<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:yandex="http://news.yandex.ru">
  <channel>
    <title>Alexander Kolosov Insights</title>
    <link>https://akolosov.com/en/insights</link>
    <description>First-hand notes on entrepreneurship, product strategy, B2B SaaS, mobility, AI and building companies across international markets.</description>
    <language>en</language>
    <lastBuildDate>Fri, 21 Aug 2026 12:00:00 GMT</lastBuildDate>
    <atom:link href="https://akolosov.com/rss-en.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>How I launched a healthy fast-food business by risking everything</title>
      <link>https://akolosov.com/en/insights/how-i-launched-healthy-fast-food-business</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/how-i-launched-healthy-fast-food-business</guid>
      <pubDate>Fri, 21 Aug 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Founder story</category>
      <description>The founder story behind Mr.Pit: selling an e-commerce stake, studying a French concept, rejecting a franchise and building a seven-location healthy fast-food chain.</description>
      <yandex:full-text><![CDATA[In 2013 I sold my stake in an e-commerce company and put the money into a business I had never operated before. The bet eventually became Mr.Pit: a healthy fast-food chain that grew to seven locations and about 1,500 customers a day. I was 26. On paper, my position looked comfortable: I had the London Night Club, was a co-owner of the Ferma Dance bar and held a minority stake in a large CIS e-commerce company. The reality was less reassuring. We were still repaying the loan behind London. Two crises had already hit the business. Artists, sound systems and new equipment demanded fresh investment again and again. Ferma was successful and was even named a leading bar by Time Out, but a conflict developed among the shareholders. I left the project by the end of the year. Looking back, I think I should have defended my position more firmly. At the time, however, the practical question was simpler: where could I find capital for a new launch? The decision to reinvest everything The only fast source of capital was my e-commerce stake. I sold it and reinvested the proceeds. There was no spare cash left. This was not a diversified portfolio decision; it was a concentrated entrepreneurial bet. I believed healthy fast food could become a large market. People already understood pita, falafel, shawarma and grilled chicken. The opportunity was not to invent a new eating habit, but to redesign a familiar product: better ingredients, a cleaner process, transparent preparation and a modern brand. We decided to improve shawarma rather than teach the market to eat something completely unfamiliar. Why we went to France I studied international formats and found Pita Pit: grilled chicken without oil, vegetables, yoghurt-based sauces and vegetarian options served quickly in pita bread. My partner and I travelled to Nantes to understand the system from the inside. We looked at the technology, operations, people and the product itself. We originally considered buying a franchise. In the end, we built the concept ourselves. The trip still gave us something more valuable than a licence: a working reference for product architecture and operations. We could see which elements were essential and which had to be adapted to the Russian market. What turned an idea into a chain The concept became Mr.Pit, the first healthy fast-food chain of its kind in Russia and Eastern Europe. The business reached positive EBITDA in its fourth month, served more than 40,000 customers in the first six months and later expanded to seven locations. Those results did not come from copying a foreign menu. They came from turning an observation into a repeatable system: Choose a behaviour customers already understand. Improve the product in a way people can taste and explain. Study a proven model, but localise the economics and operations. Design the production process before scaling the brand. Make the founder close to the product during the first months. The lesson I kept Entrepreneurship is often described as finding a brilliant original idea. My experience was different. The opportunity appeared at the intersection of a familiar mass-market product, a visible change in customer preferences and an operational model that could be standardised. Founder takeaway: a foreign concept is not a business plan. Use it as evidence, then rebuild the product, unit economics and operating system for the market in which you will actually compete.]]></yandex:full-text>
    </item>
    <item>
      <title>Waymo’s sixth-generation Driver: what falling autonomy costs mean for robotaxi economics</title>
      <link>https://akolosov.com/en/insights/waymo-sixth-generation-robotaxi-economics</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/waymo-sixth-generation-robotaxi-economics</guid>
      <pubDate>Thu, 13 Aug 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Autonomous mobility</category>
      <description>A mobility product analysis of Waymo’s sixth-generation Driver: lower hardware costs, winter operations, fleet economics, regulation and robotaxi scaling.</description>
      <yandex:full-text><![CDATA[Waymo's sixth-generation Driver matters because autonomous mobility is moving from technical proof toward industrial scaling. The strategic question is no longer only whether a vehicle can drive itself. It is where, at what cost and with which operating model the service can expand. In February 2026, Waymo said it was beginning fully autonomous operations with the sixth-generation Driver. The company described a streamlined sensor configuration intended to lower cost and support multiple vehicle platforms and more demanding weather. By May, Waymo said its service had completed more than 20 million fully autonomous trips across 11+ cities and began welcoming public riders to the Ojai vehicle that debuts the new Driver. Level 4 is a service, not a feature Most consumer “autopilot” systems remain driver assistance: a human must supervise and take responsibility. A robotaxi operates without a driver inside a defined operational domain — specific roads, weather and conditions. That makes deployment as important as model performance. Lower hardware cost changes the competitive equation As sensors, compute and integration become cheaper, the advantage moves toward operations: How quickly can a new city be mapped and validated? What utilization can the fleet achieve? How much do cleaning, charging, maintenance and remote assistance cost? How safely are unusual events handled? How fast can regulation and public trust be earned? Weather is an economic variable Snow, slush, fog and road spray do not only challenge perception. They change maintenance, routing, downtime and service guarantees. Sensors need cleaning and redundancy; vehicles may drive more conservatively; the operational domain may shrink during difficult conditions. That means climate affects city selection and unit economics. “Works in snow” must include procedures, fleet design and evidence of safety, not only a model update. The robotaxi P&L Removing the driver changes the largest variable-cost category, but it does not make the ride free. Capital cost, vehicle depreciation, insurance, charging, maintenance, depots, connectivity, cleaning, remote operations and regulatory compliance remain. The winning system needs high utilization without sacrificing reliability. What mobility companies should watch Cost per autonomous mile, not only disengagement statistics. Paid trips and utilization by operational area. Speed of expansion from testing to public service. Performance across vehicle platforms and weather. Incident handling and transparent safety evidence. Partnerships for vehicles, depots, charging and local distribution. My view on market entry Markets with predictable weather and bounded operating zones remain the easiest entry points. In Russia, a city such as Sochi would be a more logical early environment than a nationwide promise. But the sixth generation shows why harsher climates can no longer be dismissed as a permanent barrier. Mobility takeaway: autonomy becomes disruptive when the complete cost per safe, available kilometre beats the human-driven alternative. The AI model is necessary; the operating system decides the business. Updated primary sources: Waymo on sixth-generation fully autonomous operations and public riders and current operating scale .]]></yandex:full-text>
    </item>
    <item>
      <title>How Uber turns driver downtime into an AI data business</title>
      <link>https://akolosov.com/en/insights/uber-driver-idle-time-data-business-model</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/uber-driver-idle-time-data-business-model</guid>
      <pubDate>Mon, 10 Aug 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Mobility &amp; AI</category>
      <description>A product and unit-economics analysis of Uber digital tasks: how mobility platforms can turn idle capacity and distributed networks into AI data products.</description>
      <yandex:full-text><![CDATA[Uber is testing short digital tasks inside its Driver app so drivers can earn during downtime. From a product perspective, the important story is not the task itself. It is the conversion of an existing mobility network into a distributed data operation. In October 2025, Uber announced a US pilot of quick digital tasks powered by Uber AI Solutions. The company said similar tasks were already being tested in India and gave photo uploads for AI training as one example. Uber has since expanded its enterprise AI data offering around annotation, translation, editing and custom data collection. This matters because a mobility platform already possesses three expensive capabilities: a verified distributed workforce, payment infrastructure and software that can allocate work in real time. Adding a new task type is fundamentally cheaper than creating those capabilities from zero. The platform-economics logic 1. Lower participant acquisition costs The drivers and couriers are already in the application. Identity, onboarding, ratings and payments exist. Uber does not need to build a separate marketplace before testing demand for small data tasks. 2. Idle time becomes inventory Every platform has unused capacity. In mobility, it appears between trips. A well-designed digital task can turn some of those minutes into additional earnings for the participant and a new supply source for the platform. 3. Quality can be managed inside the product The same application can distribute instructions, compare answers, score accuracy and detect suspicious behaviour. This does not make quality automatic, but it creates a controllable workflow rather than an anonymous external crowd. 4. Operational data can become a B2B product Urban images, map validation, local-language samples and field observations can support mapping, speech, logistics and other AI use cases. The durable advantage is not generic labour; it is repeated access to a network operating in the physical world. What other platforms can learn The model is not limited to ride-hailing. Couriers can validate entrances and addresses. Merchandisers can document retail displays. Service technicians can classify equipment. A platform with field participants can add a “data shift” if the work is voluntary, measurable and economically meaningful. Before building it, a founder should answer four questions: Which microtasks naturally fit the participant's location and downtime? How will quality be measured independently? Why is this network better than a generic annotation marketplace? How will payment, consent and data rights remain transparent? The risks are part of the model Opaque pricing, worker fatigue, privacy and the risk of making optional tasks feel compulsory can destroy trust. Regulation will also evolve. A platform should therefore treat consent, task pricing and data provenance as product requirements rather than legal footnotes. Product takeaway: the strongest platform extension often uses an existing capability in a new market. Uber is not merely adding a feature for drivers; it is testing whether identity, payments and global field distribution can become infrastructure for enterprise AI. Primary sources: Uber newsroom announcement and Uber AI data platform release .]]></yandex:full-text>
    </item>
    <item>
      <title>Falling sales may be demand migration: how to see where the customer went</title>
      <link>https://akolosov.com/en/insights/falling-sales-demand-migration</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/falling-sales-demand-migration</guid>
      <pubDate>Fri, 07 Aug 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Market strategy</category>
      <description>A practical framework for diagnosing falling sales by tracking customer migration to cheaper products, substitutes, lower frequency and do-it-yourself behavior.</description>
      <yandex:full-text><![CDATA[When sales fall, the default explanation is often “the market is in crisis” or “marketing is weak”. Sometimes the customer has not disappeared at all. Their budget has moved to a cheaper tier, a substitute, lower frequency or a do-it-yourself solution. NIQ's Consumer Outlook for 2026 found that among surveyed consumers who felt worse off, 73 percent attributed the decline primarily to the higher cost of living. This is a global indicator, not a forecast for every category or country. But it supports a practical conclusion: pressure changes the composition of demand before it eliminates demand. Why revenue alone is a late signal Revenue tells you that something changed, not where the customer went. Teams then improve advertising, service and sales scripts inside the old model. If the customer has changed the category, frequency or price tier, those improvements may have little effect. Four directions of migration 1. Down-trading The customer buys the same category at a lower price, smaller package or simpler service level. 2. Substitution A different product solves enough of the same problem. A taxi trip becomes public transport; a consultant becomes software; a restaurant visit becomes prepared food. 3. Lower frequency The customer stays but buys less often. Annual retention may look stable while order frequency and contribution decline. 4. Do it yourself Education, tools and AI make some services easier to perform internally. The demand shifts from the finished service to tools, templates, components or guidance. Build a migration report Revenue, margin and frequency by price tier. Lost-customer interviews focused on the replacement, not satisfaction alone. Search, referral and competitor signals by category. Changes in package size, contract length and payment terms. Requests for cheaper, modular or self-service versions. Cohort movement between products and segments. Respond without damaging the brand Do not immediately discount the core product. Test a separate entry offer, smaller unit, modular package, subscription pause, self-service tier or adjacent substitute. Keep the value architecture clear so the cheaper offer does not simply move profitable customers downward. Know when not to chase Some migrated demand is unattractive. If the new segment has no margin or does not fit the company's capability, following it can accelerate the decline. The goal is not to keep every customer. It is to identify the new profit pool early. Market takeaway: do not ask only why the customer stopped buying from you. Ask what they bought, postponed or learned to do instead. That answer is often the beginning of the next product. Consumer context: NIQ Consumer Outlook: Guide to 2026 .]]></yandex:full-text>
    </item>
    <item>
      <title>Outcome-based pricing for AI SaaS: charging for value, not seats</title>
      <link>https://akolosov.com/en/insights/outcome-based-pricing-ai-saas</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/outcome-based-pricing-ai-saas</guid>
      <pubDate>Tue, 04 Aug 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>AI &amp; B2B SaaS</category>
      <description>How AI changes SaaS pricing: outcome-based and hybrid models, baseline metrics, contracts, unit economics and practical steps for B2B product teams.</description>
      <yandex:full-text><![CDATA[AI does not mean the end of SaaS subscriptions, but it weakens the logic of charging only per user. When software performs work rather than merely helping an employee, pricing increasingly moves toward usage, measurable outcomes or a hybrid of the two. The classic SaaS model offered predictability: a fixed monthly fee per seat and high gross margins. AI changes both sides of that equation. Inference and compute create variable costs for the vendor, while the customer may use the product to reduce labour, increase throughput or remove an entire manual workflow. As a result, the commercial conversation moves from “Which features are included?” to “What economic result did the product create?” Why the old pricing logic is under pressure AI has different delivery economics Traditional software can serve an additional user at a very low marginal cost. An AI agent consumes models, infrastructure and monitoring every time it works. AlixPartners estimates that variable margins for agentic AI can be materially lower than for traditional SaaS. A vendor therefore needs pricing that scales with real consumption and cost. Customers want a financial value story A CFO can defend a budget more easily when the contract connects to hours saved, write-offs reduced, revenue added or risk removed. A list of features is not enough when an AI project competes with many other transformation initiatives. Three pricing models that can coexist Usage-based: payment per task, document, minute, token or transaction. Outcome-based: payment tied to verified savings, revenue or another business KPI. Hybrid: a platform fee that covers availability and support, plus a variable component for usage or outcomes. Pure outcome pricing sounds attractive, but it is not appropriate everywhere. The vendor may not control market demand, staffing decisions or the quality of the customer's data. A hybrid structure usually allocates those risks more realistically. How unit economics changes CAC may remain high because enterprise sales still requires discovery, security review, integration and a pilot. LTV becomes more dependent on demonstrated effectiveness. Gross margin depends on model and infrastructure costs. Forecasting becomes harder because revenue and cost can both vary with usage. The product team therefore needs a commercial measurement layer: A clear baseline before implementation. Two or three metrics that both parties can audit. A BI dashboard or independent measurement process. Rules for seasonality, external shocks and changes in customer behaviour. Cost controls for model usage, retries and human review. What to do before offering outcome pricing First, choose a result that is close enough to the product's actual influence. “Increase company profit” is too broad; “reduce manual processing time per invoice” is measurable. Second, run the pilot with the future pricing metric already instrumented. Third, write the baseline and exceptions into the contract before the result is known. Outcome pricing is particularly powerful for a younger vendor entering an enterprise account. “If we do not create the agreed value, the variable fee is zero” can reduce buyer risk. But the promise works only if the vendor understands the customer's economics and can survive variance in payment timing. Product takeaway: do not select a pricing model because it sounds modern. Start with the value mechanism, identify what the product genuinely controls, measure it independently and then divide risk between the customer and vendor. Further reading: AlixPartners on AI SaaS pricing and its analysis of usage- and outcome-based models .]]></yandex:full-text>
    </item>
    <item>
      <title>AI business models beyond cost cutting: where new revenue actually appears</title>
      <link>https://akolosov.com/en/insights/ai-business-models-beyond-cost-cutting</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/ai-business-models-beyond-cost-cutting</guid>
      <pubDate>Sat, 01 Aug 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>AI strategy</category>
      <description>How founders can use AI to create new revenue, products and data advantages rather than treating it only as a tool for automation and headcount reduction.</description>
      <yandex:full-text><![CDATA[Most AI presentations start with savings: fewer manual operations, faster support and lower production cost. Useful, but incomplete. The larger opportunity is to create a product, revenue stream or market that could not exist with the previous cost and speed structure. The 2025 McKinsey global survey illustrates the gap. Eighty-eight percent of respondents said their organizations regularly used AI in at least one function, yet most were still experimenting or piloting. Only 39 percent reported enterprise-level EBIT impact. Adoption is broad; transformation is not. Efficiency is the entry point Automation is measurable and easier to approve. A company reduces handling time, improves code delivery or supports more customers with the same team. The danger is that every competitor can buy a similar efficiency tool. Cost advantage erodes when the capability becomes standard. New revenue starts with a changed constraint AI makes some activities cheaper, faster or individually configurable. Ask what becomes sellable after that change: A service that was too expensive for smaller customers. A product personalized for each account. A workflow delivered continuously rather than as a quarterly project. A result priced by usage or outcome instead of access. A data product created as a by-product of operations. Five patterns worth testing 1. Expert workflow as software AI can package part of an analyst, planner or operator workflow into a repeatable product. The value is not generic chat; it is domain data, process design, verification and integration. 2. Personalized production at scale Content, recommendations, training and interfaces can adapt to the account while sharing one production system. The business model may combine a platform fee with usage. 3. Outcome delivery If the system completes a workflow rather than merely providing a tool, the price can move closer to the verified business result. This requires measurement and clear control boundaries. 4. Data network effects A product improves as it sees more legitimate, consented and relevant operational data. The moat is created by the feedback loop, not by calling an external model. 5. Human-and-agent operations A service can combine AI speed with human responsibility for exceptions. This hybrid may reach markets where full automation is still unsafe or untrusted. Questions for a founder Which customer constraint changes because of AI? What new willingness to pay appears? Which part is proprietary: data, workflow, distribution or trust? How do variable model costs affect margin? Where must a human remain accountable? What gets stronger after the thousandth transaction? AI strategy takeaway: do not measure only how many hours AI removes. Measure how much new value the company can deliver and capture because the old constraint has disappeared. Updated data: McKinsey, The State of AI 2025 .]]></yandex:full-text>
    </item>
    <item>
      <title>The eternal beta syndrome: five traps that delay product launches</title>
      <link>https://akolosov.com/en/insights/eternal-beta-product-launch-traps</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/eternal-beta-product-launch-traps</guid>
      <pubDate>Wed, 29 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Product management</category>
      <description>Five product-management anti-patterns that keep teams from launching: feature creep, perfectionism, endless refactoring, consensus and excessive testing.</description>
      <yandex:full-text><![CDATA[A product that is “almost ready” for several quarters usually has a decision problem, not a development problem. Eternal beta is a mix of perfectionism, fear of error and unclear accountability. The cure is not lower quality; it is a clearer definition of what must be learned from the release. I have seen this pattern in startups and large corporations. Teams work intensely, add requirements, improve architecture and schedule more reviews. Everyone appears busy, but the most important event never happens: a real customer uses the product and gives evidence. Trap 1: feature creep The team adds functionality faster than customers can validate it. Every stakeholder has a reasonable request, but together they turn the first release into a complete platform. What to do: define one North Star learning goal for the launch. Prioritise with RICE or ICE, but use the framework to remove work, not merely rank an oversized backlog. If a feature does not change the decision you will make after the pilot, it probably does not belong in the MVP. Trap 2: hunting for visual perfection Designers and frontend engineers polish every screen before the core hypothesis has been tested. Quality matters, but visual precision cannot rescue an unwanted product. What to do: set a fixed polishing budget — for example one sprint — and define the minimum accessibility, trust and usability bar. After that, expose the product to a small real audience and improve it with evidence. Trap 3: endless refactoring Engineers are understandably uncomfortable with technical debt. The problem begins when internal elegance becomes more important than delivery. The market can move while the team perfects code that no customer has used. What to do: make technical debt explicit. Reserve a stable percentage of each sprint for architecture and reliability, while protecting the capacity required to deliver customer value. Critical security or data risks remain blockers; aesthetic code improvements do not. Trap 4: the council of elders When accountability is distributed across many committees, everyone can influence the product and nobody owns the outcome. In a large corporate environment, I watched a feature worth about EUR 15,000 spend eight weeks in discussion while a much larger commercial opportunity disappeared. What to do: assign one directly responsible individual for the release decision. Consultation should be broad, but decision rights must be narrow and visible. Trap 5: tests without a release strategy Quality assurance can become a reason to postpone any contact with the market. The alternative is not to ship a dangerous product. It is to match the release method to the risk. What to do: automate the critical path, use feature flags, roll out to 5–10% of the audience and monitor predefined stop metrics. A reversible release is often safer than a huge launch after months without production feedback. A practical launch contract Before development, the team should agree on five things: The customer problem being tested. The smallest complete user journey. The metric that will support a continue, change or stop decision. The risks that genuinely block release. The person who makes the final launch decision. Product takeaway: an MVP is not a poor version of the final product. It is the smallest credible system that can replace internal opinions with customer evidence.]]></yandex:full-text>
    </item>
    <item>
      <title>Business rotation versus pivot: how to move toward demand without destroying the core</title>
      <link>https://akolosov.com/en/insights/business-rotation-vs-pivot</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/business-rotation-vs-pivot</guid>
      <pubDate>Sun, 26 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Business strategy</category>
      <description>A practical strategy for redirecting resources toward growing demand while preserving the capabilities, customers and cash flow that already work.</description>
      <yandex:full-text><![CDATA[When an industry loses growth, founders often hear one word: pivot. But replacing the product, customer and business model at the same time can destroy the capabilities and cash flow that still work. In many cases the better move is rotation. Rotation means redirecting attention, resources and positioning toward growing demand while using the company's existing assets. It is not denial and not a cosmetic rebrand. It changes the trajectory without pretending the company is starting from zero. Four signals that rotation may be necessary 1. The growth driver has moved A smaller segment or adjacent service is growing faster than the core, while the core margin is becoming thinner. The new demand may already exist inside the revenue mix but remain hidden by totals. 2. Customers repeatedly ask for an adjacent outcome One custom request is noise. The same request from multiple customers is product evidence. If the team keeps delivering a similar workaround, it may be building the next offer without naming it. 3. Existing acquisition channels are weakening Organic demand, referrals or conversion decline while the cost of preserving the old growth rate rises. More advertising cannot permanently repair a shrinking need. 4. The unofficial business is becoming repeatable Teams often protect the formal strategy while revenue comes from “exceptions”. When exceptions share a customer, problem and delivery pattern, they deserve a strategic test. How rotation differs from a panic pivot A pivot asks, “What completely different company should we become?” Rotation asks, “Which existing capability has more value in the direction demand is moving?” That capability may be distribution, data, operational infrastructure, trust, supplier access or a technical platform. A disciplined rotation process Review revenue and contribution margin by product, segment and channel every month. Identify adjacent demand that has appeared repeatedly for at least several cycles. Define which existing capability creates an unfair starting advantage. Run an 8–12 week experiment with a real price and delivery model. Measure retention, attach rate, payback, gross margin and operational effort. Move resources gradually while protecting the cash-generating core. Use a portfolio, not a slogan A practical allocation can reserve most resources for the core, a meaningful minority for the rotation and a small share for longer-term exploration. The exact percentages depend on cash position and urgency. The principle is more important: do not bet the company before the new demand produces evidence. A company should be a machine capable of turning, not a monument to the founder's first idea. Strategy takeaway: preserve the mission and question the form. Rotation works when the company follows the customer's changing problem using capabilities it already knows how to operate.]]></yandex:full-text>
    </item>
    <item>
      <title>How to choose a co-founder: the lesson from a startup I never launched</title>
      <link>https://akolosov.com/en/insights/how-to-choose-business-partner-cofounder</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/how-to-choose-business-partner-cofounder</guid>
      <pubDate>Thu, 23 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Founder lessons</category>
      <description>A founder lesson from the unlaunched 2Know education platform: how commitment, timing, roles, conflict and risk alignment determine co-founder compatibility.</description>
      <yandex:full-text><![CDATA[In 2014 I had an idea for an online-education platform called 2Know. We built a prototype, presented the project at a startup competition and then stopped. Two years later, similar models were becoming large companies. The hardest part was knowing that the opportunity had not failed — we had failed to organize ourselves around it. I approached a close friend who was a doctor and lecturer. His domain expertise was exactly what the first medical courses needed. The idea interested both of us, but our practical commitment was different. He worked very long hours. I was operating two hospitality businesses. I wanted us to go all in; our calendars said otherwise. Friendship is not founder compatibility Trust is valuable, but a startup also requires compatible ambition, risk tolerance, pace and availability. People can respect each other and still be a poor founding team for a specific company. Discuss commitment in observable terms “I am committed” is too abstract. Ask how many hours each person will work, when full-time participation begins, what income they require, how long they can operate without salary and which personal constraints cannot move. Make the role asymmetry explicit Founders do not need identical roles. They need complementary responsibility with no invisible gap between them. Who owns product, sales, technology, operations and fundraising? Who makes the final call when the functions conflict? Test the partnership before dividing equity Run a real project together: customer interviews, a paid pilot or a four-week product sprint. Pressure reveals communication, pace and standards better than a theoretical discussion. Use vesting so long-term ownership follows long-term contribution. Write down the difficult scenarios What happens if one founder cannot go full time? How are salaries and expenses approved? Which decisions require unanimity? How is deadlock resolved? What happens to equity if somebody leaves? Can a founder start another project? How will poor performance be discussed? Do not outsource conviction I expected my partner to see the opportunity exactly as I saw it. That was unfair. A founder must arrive at conviction independently. Persuasion can start a project, but it rarely sustains years of uncertainty. Founder takeaway: choose a co-founder for the company you are about to build, not only for the relationship you already have. Alignment is demonstrated through decisions, time and shared risk.]]></yandex:full-text>
    </item>
    <item>
      <title>Why I choose entrepreneurship after almost 20 years in business</title>
      <link>https://akolosov.com/en/insights/why-i-choose-entrepreneurship</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/why-i-choose-entrepreneurship</guid>
      <pubDate>Mon, 20 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Entrepreneurship</category>
      <description>Alexander Kolosov explains why he remains an entrepreneur: freedom to choose problems, accelerated learning, responsibility and long-term economic upside.</description>
      <yandex:full-text><![CDATA[I became an entrepreneur because it gave me four things at once: freedom to choose problems, faster learning, direct responsibility and the possibility of disproportionate economic upside. Money proved the model could work, but it was not enough to keep me in business for almost two decades. My first business appeared when I was 19. At night I cleaned industrial floors and polished floors in jewellery stores. I studied, worked as a DJ and earned relatively little. Then a friend and I organised a party at the 1000 Miles club. It went well, and one evening generated more than my monthly income from the other work. We built a company around relaunching restaurants and nightclubs that had failed. During the first six months, we earned more than USD 20,000 in profit — serious money for me at that age. The number mattered because it changed my belief about what was possible. I could create value as an entrepreneur. Freedom means choosing the problem Entrepreneurial freedom is not the absence of obligations. It is the ability to decide which difficult problem deserves your attention. One period can be devoted to an IT product, another to mobility, logistics, education or consulting. The world is too large to live only one professional life. That freedom comes with instability and stress. You can choose the problem, but you cannot choose whether the market accepts your solution. The fastest learning system I know In a well-defined corporate role, a person may deepen two or three competencies over a year. In a new business, the founder repeatedly becomes a beginner: financial model, product, hiring, sales, fundraising, operations and crisis management can all arrive in the same week. No MBA can fully reproduce that feedback loop. Business education is useful for frameworks, networks and reflection. Entrepreneurship tests whether you can apply them while the outcome remains uncertain. Responsibility is an honest metric An entrepreneur voluntarily works close to the front line. A mistake is yours. A victory is yours too. That direct relationship between decisions and consequences prevents complacency. The most useful comparison is not with another founder's highlight reel. It is with the quality of your own decisions five or ten years ago. Revenue matters, but decision quality compounds before financial results become visible. Money is part of the answer Entrepreneurship can create more wealth than employment, and it would be artificial to deny that motivation. It also removes stability, concentrates risk and can consume years without a return. The relevant question is whether the combination of autonomy, learning, responsibility and upside fits the life you want. Global Entrepreneurship Monitor research shows that founders combine several motivations: independence, wealth creation, making a difference and sometimes necessity. They are not mutually exclusive. My own mix has changed over time, but the desire to build and learn has remained. Founder takeaway: do not ask only whether entrepreneurship can make you money. Ask whether you want the daily operating system that comes with it: uncertainty, accelerated learning and responsibility without a place to hide. Context: Global Entrepreneurship Monitor 2021/2022 report .]]></yandex:full-text>
    </item>
    <item>
      <title>What I learned by joining a global corporation after 15 years as an entrepreneur</title>
      <link>https://akolosov.com/en/insights/joining-corporation-after-entrepreneurship</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/joining-corporation-after-entrepreneurship</guid>
      <pubDate>Fri, 17 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Career &amp; leadership</category>
      <description>First-hand lessons from joining Capgemini in Germany after 15 years in business: responsibility, decision speed, systems, stability and founder development.</description>
      <yandex:full-text><![CDATA[After 15 years of building companies, I accepted a senior product role at Capgemini in Germany. I did not go there because entrepreneurship had stopped working. I wanted to understand how a global organization makes decisions, manages risk and delivers products at a scale a founder rarely sees from the outside. The scale is worth understanding. Capgemini reported 2025 revenue of €22.5 billion and more than 423,000 employees after its acquisitions. My own experience was earlier, but even then the difference from an entrepreneurial company was enormous. Responsibility becomes distributed A founder cannot send the final responsibility elsewhere. In a large company, ownership is divided across functions, committees and governance. That distribution protects the organization, but it can also make difficult decisions feel ownerless. I learned to distinguish healthy governance from responsibility avoidance. The first makes risk visible. The second creates another meeting because nobody wants to sign their name under uncertainty. Speed has a different meaning In a startup, speed means shortening the loop between observation, decision and customer feedback. In a corporation, speed must also account for security, procurement, legal constraints, integration and many markets. A founder may see bureaucracy where the organization sees accumulated risk. The lesson I brought back was not to copy the process. It was to identify which controls are necessary at our current scale and which are only habits. Stability changes the quality of attention A salary, calendar and predictable resources remove some founder anxiety. That can create deeper focus. It can also reduce urgency. Neither environment is morally superior; they train different muscles. Large systems expose invisible dependencies Entrepreneurs often underestimate architecture, documentation and change management until the company becomes complex. In a global organization, one small product change can affect contracts, data, support, countries and internal platforms. Seeing those dependencies made me more disciplined when designing B2B products. What founders can learn from corporations Write decisions so the team can execute without the founder in the room. Separate reversible experiments from decisions with regulatory or financial risk. Invest in architecture before complexity turns every change into a negotiation. Build a professional cadence for planning, review and learning. Do not confuse constant urgency with high performance. What corporations can learn from founders Give one person clear ownership of the result. Bring customer evidence into the decision earlier. Time-box reversible decisions instead of seeking perfect consensus. Measure the cost of delay, not only the risk of action. Leadership takeaway: moving between entrepreneurship and corporate work is not a step up or down. It is a way to learn two operating systems. The advantage appears when you can combine founder ownership with institutional discipline. Updated company context: Capgemini full-year 2025 results .]]></yandex:full-text>
    </item>
    <item>
      <title>How to build an MVP with AI and no-code without fooling yourself</title>
      <link>https://akolosov.com/en/insights/build-mvp-with-ai-no-code</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/build-mvp-with-ai-no-code</guid>
      <pubDate>Tue, 14 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>AI product building</category>
      <description>A practical workflow for building and testing an MVP with AI coding tools, no-code automation and managed services while avoiding false validation.</description>
      <yandex:full-text><![CDATA[AI has made it possible to assemble a credible prototype in days. It has not made customer demand, distribution, security or unit economics automatic. The danger is now different: founders can build the wrong thing faster and more beautifully. A weekend MVP is useful when it answers one expensive question. Will a specific customer share the data? Will they complete the workflow? Will they pay? Can the service be delivered at an acceptable variable cost? If the prototype does not reduce uncertainty, it is a demo rather than an experiment. Step 1: choose the assumption, not the feature list Write the riskiest assumption in one sentence: “We believe operations managers will pay for automatic reconciliation because the current manual process takes two days.” Everything in the MVP should test that sentence. Step 2: use AI for research, not invented certainty A language model can map competitors, draft interview questions and structure a hypothesis. It can also confidently invent market numbers. Require source links, open the sources and separate evidence from model-generated interpretation. Step 3: build only the visible critical path Modern AI coding assistants, visual UI builders, managed databases and automation platforms can connect a landing page, interface, workflow and notifications quickly. Do not automate what can be delivered manually during the first test. A human behind the interface is acceptable if the customer experience and economic event are real. Step 4: include the transaction A signup is weak evidence. Ask for a paid pilot, deposit, letter of intent with concrete conditions or another meaningful commitment. The customer should give up something — money, time, data or access — that proves the problem matters. Step 5: define the test before traffic arrives Who exactly should see the MVP? Which action represents activation? What result counts as a successful experiment? Which result means stop or change direction? How will delivery cost and manual effort be recorded? Step 6: add production discipline only after evidence A prototype is not a production system. Before real scale, review security, privacy, model reliability, access control, observability, backup, legal obligations and support. AI reduces the cost of learning; it does not remove engineering responsibility. A realistic weekend outcome The strongest result is not “we built an app”. It is “we spoke to ten relevant customers, three completed the workflow, one agreed to a paid pilot, and we discovered that onboarding is the main constraint.” That is information on which a company can be built. Founder takeaway: use AI and no-code to shorten the distance between a hypothesis and evidence. Do not use them to postpone the uncomfortable conversations that actually validate a business.]]></yandex:full-text>
    </item>
    <item>
      <title>Go-to-market strategy for startups: how to reach the first repeatable sales</title>
      <link>https://akolosov.com/en/insights/go-to-market-strategy-for-startups</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/go-to-market-strategy-for-startups</guid>
      <pubDate>Sat, 11 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Product strategy</category>
      <description>A practical GTM framework for startups: ideal customer, positioning, pricing, acquisition, sales motion, launch sequence and metrics for repeatable growth.</description>
      <yandex:full-text><![CDATA[A good product can fail because the market never understood who it was for, why it was different or how to buy it. Go-to-market strategy turns those separate questions into one operating system. GTM is not a marketing campaign and not a launch calendar. It is the set of choices that connects a specific customer problem to acquisition, sale, onboarding and retention. For an early startup, the goal is not maximum reach. The goal is the first repeatable path from attention to value. 1. Define the narrowest useful market “Small businesses” or “corporate clients” is not an ICP. A useful ideal customer profile includes context: industry, size, geography, current process, urgency, decision-maker and a trigger that makes the problem expensive today. The narrower starting point improves the product message, sales list and onboarding. Expansion becomes easier after one segment works. 2. Describe the problem in the customer's language Customers rarely buy “AI”, “automation” or “a platform”. They buy fewer failed deliveries, faster reimbursement, lower idle fleet time or a shorter sales cycle. The problem statement should sound like something the buyer would say in a meeting. 3. Build positioning around the alternative Your competitor may be another startup, an Excel file, an outsourced team or simply doing nothing. Positioning becomes sharper when it explains why the product is better than the real alternative, not only why its feature list is longer. 4. Choose a pricing hypothesis Price communicates the value mechanism. Per-seat pricing fits some collaboration products; usage, transaction and outcome components fit others. Test the price during discovery. “Would you use it?” is weak evidence without “Would you pay this amount?” 5. Match the channel to the deal A low-cost self-service product cannot support an enterprise sales process. A complex B2B platform rarely closes through a landing page alone. Decide whether the first motion is founder-led sales, partners, outbound, product-led adoption or a focused combination. 6. Design the complete path to value The sale is not the end of GTM. Map what happens after signature: data access, integration, training, first result and renewal. A product with strong demand can still fail if time-to-value is too long or onboarding requires heroic manual work. 7. Measure the assumptions Qualified conversations by segment. Conversion between discovery, pilot and paid use. Sales-cycle length and acquisition cost by channel. Time to first measurable value. Activation, retention and expansion. Contribution margin and CAC payback. A one-page GTM test Before a large launch, write one page with the ICP, urgent problem, alternative, promise, proof, price, channel, sales owner, onboarding path and three success metrics. If the team cannot agree on that page, more advertising will not solve the disagreement. Product takeaway: a GTM strategy is a chain. If one link — segment, message, price, channel or onboarding — is based only on opinion, test that link before increasing the budget.]]></yandex:full-text>
    </item>
    <item>
      <title>Unit economics for startups: the numbers to calculate before scaling</title>
      <link>https://akolosov.com/en/insights/unit-economics-for-startups-guide</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/unit-economics-for-startups-guide</guid>
      <pubDate>Wed, 08 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Business economics</category>
      <description>A practical founder guide to contribution margin, CAC, LTV, payback and break-even: how to test whether a startup can scale before spending heavily.</description>
      <yandex:full-text><![CDATA[A startup can grow revenue and still move closer to failure with every sale. Unit economics answers the question hidden behind the top line: does one additional customer, trip, order or contract create value? I learned the importance of this question through both successful and failed launches. A large market and an attractive product are not enough. If the transaction does not work at the unit level, scaling usually magnifies the problem. First choose the correct unit The unit must represent the repeatable economic event in the business. For an e-commerce company it may be an order or customer. For a mobility platform it may be a trip, active corporate account or vehicle-month. For SaaS it is often a customer account or recurring contract. A weak choice hides the truth. If acquisition happens at account level but revenue is measured per user, the model may compare incompatible numbers. Start with the unit that connects acquisition, delivery and repeat behaviour. The essential calculations Contribution margin Take revenue from the unit and subtract costs that grow directly with it: payment fees, delivery, contractor compensation, cloud inference, support attributable to usage and other variable costs. What remains contributes toward fixed costs and profit. Customer acquisition cost CAC is not only advertising spend divided by new customers. Include the sales and marketing resources required to win them, and separate channels. A blended average can hide one profitable channel and another that destroys cash. Lifetime value LTV should be built from contribution margin and retention, not from optimistic revenue. For an early startup, a range is more honest than one precise number. Show what happens under conservative, base and optimistic retention assumptions. Payback period How many months of contribution margin are needed to recover CAC? A model can show positive LTV and still run out of money because the payback is too slow. Break-even volume Divide fixed costs by contribution margin per unit. The result is the approximate number of units required to cover the operating base. Then ask whether the market and the team can realistically reach that volume. Segment before you average The most useful unit economics is segmented by customer type, product, geography and channel. An enterprise client and a small customer can have different acquisition, onboarding, support and retention patterns. Averaging them may produce a number that describes nobody. How to use the model before a launch Write down every assumption and mark whether it is evidence or opinion. Test willingness to pay before building the complete product. Run the full transaction manually to discover hidden variable costs. Model conservative retention and slower sales cycles. Compare the result with public economics of analogous companies. Update the model with real cohorts instead of defending the original spreadsheet. If you cannot measure the economic unit, you cannot improve it — and you definitely should not scale it. Founder takeaway: unit economics is not an investor slide. It is a decision system. It tells you whether to improve price, reduce delivery cost, change the segment, redesign the product or stop before a larger loss.]]></yandex:full-text>
    </item>
    <item>
      <title>How not to waste money on an IT startup: my failed launch</title>
      <link>https://akolosov.com/en/insights/how-to-avoid-wasting-money-on-startup-launch</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/how-to-avoid-wasting-money-on-startup-launch</guid>
      <pubDate>Sun, 05 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>Startup lessons</category>
      <description>A founder case study about investing RUB 6 million before validating an O2O beauty-service model, discovering broken unit economics and closing responsibly.</description>
      <yandex:full-text><![CDATA[The most expensive mistake in this startup was not the code. It was investing roughly RUB 6 million before validating the core unit economics with a real MVP. The most valuable decision was to close before belief turned into denial. In 2016, my partners and I launched an online-to-offline beauty service: stylists, hairdressers and nail specialists on demand. The Russian market appeared open, while traditional salons were growing rapidly. International examples suggested that the model could work. We conducted research, but it was not deep enough. We did not create the cheapest possible operational MVP before committing capital. We believed the product could take the market and started building. Competition was not the main problem Two competitors appeared with more than EUR 2 million in funding each. Our project was largely bootstrapped with approximately RUB 6 million. It would be easy to explain the outcome as a capital disadvantage, but that would be the wrong lesson. The fundamental problem was that the economics did not work well enough. Customer acquisition, specialist travel time, schedule density, service consistency and repeat usage created a difficult equation. More investment could have delayed the conclusion without changing it. We closed earlier than the better-funded competitors. That decision protected part of the capital and, more importantly, our reputation. A founder is not obliged to save every hypothesis. A founder is obliged to face evidence. Twelve checks I now use before a serious launch Map the market: study direct, indirect and international competitors. Run customer discovery: verify the problem, frequency and willingness to pay. Model the full economics: include acquisition, operations, support, refunds, capital and founder time. Benchmark assumptions: compare your cost structure with public evidence from similar businesses. Build an operational MVP: use manual workflows and no-code before custom development. Sell personally: the founder should learn the objections before hiring a sales team. Start with constrained marketing: prove that demand exists before buying scale. Avoid one-channel dependence: acquisition that works only through one auction is fragile. Test seasonality and density: especially in local and on-demand services. Delay external capital when possible: evidence increases both valuation and negotiating power. Choose partners through demonstrated commitment: enthusiasm in a meeting is not ownership. Define stop conditions in advance: decide which evidence will make you close or pivot. Closing can be a product decision Founders often treat closure as a personal verdict. It is more useful to treat it as portfolio discipline. A failed model can release capital, attention and reputation for a better opportunity. The danger begins when sunk costs become an argument for additional spending. A good stop decision needs three things: trusted data, a deadline and the emotional ability to separate the hypothesis from your identity. Founder takeaway: validate the operational truth before financing the technological vision. An MVP should test not only whether customers click “order”, but whether the complete transaction can create repeatable contribution margin. This article adapts my earlier founder case published on vc.ru .]]></yandex:full-text>
    </item>
    <item>
      <title>Cross-cultural business communication: what Russia, Germany, Britain and France taught me</title>
      <link>https://akolosov.com/en/insights/cross-cultural-business-communication</link>
      <guid isPermaLink="true">https://akolosov.com/en/insights/cross-cultural-business-communication</guid>
      <pubDate>Thu, 02 Jul 2026 12:00:00 GMT</pubDate>
      <author>kolosov@yoyomobility.fr (Alexander Kolosov)</author>
      <category>International business</category>
      <description>First-hand lessons from working across Russia, Germany, Britain and France: how different decision, feedback and trust-building styles affect international business.</description>
      <yandex:full-text><![CDATA[International business is rarely lost because somebody translated a word incorrectly. It is lost when both sides use the same words but attach different meanings to speed, trust, disagreement and responsibility. I have built businesses in Russia and France and worked with teams in Germany and the United Kingdom. Each move forced me to rebuild part of my operating system. What looked like indecision to me could be a demand for evidence. What sounded like agreement could be polite doubt. What I considered healthy speed could look like an unnecessary risk. National culture is not a personality test. Individuals differ, companies create their own norms, and stereotypes are dangerous. But cultural patterns are still useful when treated as hypotheses rather than verdicts. Russia: progress through uncertainty Russian entrepreneurial teams often become very good at acting with incomplete information. Constraints change, plans break and personal trust can matter as much as process. The advantage is speed and ingenuity. The risk is that an organization starts relying on heroic improvisation where a repeatable system should exist. Germany: evidence before movement In German corporate work I encountered a stronger preference for preparation, clear ownership and documented decisions. At first, the pace can feel slow to an entrepreneur. Then you see the other side: once the decision is made, implementation is more predictable because dependencies have already been discussed. The practical lesson is not “write more documents”. It is to show your logic. What exactly are we deciding? Which assumptions are facts? Who owns the risk? What happens if the plan fails? Britain: listen to what is not said British communication can be diplomatic and indirect. A phrase such as “that is interesting” is not always approval. Context, tone and follow-up matter. If you come from a more direct culture, the safest response is to confirm the decision in plain language rather than interpret politeness as commitment. France: win the argument before the calendar In France I learned the value placed on intellectual coherence and a well-constructed position. A meeting may spend more time debating the logic behind a proposal before moving to execution. Relationships and formality also matter, especially at the beginning. For a foreign founder, patience is not passive. It is part of market entry: understanding who influences the decision, how credibility is established and which objections must be resolved before action becomes possible. A practical protocol for global teams Agree whether a meeting is for discussion, recommendation or final decision. Write down the owner, deadline and definition of “done”. Ask how disagreement is normally expressed inside the team. Separate urgency from importance; cultures interpret both differently. Confirm decisions after the meeting in simple, neutral language. Adapt the communication style without compromising the substance. International takeaway: localization is not only language, pricing and regulation. It is also the operating model for trust and decisions. The founder who learns that model early avoids months of invisible friction.]]></yandex:full-text>
    </item>
  </channel>
</rss>