Company’s Profile and Objective
Our client is a US-Based Out of Home or OOH service provider. They help their customer analyze their brands’ reach to the Customers through Media ads and commercials. Using audience and advert data analysis, they are able to analyze the media value of their brands. This global OOH leader boosts the omnichannel approach of brands by connecting them to their audience.
While working with multiple brands who run commercials on TV channels, they couldn’t generate audience analysis due to delays in data arrival. The data that our client receives from TV channels about audience engagement is delayed by 14 days after the programs were aired. This data helps a brand decide its budget for launching and running commercials on TV channels on various days. By the time they receive data, the audience analysis that decides the ad spend is often stale or irrelevant.
When our Logesys team was onboarded, we had the challenge of predicting audience volume for the TV channels, which the brands would utilize for their TV commercials strategy. And our challenges weren’t fewer.
Our target was to analyze the last 1 year’s data and predict the audience forecast. This is not a small dataset to work with when the data feed is 4x an hour.
Time Sensitive Analysis
Planning a marketing strategy for a TV channel is a time-sensitive target. We needed to synthesize and analyze the latest data for predictions that might fluctuate during weekends and holidays. This needed an approach that could process data at high speed.
Time Series Inaccuracy
We tried to use multiple algorithms from the existing time series solutions to analyze the past audience data. But somehow, the accuracy was still a far-fetched dream.
The OOH company receives raw data from another vendor—an expert in data gathering. But this data, being raw and unprocessed, doesn’t help predict the audience engagement with a brand over the TV channels. We were tasked to devise a machine learning algorithm that can process this data fast without posing a performance threat and present an audience prediction.
Our Logesys tech team analyzed the incoming data and finalized the following tools and technologies for the proposed machine-learning algorithm:
Machine Learning Algorithm
The existing time series algorithms like ARIMA and LSTM couldn’t produce the expected results. So, we used the Random Forest Regression model to solve the issue. This model combines the outputs of multiple machine learning regression models and makes a better prediction altogether.
Since this isn’t a time series model, we converted it into a time series solution by extracting date and time information from one single time-stamped column. We have derived quarters, months, weeks, days, hours, minutes, etc., from the column to run the algorithm.
The 1-year data that acts as input for us is stored in SQL Server. The data comes in batches of 15 minutes for every channel separately, and we process each bunch for audience prediction. We flag the data with weekends, special holidays, and government holidays because these days drive more audience engagement.
We’ve used Power BI to develop the dashboards. We were able to visually present and analyze the Audience engagement, Brand’s reach, media value, etc.
Our client can run a Power BI report showing them the actual audience volume from the past and predicted audience engagement for a particular time slot. The report has generated quite accurate data so far. With the aid of the graphical data, the OOH client has been able to help their partner brands zero in on a strategy for investing in TV commercials.
For instance, if a brand knows beforehand that channel A will receive more audience than channel B, they can elevate the budget for the first one and cut the investment for the latter. Likewise, they can utilize the budget for holidays and weekends more when the footfalls are expected to be higher.
The solution offers multiple benefits to a brand sponsoring the TV channels or programs, such as:
Logesys team worked with the OOH client for 3 months and delivered the output to their satisfaction.
The project has been delivering the forecasted data to the brands with an accuracy of 90% compared to the actual data received. Since there is no lag of 14 days, they can effectively tweak their visual media marketing plans on time to accommodate unexpected fluctuations.
The solution has been running for more than a year now and is expected to help brands hit the right balance in finer ways with audience engagement. After all, a machine learning algorithm grows its accuracy with time.
The dashboard is compatible with all handheld devices and is highly accessible and swiftly sharable.
Our client’s marketing team has nothing but words of appreciation for the forecasting this project offers them.