Deep learning can be integrated in many areas of digital marketing. Even now, the use of AI technology shows impressive results: thanks to deep learning, the number of orders grew by 170% for RTB House’s customer (Modalia.com). We talked to Anton Melekhov, Country Manager at RTB House Russia and speaker of AI Conferenceto find out how the company uses AI and what results it has shown.
Anton Melekhov is an expert in digital marketing and e-commerce. Previously, he worked for Gemius consultancy agency and managed online promotion at Independent Media publishing house.
Interviewer: AI Conference (AIC)
Respondent: Anton Melekhov (А. М.)
AIC: In the article for Rusbase you said that deep learning could make advertising activities up to 50% more efficient. How?
А. М.: From the start, AI was used by Netflix and Amazon to provide user recommendations based on their query and viewing history. As a result, sales increased and the use of the deep learning technique (a branch of machine learning in AI technology) started to gain momentum.
In digital advertising, efficiency of campaigns depends on the accurate definition of the target audience and its needs. The first step for advertisers is the use of machine learning and AI methods. However, at a later stage one should move on to use deep learning deploying its algorithms and data models for accurate analysis and identification of user needs.
E-commerce uses this approach very actively due to the wide range of offerings. The decision to show one banner or another must be taken in a millisecond basing on the online behavior of every user. Detailed analysis performed by deep learning algorithms ensures that personalized ads are shown quickly and are highly adapted to every unique user.
I will use the example from our practice – advertising campaign for Spanish multibrand online clothing store Modalia.com. We realized a personalized retargeting campaign using deep learning algorithms with the aim to increase sales and ROI.
Moreover, we offered a flexible approach to the creation of banner ads based on the recommendations mechanism that provides a more accurate selection of goods displayed on the banners. The customer needed customizable banners with seasonal special offers. They had to change dynamic banner to static banner ads featuring promotional offers.
In the end, we came to the solution capable to meet specific needs of the customer within the specified budget. The campaign helped to grow ROI from 33% to 198%, the number of orders increased by 170%, conversion rates – by 27%, and CTR – by 1.7%.
AIC: In which business areas are AI technologies in-demand most of all? Why?
A. M.: AI technologies are used in the economy sectors where handling big data is important. Today the term ‘big data’ has a synonym ‘new oil’, an important resource for the development of finance services, production, medicine, and education. Undoubtedly, any of these industries needs to promote its services and products to end consumers. It means that digital marketing will be rapidly growing and we will have more work to do.
AIC: How is deep learning used in these areas and what results has it shown? Please, give two or three examples.
A. M.: Russia adopted the development program called Data Economy 2024 intended to shift all our production and everyday life to constant interaction with AI mechanisms. For example, Yandex implemented speech recognition technologies SpeechKit Cloud to transform speech into text.
Here is the example of one of the outstanding advertising projects: last year, Rothco agency and The Times newspaper won the Grand Prix in the Creative Data category at Cannes for the reproduction of the speech of the US President John Kennedy (project called JFK Unsilenced). Using AI, they collected and processed records of 831 interviews and 233 thousand phone calls, and managed to reproduce the style of Kennedy’s speech.
AIC: In 2017, RTB House announced the algorithm for online ad retargeting within the Real-Time Bidding model based on deep learning. What operating principle does it have?
A. M.: I will say more about it, as it is an important tool of RTB campaigns, which we have significantly improved since that time.
Before you see a banner page on the Internet page, several processes take place. The whole procedure of placement and sale starts when the user initiates the attendance of the advertising platform. This activates bid queries for specific ad placement places.
Before the auction starts, the platform shares important data about the user (geographical location, device type, etc.). At the next stage, it offers places to ad exchanges that further report about available positions to their DSP partners.
Demand-side platforms (DSP) at first evaluate the value and risks of purchasing a specific ad placement basing on all available data. Then they check ad campaigns run by advertisers and assess which of them match the user most of all. An important part of this assessment is the security filtering that prevents ad displays on websites with compromising or insulting content.
After the assessment, DSP inform chosen advertisers about the specific user and available places. Advertisers choose the price they are ready to pay for an ad display at the auction. At this time, the banner design is prepared: the choice of the correct format, decision on the specific design pattern and product offer.
After the price is defined and the banner is created, time comes for the optimization of placement. Currently, as the market standard shows, platform owners collaborate with around six platforms. To optimize expenses, buyers should decide where to purchase an ad space – at an open or private auction. After the decisions round, DSP make bids on behalf of advertisers. Together with offers, they send readymade banners.
AIC:What results has the algorithm shown at this stage?
A. М.: The analysis of the campaigns’ results before and after the use of the deep learning method has shown that it has increased the efficiency of campaigns by 50%.
AIC: What will you be speaking about at AI Conference?
A. M.: I will focus on deep learning algorithms and principles of their use in retargeting. I will tell what benefits they bring in the work with product recommendations and what practical efficiency the technology provides for business in general.