Skip to content
HighLoad++ · November 8, 2018

tinkoff.ru Acquisition from banner click to personalized page

How business goals turned into architecture, processes, and teams around the public web

/ tinkoff.ru Acquisition · HighLoad++ 2018

Slide contents

  1. 1. tinkoff.ru Acquisition from banner click to personalized page

    How business goals turned into architecture, processes, and teams around the public web

  2. 2. What this talk covers

    Goals and engineering principles

    Process and performance marketing

    Frontend and teams

    SEO, personalization, tests

  3. 3. What efficient acquisition means

    Three goals, one system

    Many and cheap — Channels scale, not isolated wins

    Good UI/UX — Banner-to-application path is fast and relevant

    Scalable processes — Campaigns and tests without release bottlenecks

  4. 4. 01. Engineering principles

    Abstraction, simplicity, feedback, and communication structure

  5. 5. The model matters more than channel details

    Process model ≠ channel details

    Entities: campaign, form, report

    The level where business meets IT

  6. 6. Universality costs more than it looks

    No universality until needed

    Compact core, specifics outside

    A simple model serves the goal

  7. 7. Without Check and Act it is not a loop

    Plan/Do: goal, campaigns, tests

    Check: KPIs and deviations

    Act: bids, content, resources

  8. 8. Architecture mirrors communication

    Chaotic teams → sprawling system

    Topology holds ownership boundaries

  9. 9. 02. Acquisition process

    The acquisition loop and the team around it

  10. 10. End-to-end acquisition loop

    Campaigns, content, client, tracking, and action — one system

  11. 11. The matrix works only with explicit boundaries

    One loop for business and IT

    Business sets the unit economics

    IT covers many technology zones

    Core gives tools, products deliver features

    Works only with explicit boundaries

  12. 12. 03. Performance marketing on the web

    Automating paid search, data, and decisions

  13. 13. The channel scales only through daily work

    Campaigns: search, display, targeting

    Scaling winners, killing losers

    Analysis by channel, keyword, product

    Semantics, bids, fast hypotheses

  14. 14. Without data it is not a loop

    Campaigns, platforms, marts, and reporting in one process

  15. 15. Reporting closes the bid-management loop

    Data, marts, clustering, bids, and reporting

  16. 16. Owner holds the goal, lead holds coherence

    Product owner: goal and priorities

    Lead: technical coherence

    Developers and analysts: data and reports

  17. 17. A minimal stack, not a universal one

    Python + Postgres + Rabbit

    PHP + React + Material UI

    Metabase for reporting

  18. 18. 04. Frontend tinkoff.ru

    The main communication channel with potential clients

  19. 19. Public web: main acquisition channel

    Every business unit is a stakeholder

    Unauthenticated zone: first touch, forms

    Public web ≠ internet bank

    Every unit changes content fast

    Frontend is a delivery platform

  20. 20. Thin frontend, thick engines

    Thin frontend per product, thick backend of reusable engines

    Frontend

    Split by products

    Shared blocks, forms, tracking

    Change velocity

    Backend

    Separate services and engines

    Content, rules, forms, events

    Less duplication

  21. 21. Speed to products, reuse to engines

  22. 22. Team structure follows architecture

    Core owns engines; product teams own business functionality

  23. 23. A wrong boundary turns speed into coupling

    What stays in core and what goes to product teams

    Core

    Page and block engine

    Form engine

    Tracking, infra, contracts

    Product teams

    Business-line functionality

    Pages, scenarios, blocks

    Hypotheses, A/B, product P&L

  24. 24. 05. SEO

    Where engineering principles meet delayed feedback

  25. 25. SEO shows where clean principles stop applying

    Tasks

    Optimize product pages

    Unify pages and landings

    Behavioral factors via personalization

    Problems

    Lag between action and result

    Algorithms not transparent

    Deadline beats the perfect model

  26. 26. 06. Personalization and tests

    Conversion through hypotheses, targeting, and real-time feedback

  27. 27. Conversion grows through hypotheses, not releases

    Hypotheses without a heavy release

    Relevant content by targeting

    Device, channel, and behavior targeting

    Fast statistics back into decisions

  28. 28. Personalization sits on every user's path

    A/B tests become system architecture

    Load

    On every user's path

    Decisions must be fast and reliable

    Failures show in conversion

    Data

    Many sources of segments

    Real-time stats per variant

    Logs → business signals

  29. 29. Personalization is architecture, not an interface

  30. 30. The personalization team is a platform team

    Owners appear for domains, data, and infrastructure

    Senior product owner: goal and priorities

    Product owner: the system's backlog

    Lead and infrastructure: platform reliability

    Customers: experimentation loop

  31. 31. When principles can bend

    Exceptions usually trace back to the project triangle

    Time critical → targeted bypass

    Budget limited → local solution

    Volume grows → stricter prioritization

    Name the compromise's reason and lifetime

  32. 32. A compromise is fine when its reason is named

  33. 33. Takeaways

    Goal first — then process and architecture

    Core where needs repeat across products

    PDCA rests on tracking and analytics

    Conway's Law shapes system and teams

    efficient acquisition = process + architecture + team

  34. 34. Materials

    Talk recording — HighLoad++ 2018

    HighLoad++ 2018 abstract: highload.ru

    Evolution of tinkoff.ru over 3 Years

  35. 35. Thank you!

    polomodov.tech

    All slides and links are in the Telegram channel

    Alexander Polomodov, Development Manager, Tinkoff

    @book_cube