Brand presence

About

Companies and brands have always invested in having a presence; either by appearing in certain ads or through influencers on social media.

However, this investment is useless if there is no traceability of where the brand has appeared, for how long, for how many people and, most importantly, what impact and return on investment we have had on our business; as well as answering other crucial questions.

Idea

Providing traceability of our brand presence is a task that involves a lot of effort due to the volume of data processing that it requires. A successful analysis of this data will help to carry out market researches and correlations to know where to invest in future campaigns to maximize our business objectives. The objective of this experiment is to be able to exploit this huge amount of information that is generated from different data sources and its subsequent analysis for decision-making.

Thanks to artificial intelligence and parallel data processing, this is now possible.

Process

First, we must bring together all the data sources where our brand appears (TV channels, radio, social networks, Twitch, YouTube channels, etc.). Several processes in the Azure cloud will be in charge of carrying it out. Subsequently, thanks to parallel data processing and sophisticated Artificial Intelligence systems from both Cognitive Services and ad-hoc trained neural networks (all this also put into production in Azure), our brand will be able to be identified. In addition to this, the current audience level, as well as other relevant information, will be extracted at all times. Then all this information will be stored in a corporate Data Warehouse after an ETL process that will bring together all the information. Finally, an analysis will be carried out in Power BI to be able to transform all this information into knowledge that can be exploited by the analysts for decision making.

Utility

Obtaining brand presence is a good practice where a large part of capital is invested. Providing traceability and being able to analyze the impact it has is a difficult task if it’s not done with the right tools.

Thanks to Artificial Intelligence and Big Data techniques, we offer to bring all this knowledge to analysts in order to answer all their questions, maximize return on investment (ROI), as well as achieve business objectives.

Introduction

Companies have always struggled to have presence in the sector. Each year, companies invest huge amounts of money to appear in strategic locations and to be able to achieve a greater volume of sales, customers and what–nots. But, do they conduct studies and traceability of:

• Where has my brand appeared?

• For how long?

• How many people have seen my brand?

• In which medias?

• What impact has it had?

There is no doubt about the repercussion that having a presence in these strategic places provides. However, if we enrich it with a posteriori studies that are capable of providing us with information for decision-making, our return on investment (ROI) is guaranteed.

This problem creates real headaches when having to deal with large data from different sources and their subsequent analysis. But, thanks to artificial intelligence and parallel data processing, this is possible.

Architecture

The high-level architecture capable of satisfying the earlier mentioned needs would be the following:

Roughly speaking, what we would have would be a series of origins from which we will analyze when the brand is present and how many viewers it is reaching. Among these origins, we can find:

• Twitch platform

• YouTube

• Sports such as football, tennis, MotoGP, basketball…

• TVchannels

• Radio

• Socialnetworks like Instagram, Facebook…

These are just a few examples of the wide range of possibilities that we have, being able to add as many as we want. Let’s see it a little more in detail:

We would start from the base of having a number of services in Python that would be in charge of capturing and connecting to the different APIs and sources. In addition to connecting, they would make the necessary cuts to keep the data of interest.

This data would then be sent to both Azure Cognitive Services and custom AI models that will be the responsibles for identifying the brand in each of the data sources.

Finally, all this information will travel through Azure Data Factory where the Business Intelligence’s own Extract, Transform and Load (ETL) process will take place and this information will be transformed, analyzed and joined to the other data in our central data unit constituted by the Data Warehouse.

This last part will be where we nurture the subsequent analyses performed with PowerBI that will be what we make available to the corresponding department for decision making.

Preparation

The sources from which we are going to extract the information are totally heterogeneous, which forces us to carry out a set of techniques and scripts to capture the data that we need. In this case, we decided to use Python programming language, which offers a suite of tools capable of executing this heavy task.

For different social networks we have at our disposal the corresponding API to obtain the desired information (publications, likes, visits, etc.). And, both from Twitch and the different television channels, we have direct access to the signal.

Later and thanks to artificial intelligence, we will be able to detect if the brand is present, like in the next images:

Services

Once the information has been collected, different services will be in charge of detecting the presence of the brands in each image. Apart from customized models that will improve and enrich our system, we will use Computer Vision, which is one of Azure’s Cognitive Services:

Analysis of the information

Once we have the architecture assembled and the data has been processed by the Cognitive Services, the next step is extracting information from the pre-processed data to make it available to the analysis departments, which will use it for decision making.

To tackle this last part, we have decided to make use of a great ally such as Power BI.

Understanding the information

Next, we will explain how the Power BI’s information is structured.

At the top we find the sources:

• TV channels (for example)
o Tele 5.

And within each channel we can find different categories, like:

• Programs
o The comedy club => We detected a brand on:
 the stage
 the comedian’s clothes
-> We identify the comedians and detect the number of spectators

• Sports
o Football matches => We detect a brand on:
 the stadium
 the players´ clothing
-> We identify those players and detect number of spectators

• Ads (again, same process)

In Social networks:

• Instagram
o FC Barcelona account
 Publications: We detect the brands
-> We identify the player and identify the number of visits and likes

• Twitch: Same concept as the previous ones. We would be able to navigate the different streamers, identify them and detect how long a brand has been seen on the screen and for how many viewers.

• And so on…

Analysis

“Impact” is the effect caused by each of the elements. For example, the impact on a football broadcast would be calculated as the product between the seconds that a brand appears times the number of spectators. However, in a social media post the impact will be equal to its reached viewers.

The “impact” metric is something totally customizable, and must be calculated as the experts in the business dictates, because each company behaves in a different way.

Let’s start with the report. In it, the different elements will appear numbered to facilitate their reading and monitoring.

Global metrics – tab

1. The percentage of impact of each social network. We can see how Instagram and Facebook predominate, the first being the winner.

2. Percentage of impact by subtype. In this case, we can see how eSports and entertainment stand out from the rest.

3. Regarding television channels, Cuatro and La Sexta are the superiors.

4. There is no doubt about the impact that Twitch is gaining. That is why we thought it would be interesting to compare it with the impact of both social networks and the TV. In this case, we can see how Twitch could practically be put at the same level than the other two, so the investment made in this platform is being truly effective.

5. Finally, in terms of general impact by source (without breaking down by subtype), Twitch would be the one performing most satisfactorily.

Social media – tab

In this tab we are going to enter into a study between the different social networks:

1. Number of total publications vs the target set. We set a goal of 1000 and we are 67 publications above.

2. Although the number of publications has been exceeded, we have not received as many “likes” as expected.

3. However, our custom metric that relates both “likes” and visits to publications tell us that we have reached a lot more people than expected, so the objectives are being achieved.

4. In this chart we can see the impact of our publications by date. February was clearly the busiest month.

5. One of the studies to be carried out is when (time) our publications have the greatest impact. Thanks to this visual we can see how from 8 to 9 and from 13 to 17 is when we have a greater impact.

6. Here we can see the impact of each of the social networks. In addition, it also works as a filter.

7. If we want to know which person in the publication has the greatest impact, this chart will tell it. Here Cristiano Ronaldo and Lionel Messi meet tied at the top of the list. However, Iker Casillas seems to be the one with that least impact creates.

8. We also have at our disposal a couple of filters to filter by “Service”, this means, by a social network account in particular or by a particular person.

9. Finally, we can see the number of publications by type (image vs video).

If we want to filter by a specific account (FC Barcelona, for example) this would be what we would see:

And here an example of FC Real Madrid:

As the analysis shows us that the posts with Cristiano Ronaldo or Lionel Messi have most impact, the decision-makers interest would be in having these two appearing in most of the posts in order to generate even more impact. And as the publications on the FC Barcelona account work very well between 8 and 9, the account should try to publish at that specific time, or between 1 and 3 o’clock, which is a good time to publish for both accounts.

Twitch – tab

As we have seen before, Twitch is one of the leading platforms at the moment. And to many, the new television.

1. Personalized impact vs the objective set.

2. Seconds that appear on the screen vs the target.

3. Number of people we have reached vs the target.

4. When the post has had most impoct. In social networks, it is crucial to know the best time slot to publish, so this chart is very important.

5. Impact per person. In addition, it also works as a filter. We can see how Ibai is the leader, followed by AuronPlay.

6. The impact by date.

“TV Channel” tab

This tab shows the analysis of different television channels and programs:

1. As in the previous tabs, here we can see the customized impact vs the objective.

2. Total seconds vs the goal.

3. People we have reached vs the objective.

4. Time period in which the most impact has been achieved, this being mainly between 6:00 p.m. and 3:00 a.m.

5. Impact by service. As we can see, the ads have greater impact than a football game. But, we have to keep in mind, that the costs for appearing in an add are different for the costs to appear in a stadium or on the players’ shirt, and that, in this case, we should also elaborate metrics such as impact vs. invested money.

6. The impact in different TV channels.

7. Impact per person who appears on the screen with the brand. This info can be used in future ads and choose one person over another to appear in the ads, in order to achieve a greater impact.

If you want to read more go to this post.

Stay up to date: idiwork.com

Follow us on Twitter

Links

• Azure
• Azure Cognitive Services
• PowerBI