Customer Story. Real-time bidding to place online ads on websites like the one you are looking at now is fuelling enormous growth in stored bid records for analysis by ad spot sellers and buyers. With billions of potential page viewers, cut myriad ways by gender, age, income, location, occupation, interests, etc. and thousands, if not tens of thousands of online outlets, then analysing bid data to find the best way to price ad spots and bid for them is mind-bendingly complex.
Ad spot buyers want to maximise the effectiveness of ad campaign spending while ad spot providers want to maximise their income by optimally pricing their ad spots, such as banners, side bars, etc. according to time of day, week, month, year, etc.
Both sides need to analyse as much data as they can as quickly as they can to get the most accurate decision-making information – but they haven’t been able to look at the gold standard source material – the raw bidder log data – because there is simply too much of it.
This has meant either analysing an extracted subset of the source data (estimation), meaning it’s incomplete, or using a background batch-type job analysis (pre-aggregation), which is slow and incompatible with the kind of near-instant analytics needed by adtech professionals researching both the supply and demand sides of the adtech business. The third option is to develop a custom system with hardware, database software and access software components – which is expensive, time-consuming and complicated.
Ocient has built a massively scalable parallel-access relational database storage system, using NVMe SSDs, for such data, and it can be accessed using SQL. It’s sold a hyperscale campaign forecasting system to MediaMath, an adtech demand side platform (DSP – see graphic above) provider for online ad spot buyers. Its staff now have near-instantaneous (sub-5 second)access to SQL analytic runs on massive raw bid log datasets.
The stats are, well, prodigious sounds the right word, as MediaMath handles more than six million bid opportunities per second with 10–12 petabytes of new records per day for more than 3,500 advertisers. By avoiding having to write a custom system that would have otherwise taken multiple engineers 9–12 months, MediaMath with Ocient was able to save about 50 per cent in time and cost while bringing on and supporting about $100 million of new and existing business.
MediaMath’s Ocient system features”
- Low-latency analytics optimised for highly parallelised processing leveraging NVMe solid state hard drives with compute adjacent to storage;
- Continuous loading and transformation of raw, semi-structured bidder log data with streaming data available for query within seconds;
- Support for geospatial data types and complex data formats, including arrays, tuples and matrices, to handle deep arrays of audience data and bid opportunity contextual data sets;
- ANSI SQL queries with powerful windowing, aggregation and machine learning models that process thousands of unique campaign targeting criteria;
- Zero Copy Reliability leveraging parity coding to ensure data availability and reliability with 60–80 per cent less storage footprint than copy-based data storage and analytics;
- GDPR and CCPA compliance;
- Deployed as a fully managed service in OcientCloud for ultimate flexibility and low operational overhead.
The loading and transformation of data could reach Tbits/sec speed but that’s not guaranteed with MediaMath.
Anudit Vikram, chief product officer at MediaMath, said “Working with Ocient enabled us to get to market months faster. We left the customisation of our solution to Ocient, meaning we didn’t need ten or more of our own engineers dedicated to making it work. We can now grow our market share and pursue massive new business opportunities while reducing costs and eliminating the resource-intensive process of customising our own solution internally.”
Check out a 22-page Ocient technical backgrounder (Next Generation Campaign Forecasting) on real-time bid analysis to get a better picture of what is going on. But, if you are not an adtech professional, put aside a good hour for this. The terminology is complex and the database details difficult to understand if you are a lay person – like me. I’m more sadtech than adtech.