Research Blog

October 14, 2019

Applying AI to Enterprise Integration: How Ready Are We?

Introduction

Lately, it seems that we hear about artificial intelligence (AI) every day and see it popping up practically everywhere. AI can do many things humans can - and do them thousands to millions of times faster. This enables us to take humans out of some workflows and instead rely on artificial intelligence to drive automation in many use cases.

Because AI has the potential to redefine multiple technology landscapes, it deserves careful attention. In this case, we have studied AI from an integration lens, exploring how and where it can change the integration landscape. In the context of our research, we have defined integration as, “the use of technologies that enable typically diverse systems within or between organizations to work together to achieve common business goals. This blog summarizes our findings.

AI-Related Integration Use Cases

First, as a part of our study, we identified eleven classes of AI-related use cases in the integration domain.

  1. Integrations required to enable AI - collecting data required for AI and carrying out actions suggested by AI, which creates many use cases
  2. Data integration - use of AI to reconcile data from many data sources with different formats and semantics into meaningful records
  3. Extracting useful information from human inputs - use of AI to extract meaningful data for inputs, such as audio, video, and natural language
  4. Increasing usability - use of AI to enhance the usability of systems
  5. AI support within existing data stores - supporting AI as turnkey solutions within existing data stores
  6. Algorithm economy - combining algorithms from an algorithm marketplace to build systems
  7. Self-driving operations (DevOps and regulatory compliance) - use of AI to reduce or eliminate the need for human intervention while managing systems
  8. Automatic and self-service integration - use of AI to simplify the process of creating integrations
  9. Security - use of AI to detect potential attacks and fraud
  10. Business automation - use of AI to automate business operations
  11. API marketplace - use of AI to provide recommendations and ratings in API marketplaces

Key Observations

In evaluating artificial intelligence in the context of the integration use cases we arrived at a number of observations. Notably, AI indirectly creates demand for integration. This is due to the fact that enabling AI in an enterprise involves collecting, cleaning up, and creating a single representation of data, as well as enforcing decisions and exposing data outside, each of which leads to many integration use cases.

Second, weaknesses in AI decisions can be much more harmful than a weakness in a human decision for three reasons:

  1. AI is being integrated into most aspects of our lives.
  2. Since the per-repeat cost is small, AI is used more frequently than humans.
  3. Different people have different opinions, which provide oversight and balance out the effect of mistakes, whereas AI applies the same bias broadly. For example, human interviewers will have a broad range of bias while AI will have one kind of bias.

Therefore, the teams implementing AI use cases - whether or not they involve integration - should carefully consider the ramifications of those challenges, such as bias and privacy concerns.

Third, challenges, such as a shortage of skilled professionals and data and model interpretability, are limiting the effectiveness of AI adoption. For example, when applied to competitive fields, such as security or stock markets, AI-based systems can end up competing with each other which might lead to worse-than-before situations. One example is stock market flash crashes.

Fourth, multiple forces are driving future AI applications that will be offered as cloud APIs.

  • Since AI needs data, having exclusive access to data in some cases leads to significant competitive advantages. This is called the “data network effect.” AI applications in the cloud give vendors full access to the data.
  • Due to a lack of expertise and data, custom AI model building will be limited to large organizations. It is hard for small and medium-size organizations to build and maintain custom models. Many AI use cases will be solved as cloud APIs that let vendors concentrate data and expertise.
  • AI deployment and management is complex, and they complicate the deployments while professionals who can deal with those complexities are in short supply. Using cloud APIs will let AI consumers sidestep those complexities.

In our study, we explored how each class of use case is affected by the opportunities, challenges, and risks identified above. Among the use cases, we have found that four already have systems available and are likely to find more adoption. These are “security,” “data integration,” “extracting useful information from humans,” and “AI support within existing data stores.”

On the other hand, we have identified three use cases that do not face critical challenges but will need 5 to 10 years of development before wide adoption can occur. These include, “increasing usability,” “self-driving operations (DevOps and regulatory compliance),” and “automatic and self-service integration.”

Finally, we have found that the “business automation” use case, while accessible for large corporations, will often be out of reach for small-to-medium enterprises (SMEs), if they are not supported as a cloud API.

Conclusion

In our study, while critically evaluating how AI can affect the integration domain, we identified eleven classes of use cases that can significantly affect integration outcomes. AI has a significant developer community and a rich set of tools. Many real-life AI use cases have demonstrated their effectiveness. Despite the promise, there are obstacles. Risks, such as bias and privacy concerns, along with challenges, such as model interpretability and a shortage of skilled professionals, are limiting the adoption of AI. Our report, How Will AI shape Future Integrations? discusses the opportunities, risks, and challenges in detail, evaluates the feasibility of each use case, and identifies four classes of use cases ready to be exploited.

Learn more about our research here.