X-Sell: Sales Channels Need a Shared Brain

Nikita opens with three facts about himself: he was born in Vladivostok and lived there some thirty years, studied computer security at the local university, and is a mathematician by degree — hence, he admits, a professional bias towards security, which he still handles as a manager with a security partner. He started selling at the age of ten: he hauled to the flea market anything lying loose at home — collectible caps, books, video cassettes — for pocket money; Alexander reckons his sales experience counts from there, which makes it twenty years. His most expensive mistake cost 700,000 euros: in a processing centre the accounting sums were computed by reporting-tool templates that queried the database with raw SQL, and the queries were wrong. Alexander notes that a bug's effect is usually hard to price; Nikita agrees that reliability is harder still: the scale takes ages to assess, and reputational risk sits on top — a customer got no service just as the deal of their life fell through.

The guest runs through his career: programming at university, a first job doing general IT support at his own institution, then adapting e-learning systems, where he settled on Java, and a back office for accounting at a company that connected online shops to acquiring — ledger entries, wallets, commissions, and the front end on Angular. Next came the Moscow Exchange and the Finuslugi marketplace: an interview in Vladivostok, a move, a landing on 1 April, lockdown day, a step down to developer, and a return to management three months later. He left in search of product experience: his Java interview was run by the current CTO Artem Ivanov, and Igor Maslov checked his algorithms. He disliked the first team so much that he wrote HR a full A4 page — and had nearly given up when a future manager offered him the sales domain to build from scratch. In half a year he single-handedly shipped an origination system to production, and only then hired an analyst. Today he grumbles at the monolith it became; Alexander objects that a perfect first version would have been an overbuilt contraption taking a year to launch.

Nikita boils marketing and sales down to three questions: what we sell, how we sell, and to whom. The "what" is shaping relevant offers, keeping a history the system learns customer behaviour from, and prioritising them. The "how" is channels: the mobile app, the web, field representatives, call-centre operators and offline partners at an airport; channels, in his words, must share a brain. Hence an argument about terms: cross-channel is simply the ability to deliver an offer through several channels, whereas omnichannel is shared context — a customer stops at the terms-approval step in the mobile bank, and the operator who rings them knows they got exactly that far. Alexander recalls omnichannel being sold to him differently, as knowing which channel moves a customer through the checkpoints; Nikita disagrees, calling that a communication strategy. The third question, "to whom", is classic segmentation and targeting, which takes the conversation into experiments.

Nikita describes control segments and the rule of comparing like with like, and singles out availability: you must not offer what a customer plainly cannot get, and during a test a customer may take the product and drop out of the segment. Alexander adds a complaint: a lift measured on a narrow segment gets stretched over everyone, although sometimes there is nobody to stretch it to, and instead of a scalable experiment you get a set of per-segment strategies — a mix rather than a strategy. Nikita answers that this is what triggers are for: the system running the test must learn as fast as possible that the customer has changed state. Another trouble is interference: advertising, origination and delivery run tests at once and the result falls apart, like a microscope whose table a friend shakes; the cure is end-to-end tests on shared references, not rolled out yet. Alexander also recalls metric trading: at YouTube the choice was between another million in ad revenue and losing viewers.

Nikita explains X-Sell through a shop: you came for an iPhone and also bought a case and a screen protector, so the average check went up; you were offered another phone instead, so the check fell but the customer still left with a purchase. Inside the company the point is maximising LTV, and loyalty travels with it. Alexander unpacks the metric's history: each product used to live by its own NPV — acquisition cost against what a risk model says the customer will bring, with future money discounted to today — but as the company grew to some fifty million customers the ecosystem effects kicked in, NPV started to pinch, and the model had to grow probabilities of second and third products. Nikita calls X-Sell a business platform: one business capability that can be switched on for a business line, metrics such as LTV and product coverage, domain experts beside the IT teams. Alexander sees a service line with a single front door, which also guards against the tragedy of the commons: without rules for sharing touchpoints, customers get buried in banners as a common field is trampled by extra cows.

Technically the domain rests on several systems. The offer system computes proposals in advance with a 99.9 approval rate — the round hundred was left out deliberately; its Apache Flink engine picks the terminal event within a time window when "create the offer" and "delete the offer" arrive together, while a Task Engine acts as a kill switch recalculating the whole base when, say, the key rate changes. Cassandra, the store, has ironically become the read-side bottleneck; they want 30,000 RPS out of it. The trigger platform sells not a stream of events but a subscription: millions in, one useful event out — and by predicting the right moment to call, the team raised the pick-up rate by tens of per cent. Nikita breaks his own role into narratives and calls his vision LUPA: L for lean, U for evolutionary (the letter, he admits, does not quite fit), P for partnership on shared goals, A for affordable. Alexander adds that change management fails not because the architecture is ugly but because people never bought into the change.