We are living in the midst of a global revolution right now. The business world of today is more interconnected than ever, not just in terms of technology, but also in the number of partners that businesses typically deal with.
External partnerships have become essential to building and operating a business nowadays for the simple reason that they often lead to increased productivity and create new opportunities for both parties.
The number of partner networks is also increasing. An IBM study titled “Evolution of the API economy” highlighted that 70 percent of businesses want to expand their external partnerships.
For example, in the image below you can find an overview of a digital business ecosystem in the travel industry. This is comprised of:
- Supply side: This includes suppliers such as hotels and flights
- Demand side: This includes services like airlines and travel agencies
- Partner network: This connects the supply and demand side and can include entities such as supply banks, exchanges, or marketplaces
- Third-party applications: This can include businesses such as payment providers, CRMs, and so on
Regardless of the size of the company or the industry, one of the commonalities that exist in all external partnerships is that you’re sharing data between each other. As you can imagine, sharing data with external partners in any capacity adds additional complexity and vulnerability to both organizations. In particular, the APIs that bridge these partnerships are incredibly valuable, but they are also highly prone to errors, downtime, and cybersecurity threats. As a result, when one of these providers fails to meet a defined SLA, end-users are affected, revenue is lost and data breaches occur.
As the use of APIs and the size of these partner networks continue to grow, detecting these glitches and threats within a reasonable timeframe becomes a non-trivial task for even the most advanced technical teams.
The traditional way of monitoring metrics related to partnerships – for example traffic, referrals, and revenue – has been to feed them all through an IT monitoring or Application Performance Monitoring (APM) system.
The issue with these traditional approaches, however, is that monitoring machines and monitoring business KPIs are fundamentally separate systems and need to be treated as such.
In particular, business metrics related to external partnerships can often be much more volatile than monitoring machines. Not only that, but KPIs can also be influenced by external forces, such as seasonal human behavior. This means that, unlike machines, business metrics simply cannot be measured and evaluated in absolute terms.
Another key difference between monitoring partnership KPIs and monitoring systems is measuring topology or the relationships between these metrics. When we’re monitoring data from machines, we can often identify clear relationships between them. When it comes to monitoring business KPIs, on the other hand, there are often millions or billions of metrics to keep track of. This sheer volume of data means that many of the relationships between KPIs will be nonlinear, making them much more difficult to identify.
Luckily, this is exactly the problem that machine learning helps us solve.
Before we discuss how machine learning can be applied to partnership monitoring, let’s first review a few common types of partner networks that many businesses are already a part of.
Types of Partner Networks
Let’s discuss the two main categories of partner networks – affiliate networks and programmatic advertising – and the API technology they rely on.
An affiliate network acts as an intermediary between publishers (i.e. affiliates) and merchant affiliate programs. As you can imagine, managing large affiliate networks that have thousands of accounts is an exceptionally complex task. One of the challenges faced by these large networks is automating their tracking processes in order to prevent revenue loss and churn.
Affiliate tracking systems require account managers to track countless metrics such as traffic volume, conversion rates, return on ad spend, and many others. In order to manage these factors at scale, affiliate networks are now turning to autonomous anomaly detection solutions that provide real-time alerts to incidents in the network. For example, if the solution identifies drastic changes in metrics such as a drop in referrals, this could mean the account is at risk.
The result of implementing a machine learning system for affiliate monitoring means that instead of constantly monitoring changes in these metrics themselves, account managers are able to focus on more high-value tasks such as relationship management.
Another common partner network that many businesses rely on today is a programmatic advertising platform. Programmatic advertising networks connect advertisers with publishers, in which advertisers are bidding on inventory (i.e. ad space) in real-time, otherwise known as Real-Time Bidding (RTB).
These partner networks connect businesses that manage huge advertising budgets in a completely automated fashion, which means the platform also needs to monitor countless metrics including partner impressions, clicks and conversions. The number of metrics that these ad tech platforms are tracking each day can often reach into the billions, and are required for both the supply and demand side.
As you can imagine, one improperly tracked metric or error in the network can lead to millions of dollars in losses within seconds.
In order to solve the challenges faced by programmatic advertising networks, AI-based monitoring systems have become essential.
Machine learning is particularly well suited for this task as it’s able to handle a huge amount of data from the network, extract meaning from the data and identify potential incidents in real-time.
One of the reasons partner networks are so hard to monitor is because of the underlying infrastructure. They typically connect through APIs, or Application Programming Interfaces, which act as a software intermediary so that two applications can communicate with one another. One of the main challenges of monitoring APIs is not only that there are huge amounts of data being transferred each second, but also the fact that they have such low visibility, which means we may not always know when something has broken within the protocol.
An API error can often result in downtime for many other applications that are dependent on it, which, as you can imagine, creates a snowball effect of potential issues. Specifically, gaps in API performance can impact user experience, disrupt workflows, and severely damage a brand’s reputation. As a result, monitoring APIs at all times is essential to avoiding significant revenue loss.
AI-based monitoring can be used for APIs by learning the normal behavior of every single metric and then providing real-time alerts about anomalous events and potential incidents. One of the advantages of a machine learning-based system is that you don’t need to define exactly what to look for when monitoring the API. Instead, the system learns how to monitor the normal functionality, performance, correctness, and speed of every API call on its own.
The Traditional Approach to Monitoring Partner Networks
Monitoring partner networks have traditionally been done with the use of Business Intelligence (BI) dashboards alongside a manual alert system if irregularities are detected. The first challenge with this approach is the obvious inability to scale, as a team of analysts can only monitor a certain number of metrics manually, which is certainly well below the billions of data points created each day within partner networks.
Another issue with this approach is latency or the time delay between an incident and the team’s response. This is referred to as the mean time to resolution (MTTR), and as you can imagine, an increase of even an hour in your team’s time to resolution can have a huge impact on the cost to the business.
Finally, partner networks are at their core a relationship business. So if you’re not as proactive as possible in finding and solving issues, it’s quite easy for these relationships to be irrevocably damaged. Of course, issues in partner networks will always keep happening, but if your partners know that you’re using an AI-based monitoring system as opposed to a team of human analysts simply “eyeballing” incident logs and trends, that can go a long way in building trust.
Automated Partner Monitoring with AI
Now that we’ve discussed a few of the issues associated with traditional approaches to partner monitoring, let’s review how a machine learning-based approach can automate and improve the entire process.
The first advantage of AI for partner monitoring is speed. We mentioned earlier that a company’s mean time to resolution is a key metric in keeping the costs down when incidents inevitably do arise. The fact is that machine learning algorithms can monitor billions of data points each second, which is something a team of analysts simply will never be able to do on their own.
In the example below, anomalies were detected by automated business monitoring platform Anodot, which helped the team resolve them within the hour.
On top of monitoring data points, a machine learning-based system can automatically learn the normal behavior of each metric individually, so it can detect even the slightest deviation from the expected behavior. With real-time alerting built into the system, this means that your team is aware of and can respond to incidents much faster than traditional monitoring systems.
Another advantage of AI-based monitoring is scale. Regardless of how big of a team you build to monitor systems, there is likely going to be diminishing returns on human capital when it comes to the scale of data that partner networks have to deal with. Automated AI monitoring also doesn’t take breaks, so no matter what time of day it is, it can be used to extract meaning from huge datasets so your technical team has a comprehensive overview of what’s going on in the network.
A final advantage of an AI-based approach is the granularity it provides. Not only does AI-based monitoring provide a real-time overview of the entire network, but it also inspects the most granular metrics of the network.
As mentioned, even with a team of analysts watching over the network, these metrics can easily get lost amongst the sea of data. In particular, if there’s an incident in the network, an AI-based solution can group related anomalies and events into a single alert, so that you don’t get “alert storms” for every little incident that occurs. This ability to find correlations also means that you can identify the source of the incident, all the way down to the location, device, and browser.
As we’ve discussed, an AI-based partner monitoring solution has the ability to track every data point in a network at the most granular level, while also providing an overview of the network as a whole. This means that you can be as proactive as possible when it comes to optimizing your systems, finding incidents as soon as possible, and identifying the source in order to improve your time to resolution.
As a result, AI for partner monitoring allows you to build more trust and sustainable relationships within the network.
About the Author
Amit Levi is VP of Product and Marketing at Anodot. He is passionate about turning data into insights. Over the past 15 years, he’s been proud to accompany the development of the analytics market. Having held managerial positions in several leading startups, Amit brings vast experience in planning, developing, and shipping large scale data and analytics products to top mobile and web companies. An expert in product and data, his mantra is “Good judgment comes from experience and experience comes from bad judgment.”You can also read this article on our Mobile APP