Optimize ad spend with machine learning

Client Snapshot

An online travel and hospitality company was spending billions annually on online advertising but was concerned with the inefficiency of much of that spend. The company identified the need to optimize spending on metasearch platforms at the property level. Blueprint employed advanced data science and machine learning to build a system capable of predicting ROI with the desired level of granularity, reducing overall spend and increasing the profitability of the remaining spend.

Work Summary

The problem

Most people have used a metasearch engine, which aggregates results from multiple websites to display price comparisons for anything from fans to cell phones to hotel stays. Within these aggregation sites, brands often bid for placement in search results and pay per click whether their ad converts to a sale or not, meaning brands can wind up losing money on an ad, even if it does convert.

Online travel and hospitality companies are heavy users of metasearch engines because customers love to plan their trips and compare prices for everything from hotels to flights to cars to concerts. Metasearch represents a powerful online channel that drives traffic and sales, but performance advertising has razor-thin margins, so ensuring profitable ad spend is essential for success.

An online travel company with multiple distinct brands approached Blueprint to help optimize and predict the return on advertising spend for hotels appearing in metasearch results. With a performance advertising budget topping one billion dollars, any improvements to the process of spend allocation would reap significant returns. Luckily, the data landscape in the online travel industry is vibrant, with thousands of data points per property that can influence the success of a predictive model. Blueprint recognized that the solution would involve utilizing that existing data in new ways to improve performance.

The Blueprint Way

As with any data science project, the engagement began with an effort to gain a deep understanding of the customer and their situation, followed by heavy data engineering to prepare the data and build the pipeline to feed and train the machine learning data models. The company had multiple brands, each with massive amounts of data that needed to be extracted and housed in one location, so rather than wait for the in-house engineering team to transform and load that data, Blueprint took on that work to return value more quickly.

Machine learning applied between customer data sources and customer business systems

Blueprint built an end-to-end AI/ML data pipeline, a marketing campaign optimization engine and five machine learning models that fed into a performance optimization algorithm. The resulting system accurately identified the ROI of individual combinations of brand and property on metasearch, allowing the company to decrease wasteful spending and increase profitable spending.

Successful machine learning can take multiple models working together to accurately and reliably predict future performance. For this engagement, Data Scientists needed to factor in customer behavior, ad placement and spending, and to do that, Blueprint built models to predict:

  • Click-through rate
  • Whether a property will display in a search
  • Probability of conversion
  • Cost-per-click
  • Property profitability

“It is so simple, yet so complicated at the same time,” said a Blueprint Data Scientist who worked on the engagement. “All these things are interconnected, and we needed them to work together.”

In addition to identifying which hotel and brand combinations warranted bids for ranked ad placement, Blueprint built another algorithm to specify which combinations merited bids for the No. 1 ranked position. All the models worked as a synchronized whole, receiving and returning continuous information streams in the company’s data ecosystem.

Armed with new prediction capabilities, the travel and hospitality company nearly eliminated wasteful metasearch spending, optimizing their return on advertising spend. In the first three weeks of using these models, the company saw a 26% improvement in efficiency, resulting in considerable increases in company-wide revenue and profit.


Share with your network

You may also enjoy

Blueprint collaborated with our client’s engineering and data teams to acquire and organize this data for analytics, leveraging the Lakehouse Optimizer to ensure cost transparency and continuous improvement within their Databricks platform.
With a focus on migrating from Cloudera to Databricks within a tight timeframe, the Blueprint team executed a seamless transition while addressing critical challenges surrounding data silos, and heavy storage costs.