As mobile apps take a more central role in the relationships between companies and their customers, our expectations for them grow, while regulators and privacy watchdogs apply a higher level of scrutiny.
Edge AI is the perfect solution for digital marketers who want to provide a personalized customer experience without infringing on users’ privacy and without getting into trouble with the regulators and the App stores (Apple / Google) over unsafe data practices. With Edge AI, all personal data can stay on the device.
Less pain, more potential
At the same time, Edge AI can save a lot of headaches for app developers. App Store submissions are much easier, with regards to disclosures and data safety practices, since you probably won’t have to collect or share any data. Here are some key additional benefits for app developers:
With Edge AI your app will be able to:
- Process and analyze large amounts of data locally, combining app usage and real world behavior without needing complex server-side integrations.
- Benefit from insights that are continuously updated throughout the day, every day, so data doesn’t “go stale” and even “dormant users” can be targeted.
- Trigger engagement in real time, instead of relying on network connectivity and round-trips to the server.
- Use external data from 3rd party services to trigger messages and engage users
- Build predictive models about each individual user, so your app can anticipate the next move.
- Implement a full solution with a single point of integration.
- Avoid hidden costs and scalability limits, since the solution will work the same for any amount of users.
Bright Future, Dark Magic
Edge AI is an excellent solution for many of the challenges that app developers face. However, that elegance masks a great deal of complexity. After all, being able to take cutting edge Artificial Intelligence and server-sized capacity and making it run efficiently on a variety of handset environments requires engineering expertise and skill that is just short of magic. The sections below outline a few of the challenges to optimizing Edge AI to the point that it can run securely, reliably and efficiently on consumer-owned devices and under the constraints of mobile OSs.
Resources: Doing a Lot with a Little
Most AI projects never have to consider the amount of available resources. It is typically a matter of budget; if you have enough budget, you’ll have enough resources. For Edge AI, the opposite is true. There are different types of limitations:
Global Limitations– The maximum capacity of the device. As advanced as mobile devices have become, they are clearly limited in comparison with the server-side environments: there are CPU, memory and power (battery) limits.
Runtime and Practical Limitations– The maximum capacity allocated to any one app at any given time.
With Edge AI, as opposed to server-side processing, the app operates as a ‘guest’ on the device, and must be careful not to abuse the hospitality. This means that a lot of engineering energy goes toward which resources are utilized and, perhaps more importantly, when they are released.
Craftsmanship: Precision and Skill
Edge AI has a great deal of promise. The amount of fresh, high-quality data and the ability to process locally open up incredible opportunities. The challenges are equally high, since the tools that are available are much more limited. One developer compared it to “painting a masterpiece using only two colors”. For example:
Limited coding options
With server-side development there are a variety of technological choices to choose from, but when it comes to developing on the edge, those options are much more limited.
Interfaces and Integrations
Cloud architectures offer a greater number of options for interfacing with other solutions that are often, similarly, based in the cloud. However, in the constrained environment of a mobile phone, there is a larger emphasis on optimization and thinking through each integration.
Server-side development is deployed once, to a known platform. On the Edge, however, there are multiple permutations of device/OS and the rate of deployment is out of your control, so it is almost guaranteed that your software versions will not be homogeneously deployed and you will have to contend with multiple versions across your service.
Server-side solutions are notorious for capturing volumes of data that may or may not be relevant. With limited memory, the challenge is to continuously discard data that is not useful in order to focus efforts and resources on the “good data”.
Administration of the solution
Creating a web interface to manage your server-side data is fairly straightforward, since all data can be read and new data can be written. When developing on the Edge, you must create a unidirectional architecture in which the device can make the decision regarding the information it should ‘pull’ in order to provide a relevant user experience.
Availability of engineering resources
For all the reasons above, developing a solution centered around Edge AI requires engineers that are experienced enough with real time processing and low level coding in smartphones. There simply aren’t a lot of them around.
Fine Tuning is an Ongoing Task
Anagog is at the cutting edge of Edge AI development, recognized over the past few years as a leader and a pioneer when it comes to applying Edge AI to mobile marketing and engagement. Even now, as our SDK has been deployed in tens of millions of devices across dozens of OS versions and phone models, the task of fine-tuning and optimizing our software continues, as we find creative ways to use the data available for generating deep insights into the behavior of the users of mobile apps that have embedded the Anagog SDK. For example, we have adopted the following practices:
A ruthless approach to storing data
Selecting the data that is really useful and valuable vs. reflexively collecting every bit of data only to discover later that it is useless or stale.
We are very respective of the device resources in order to prevent unnecessary battery drain and memory “bloat”.
Excellent detecting skills
Our experts have many years of experience in mobile environments and have developed a “sixth sense” about what information can be extracted from the device by making smart connections and applying machine learning.
Developing AI data models requires evaluating all types and sources of data that may be directly or indirectly related to different user behaviors. Looking for “secrets” among the data points requires continuous modelling and testing.
Edge AI has enormous possibilities that help an app provide an engaging, high-quality user experience without relying on a cloud connection that introduces delays and legal complications. As with every powerful capability, skill and expertise are key for an effective application of the technology.