The evolution of marketing: from a small online store to a hypermarket

Published on June 22, 2016

The evolution of marketing: from a small online store to a hypermarket



    This is a story about how the site of the Sotmarket online store was created and the work with traffic attraction channels changed as the company grew.

    Note by the author. Survivors mistake
    «Живые трупы», — вот так описал историк Кевин Вильсон пилотов Второй мировой.

    Шансы уцелеть при бомбардировке были такие же, как при игре в русскую рулетку с несколькими патронами. Один раз может и повезет, но если продолжать играть, никакой удачи не хватит.



    Действительно, многие бомбардировщики не возвращались на базу. А на тех, что уцелели, оставалось множество пробоин на крыльях и хвосте. Любые улучшения защиты корпуса самолета могли спасти не одну сотню жизней пилотов.

    Командование предположило: усиления брони требуют те части самолета, на которых наблюдалось больше всего пробоин. Однако то, что казалось очевидным, было ошибочным. В действительности, места пробитые снарядами напротив демонстрировали наиболее крепкие части корпуса. Ведь с этими пробоинами самолет оставался достаточно цел, чтобы долететь до базы.

    Обнаруживший ошибку математик Абрахам Вальд позже, на базе статистических методов, не только выделил наиболее уязвимые места воздушных кораблей, но и рассчитал возможную длинну полета с определенными типами повреждений корпуса.

    Систематическая ошибка выживших — это склонность фокусироваться на одной группе данных («выживших»), игнорируя или не имея данных по другой («погибших»).


    Redesign of the project. Start


    In 2010, the store approached organic traffic saturation. About 70% of orders were made through a call center and only 30% were processed through the site. It was obvious that scaling up work with paid traffic sources when most of the orders go through the call center efficiently is impossible.


    Work on the redesign took about nine months, along with rebranding. Four months after the restart, we summarized the following statistics:
    • 74% of orders were issued entirely through the site, 18% through a call back.
    • 3.7 pages viewed per session versus 2.6 before redesign.
    • The conversion to the created order increased by 32%.

    Although at the time of launch, the recommendation algorithms on the site were raw, combined with the confirmation of an order by phone, sales margins increased by 21%, and the time for order processing by the call center operator decreased.

    Particular attention in the redesign process was paid to the search engine optimization of the site, as the main source of traffic. Here it was necessary to find compromises between content for users and technical, optimization schemes. So from 2010 to 2014, organic traffic grew almost 7 times: from 0.6 to 4.1 million sessions per month (excluding brand requests) .

    In early 2014, with the help of colleagues from OWOX, we were the first in Russia to successfully implement Google Universal Analytics with more than 100 events that track user actions on the site. We also started implementing QliqView for data visualization and analysis.

    In the same year, we launched the second large-scale redesign of the online store. The key task was to adapt and personalize the site for users of mobile devices.

    Promotions, newsletters and affiliate sales have become new sales drivers.

    Stock pages are the weakest templates for the vast majority of retailers. Usually this is either textual footcloths with links to how to find certain products in the catalog, or clumsily made-up whole pictures.

    In our case, the landing page of the action was distinguished by a minimalistic cap, from which a quick transition between existing shares, the main conditions of the action and the goods participating in the action were available. This page template was focused on affiliate and direct mail traffic. Thus, the conversion into a purchase through these channels was on average 21.7% higher than the usual illustrated text page.

    Note by the author. Placebo design
    Многие интерфейсные элементы не делают ничего. Часто встречающиеся неуклюжие «продающие лендинги» обвешаны большим количеством нефункционирующих интерфейсов.

    Ограниченные предложения с таймерами обратного отсчета, при завершении которых ничего не происходит. Уведомление о том, что кто-то еще просматривает данную акцию. Возможность снизить цену на товар или услугу. Множество визуальных якорей призвано мотивировать посетителей к действию.


    Подобный плацебо дизайн можно встретить и в оффлайне. Например, во многих общественных лифтах в США и Европы кнопка закрытия дверей лифта не закрывает их, начиная с 90-х годов и ростом общественного внимания к правам инвалидов. Эти кнопки применяются только рабочими или обслуживающим персоналом и работают только с ключом. В современных конструкциях общественных лифтов эта кнопка может отсутствовать.

    Большинство перекрестков сейчас компьютеризированы, но когда-то маленькие кнопки у пешеходных переходов позволяли людям сообщить о необходимости смены сигнала светофора. В тех же США сейчас эти кнопки почти везде отключены, но задача замены или удаления их в большинстве городов была такая большая, что их просто оставили. И люди все еще нажимают на них, потому что сигнал светофора, в общем-то, меняется.

    Engineering and design in the marketing of an online store determine the possibilities of its merchandising. Most predictive algorithms are effective only if they have the correct visual context.

    Programmatically changing the background of images. Almost all photos of suppliers' products are provided on a white background, which does not allow transferring the product image to another background without re-clipping. For the implementation in the design of colored substrates without hours of manual work, we took advantage of the imperfection of human perception.

    The human eye does not see the difference in tones, without the possibility of comparison. So, for example, we do not notice the difference in color rendition of the two monitors until they are next to the identical pictures on the screen.

    So, the solution was to impose a color layer with 96-98% transparency on top of the HTML layout. At the same time, when you hover over the product block, the opposite effect was created. Removing the overlay of the “gray” background returned the brightness and contrast of the picture, making it more attractive against the background of others.



    Product blocks in the catalog. The following varieties of block design can be distinguished:
    • By status: in stock, on order, labeled (stock, new, discount), not available, in the basket. If the product is temporarily withdrawn from sale, it is worthwhile to give the customer the opportunity to at least sign up for a notification of arrival, and when the product is already in the basket, the opportunity to quickly switch to it.
    • By type of photograph: horizontal, vertical. A narrow elongated block is not suitable for everyone. For example, photographs of TVs, cameras are often wider and narrower than mobile phones.
    • The block size allowed integrating a banner of a standard size of 200x300 pixels directly into the output. Such placement usually has a high CTR due to nativeness.

    For responsive sites, the principle of a gradual increase in information is widespread. For example, when hovering over a product block, additional content is displayed. This approach, including allows you to balance the tasks of search engine optimization and the normal perception of content.



    Interface Elements The formalized GUI makes the interface not only more intuitive, but also allows you to provide quality support for the project, which several designers and developers are working on.



    Perception management. The site header is organized into three logical hierarchies:
    • Inversion combined the basic elements of navigation: logo, menu, search and basket. In this design, color contrast is the first to attract the attention of the viewer.
    • Auxiliary navigation is highlighted by enlarged pin sizes and pictograms.
    • Additional navigation going to the third level adapts to a specific section of the catalog and tells the client the most popular sections in it.


    The footer integrates the interactive sections of the client’s personal account: selected and viewed products, help tools, as well as the end-to-end user ID, which is used by the operator to view the client’s session.

    Affiliate marketing. The Age of Clickers


    image
    In 2010–2011, the active growth of traffic and orders ensured including own affiliate program. Several hundred directly connected partners attracted orders, receiving a percentage of sales.

    The year 2012 saw the active growth of CPA networks in Russia. At that time, each affiliate network had a certain amount of unique traffic. What made it expedient to connect to each of them. By the summer of that year, in less than 5 months, the company was integrated with more than 20 affiliate networks.

    In December, more than 24 million unique users visited the Sotmarket website, about 1/3 of the Runet’s audience at that time. In January-February 2013, traffic growth continued, and sales conversion began to decline.

    It is difficult to imagine the negative consequences of paying for completed orders based on our own accounting system. However, the fact is that when working with an affiliate network, the advertiser does not have the ability to visually control the quality of published advertising: sites, context, placement formats, in contrast to the direct affiliate program. This opens up great opportunities for fraud.

    Note by the author. Cheating technology
    2013 год стал одним из пиков по количеству внутри рыночных разбирательств между партнерскими сетями и рекламодателями. Сети обвиняли друг друга в различных махинациях. Рекламодатели игнорировали существующие проблемы либо предпринимали безрезультативные попытки действовать в юридическом поле.

    Вот несколько примеров различного типа фрода:
    1. Брендированный трафик. Включается когда вы ложитесь спать. Партнер в ночь или на выходные включает запрещенный вид контекстной рекламы. Работая через сайт-прокладку с авторедиректом (блог, купонный сайт и т.п.) клиенту устанавливается партнерский cookies-идентификатор и скрывается настоящий источник привлечения.
    2. Кликандер и попандер. Включается дозировано в разрешенные каналы трафика. Как и в первом случае, пользователь проходит авторедирект через сайт-прокладку. После чего он закрепляется за данным партнером. Всплывающее окно может автоматически закрываться, быть скрытым iframe-ом. Обнаружить подобные манипуляции вручную практически невозможно.
    3. Тулбары. Включается тогда, когда покупатель уже решил совершить покупку в магазине. В браузере клиента происходит все тот же авторедирект, который прерывает сессию и фиктивно возвращает клиента на страницу покупки, но уже с привязкой к определенному партнеру. Менее агрессивные способы предполагают, что пользователь самостоятельно кликнет и перезагрузит страницу. Например, нажав на предложение скидочного купона, информацию об акции или низкой цены в другом магазине.

    Партнерские сети не умеют и не заинтересованы в эффективном мониторинге, фильтрации источников привлечения клиентов.
    Развивающийся рынок мобильной рекламы сталкивается сейчас с аналогичными проблемами при работе по модели — CPI (Cost Per Instal) в более гипертрафированном виде. Небольшое количество специалистов умеющих работать с мобильной рекламой и не достаточная прозрачность мобильной аналитики создают для этого нездоровую почву.

    The main vulnerability was the attribution of orders according to the model of the last paid click. Due to the high volume and intensity of impressions, clickman and popander traffic began to re-link customers, brought mainly by organic sources, to themselves, with an insignificant real contribution to the conversion.

    In the next 3 months, we significantly updated the terms of the affiliate program and introduced a fraud filtering system.

    Traffic analysis was based on a set of technical and behavioral indicators of a user session:
    • Technical characteristics of the session: clicks, cursor movement, number of page views, etc.
    • Entry points, inside site transitions, achievement of targeted actions during the session.
    • Frequency of rewriting the traffic source in the chains of attracting users by this partner.

    So partners were classified according to the declared type of traffic, defining their reference metrics. Deviation from the given indicators outside the given corridors was processed as fraud. In such cases, the system automatically assigned the order to the previous partner in the chain that successfully passed the test.

    Along with changes in user behavior in the reference traffic sources, filter settings were also adjusted. The system was trained manually. Not without errors, when high-quality traffic was marked as fraud. In this case, the partner was rewarded manually. Although such cases were not more than 0.4% per month.

    Paid and free traffic channels


    Distribution scheme of attracted customers and expenses by traffic channels

    The chart above shows the distribution and cost of attracted customers according to the model of the last click. Brand queries are excluded from the channels of the Display and search traffic and are included in direct inputs.

    Note by the author. Technological intensity of marketing channels
    Техноемкие каналы — источники привлечения целевого трафика, требующие существенных вложений технических ресурсов и низкую эластичность роста трафика по бюджету. Т.е. повышение рекламных расходов не приводит к пропорциональному росту целевого трафика.

    Такими каналами, например, являются: поисковый трафик, директ-мейл, партнерский маркетинг. Они традиционно наиболее эффективны на рынке электронной коммерции. В долгосрочных стратегиях коммерческие каналы продвижения — это дополнения к техноемким каналам. Последние могут составлять до 60-70% в трафике и до 80% в продажах.

    При этом средний чек в зависимости от канала и источника привлечения может существенно отличаться. Поэтому, например, большое значение в партнерском маркетинге уделяются непосредственно выбору площадок и акциям; для рассылок со скидкой или подарком можно увеличить не только объем трафика и выручку, но и не потерять в валовой прибыли. Так, зачастую больше половины перешедших из писем совершат покупку не воспользовавшись акцией.

    Современный маркетинг не может успешно развиваться без айти и продвинутой аналитики. Грамотное управление техническими ресурсами влияет не только рост выручки, но и сокращает рекламные расходы.

    Calculation of advertising costs for an order or an attracted customer is the simplest and most common method of managing advertising costs, which is often followed by small online stores. With the expansion of the range of goods, services and the increase in the marketing mix, there is an organic transition from CPO to ROMI . The traffic attribution system also changes: from last-click accounting to distributed models, when each channel, source in the customer’s attraction chain gains weight in proportion to its contribution to the conversion.

    It is also wise to consider marketing discounts in advertising costs - these are discounts provided to attract and retain customers. For example, in direct mail channels, social networks, coupon and cash back traffic, the share of discounts is the largest. However, given the high average bill, for example, subscribers can bring up to 25% of revenue. In addition, newsletters make a high contribution to the conversion of other traffic channels.



    The best mass mailing is inferior in trigger productivity by 3-5 times. Therefore, work on any project should begin with them. The task of mass mailings is to build relationships with the client, informing and solving current tactical problems.

    Bulk mailings:
    • Promotions and discounts - regular special offers, sometimes exclusively for subscribers.
    • Seasonal and thematic - new items for the summer, winter seasons or for significant holidays.
    • Content and news - information about new products that are on sale, about how you can use products, company products.

    On average, 12-16% of customers visiting the site are authorized, i.e. they know contact details. The level of customer authorization is important - this is the proportion of the audience with which direct mail communication is possible.

    Trigger mailings:
    • Behavioral - related to user behavior on the company's website. For example, an abandoned basket.
    • Event - related to external events. For example, birthday, local holidays, etc.
    • Transactional - related to purchases and customer relationship history. For example, related products to order.

    Contactless newsletters


    Up to 9% of the traffic on the retailers' website can be made by users of cash back (Sberbank Thank you, Mnogo.ru, Svyaznoy Club, etc.) or coupon services (Groupon, Biglion, KupiKupon, etc.). These are customers who have already shared their contacts with these companies. The problem is that the data is often not even aware of the fact that the user has visited a particular store, and even more so about the interest in certain products.

    This audience is interesting in that it is usually attracted by the affiliate model, i.e. Payment for advertising is made for a completed transaction .

    In order to fill the gap between the transition of a user registered with a partner and his interests in products on our website, we came up with a fairly simple way to exchange information without transmitting personal and contact details of customers:
    • A pixel in the parameters of which the partner substituted the unique identifier of the registered user when opening a letter or visiting the site. A similar technology (Cookies Matching) is used, for example, in RTB systems.
    • An XML catalog in which information was stored for direct mailing. Similar to what the store is preparing, for example, for Yandex.Market
    • The URL parser that saved the address and parsed the parameters of the visit page is asynchronous JavaScript.
    • PostgreSQL (later the Hadoop, Hbase stack) is a data warehouse that stores information about user sessions.
    • The API that was responsible for the data exchange: sent pictures, products, discounts, the logic of the chain of letters to a partner for a particular user.

    So, the user who opened the affiliate newsletter or logged in to the site was set cookies, which allows you to further monitor his actions when moving to a retailer. As soon as he visited the store’s website and, for example, formed and threw a basket, the system sent the relevant information to the partner.

    The key to the data exchange is the user ID. Formation of the newsletter took place on the side of the partner, contact details were provided to them before sending the message. Similar triggers added 25.5% of channel sales.

    It is worth noting that no marketing can replace the service and high-quality operational customer service. However, being the third largest expense item of an online store, it can become either an excruciating ballast or a powerful sales engine.