Data-driven decisions made in real time

Daniel Landsman at Aerospike describes how AI is redefining real-time business decisions

Consider for a moment that it takes about 20 milliseconds for a thought to cross the synapses in our brains. Then think about how long it takes for an online ad to load - less than 200 milliseconds, or the blink of an eye. During that time, an auction takes place among advertisers to establish which ads are placed for us to see. These technology-aided business decisions are lightning fast, and things are set to get even faster.

AI has revolutionised decision-making processes and nowhere more so than in real-time applications. Driven by demand from businesses to improve efficiency and responsiveness, gain competitive advantage and boost customer experiences, AI-enabled real-time decision-making has become pivotal. 

Where it all began

Early versions of computers created accurate monthly accounting that was used to assist in data-driven business decisions. Later, the arrival of personal computers, and then the internet, allowed organisations to make important decisions not in monthly cycles but on a weekly basis. And now, because of access to information in real-time, decisions are being made daily. 

Meanwhile, human cognitive processes remain the same. As long as we are the ultimate decision-makers, our brains are a bottleneck, and this constrains the time and efficiency with which businesses can respond to data. 

Combining AI and real-time data

The answer lies in the integration of AI with real-time data platforms. This means that large amounts of information can be processed almost instantly, supporting businesses in making faster, more informed decisions and outstripping the limitations of our own brains. 

Central to this capability is the use of real-time data platforms. These ingest millions of data points per second, supporting high frequency reads and writes with consistently low latency. These platforms manage the demands of AI applications, which makes them ideal when immediate responses are required, such as for personalised advertising or for detecting fraud.

A good case in point is Quantcast, which uses AI to orchestrate real-time auctions for ad placements. With AI/ML (machine learning) techniques and real-time data from millions of online sources, Quantcast aims to deliver suitable ads to its users fast and at scale. This enhances the relevance and impact of ad campaigns and assists in creating higher conversion rates. 

Another example is the real-time AI-driven fraud detection system that is used by PayPal, widely considered to be a standard-bearer for financial transactions. In the first quarter of this year it processed 6.5 billion transactions, using its graph database model to identify and prevent fraudulent activities.

This works by using a two-sided network with buyers and merchants sending each other transactions. The network is encoded as a graph with buyers and sellers modelled as vertices in the graph. The real-time graph platform returns graph query results very quickly or in sub-seconds, and the return query results are used in machine learning models for immediate fraud prevention.

Real-time fraud detection is essential to the integrity and security of PayPal’s transaction network.

Gazing into the crystal ball of real-time AI

Looking further ahead, there are several emerging trends that are set to define the next generation of AI applications.

Enterprise data needs securing for privacy and competitive reasons, but it also needs to be included in AI apps for improved contextualisation. We will see the use of what is described as “retrieval augmented generation” (RAG)-based AI apps, which let enterprise data provide context to large and small language models, resulting in a better quality of AI-based decisions. 

The future of AI applications also lies in multi-model capabilities—key-value, graph, document, and vector—within one platform. This approach will generate richer AI applications that can manage complex data types but still deliver real-time insights with low latency. Integrating multiple data models will help organisations develop more sophisticated AI applications for deeper insights and more effective real-time decision-making. 

As AI models evolve, continuous learning will become important for maintaining accuracy.  Traditionally, models have been trained offline and updated from time to time. But if companies are to gain access to up-to-date information that informs context-rich decisions, continuous learning will be an important next step. 

Making critical decisions in real-time

Integrating AI and real-time data together is already changing the decision-making processes across multiple industries. By processing and acting on the insights that the data provides more quickly than a human can, AI enables businesses to make informed decisions almost instantly. This is giving businesses a clear competitive edge and helping them forge long-term loyalty with customers.

As technology advances further, adopting enterprise data, multi-model capabilities and continuous learning will progress the opportunities afforded by real-time AI, bringing in a new age of rapid, data-driven decision-making.

Daniel Landsman is Global Director of Adtech, Martech and Gaming at Aerospike

Main image courtesy of iStockPhoto.com and mikkelwilliam