration, market development, new products, diversification. The episode ran the longest so far — an hour and twenty-five; next Thursday the club continues the book, and Andrey is invited back for another episode on something else.