Personalizing products, prices and customer experience

My own introduction to personalization happened about ten years ago at an eCommerce conference. One of the speakers asked the audience: “When you see contextual ads on Google, do they adapt to you?” Everyone said: “Yes, of course.” And it is true — look at a cordless drill on one site, and drills follow you around the entire internet afterwards. The speaker went on: “And does that work on your site?” Everyone said: “No.” Back then nobody really knew what to do about it.
That was the moment I understood that sites have to adapt to customers the same way Google does. Obviously, if we arrange the products on our own site as personally as Google arranges its ads, we will earn more. Because if that were not the case, Google would not be earning as much as it does.

Personalization plays a very large strategic role for our company. Beeline does not only provide mobile, broadband and home internet services — it also runs its own online store. It matters to us that a customer finds the right service or product as fast as possible, depending on who that customer is and what brought them to us. That is exactly why we decided to build out personalization.
Effective personalization is always about shortening the path a customer takes to the action they want. For Beeline’s business that shows up as a higher CR (conversion rate) into a purchase, a SIM card order, a broadband application, an add-on service or a plan upgrade. To get there we have to be as convenient and as clear as possible for everyone — for teenagers and for older customers alike.

Many large companies today are moving towards ecosystems. Becoming an ecosystem customer is almost like getting a membership in a private club. Thanks to personalization, an ecosystem customer can get every product and service they need in a single window: changing a plan, buying a device or activating a subscription.

We measure the success of every campaign with A/B tests. It helps that the Gravity Field personalization platform not only calculates uplifts on key metrics automatically, but also gives you “probability to be best” — the likelihood, based on Bayes’ theorem, that the positive or negative result of a campaign will repeat in the future.
It is worth saying that not every hypothesis is confirmed by an A/B test, while we critically need to understand the cumulative effect of using the platform. So we once ran an experiment: we switched the Gravity Field script off for a share of site users and looked at how their metrics changed. Tellingly, average order value in the test group (users who saw personalized campaigns on the site) was 17% higher than in the control group (users who saw no personalized campaigns at all because the script was off).

Recommendations on the product page. It is certainly important to show recommendation blocks at different stages of the funnel, but the product page is the page type where the user most often makes the buying decision. OMA.BY product pages currently carry three recommendation blocks: “Similar products”, “Recently viewed” and “Bought together”. We arrived at that number and that content through a lot of A/B tests — we tried 1, 2, 3 and 4 blocks and experimented with different recommendation strategies.
Another thing we are particularly proud of is the personalized category tags on the home page. They perform very well in revenue terms, and thanks to Gravity Field they appeared on OMA.BY almost at the same time as the equivalent mechanic launched on YouTube.

Recommending in-stock products on out-of-stock product pages. What matters here is not only what goes into the blocks, but where they sit. A/B tests showed us that the higher the replacement block sits on the page, the bigger the uplift in conversion and revenue. On in-stock product pages, by contrast, a block placed lower down works better. There we use the “Bought together” strategy — it helps us cross-sell products from adjacent categories and lift average order value.

One of our most recent wins is the mechanic for a members-only wine club. When a user lands on the site, we use CRM data to determine whether they are a member and pass that to Gravity Field. Depending on the user type, Gravity Field then shows a personalized banner on the home page. Users who have not joined yet are invited to sign up and get a discount on their first order; members see a banner with their personal discount.

I had seen the low-stock mechanic at other companies and had wanted to test it for a long time: adding a “Selling out soon” badge to selected products. The hypothesis was that scarcity would push users to decide faster. Oddly enough, it was not confirmed — the campaign brought key metrics down. We never fully understood why. Most likely it had to do with the specifics of our stock: in some locations the selected products really could run out, and by the time we showed the badge the user physically could not buy them.

We have a cart-page mechanic that offers users a SIM card on a particular plan alongside the device they are buying. If the user declines, we show a message explaining that the plan can be personalized — every default parameter can be changed to fit their needs. So far we have not managed to get a positive effect out of this campaign, but we keep experimenting: analysing the data and building new hypotheses. Working with personalization is fundamentally about constant experimentation.

METRO has been running a “Best price guarantee” campaign successfully for quite a while. We show badges with that message on products priced below competitors. At some point we decided to strengthen the message — to make the badges more noticeable by adding an animated shine to them. Oddly enough, the effect was the opposite. Users are probably just irritated by marketing that pushes too hard.

I will never get tired of repeating that the main goal of working with personalization is the cumulative effect. Campaigns that fail get switched off and help you generate stronger hypotheses; campaigns that win stay with you for years and help you hit your revenue targets. What matters is to analyse campaign results not only in aggregate but across many segments: to decide on optimisation or scaling based on the value delivered to each specific type of user, not on an average across the whole audience.