New AI-centric Integration Use Cases Will Boost iPaaS Adoption

In the iPaaS Emerges as the Dominant Platform for Automation and Integration article published on nasdaq.com in January 2024, we highlighted how,  over the next few years, organizations worldwide will dramatically grow their expenditure in automation and integration technology to address new use cases and requirements. We also observed how integration platform as a service (iPaaS) will emerge as the dominant form of automation and integration technology, by tripling in size and dwarfing any other segment of the market in terms of both user expenditure and adoption. We mentioned iPaaS versatility - that is, its ability to support a wide range of cloud-centric use cases - as one of the reasons for its success. Moreover, IT megatrends such as cloud, mobile, IoT, API, analytics and AI are all generating high amounts of new automation and integration requirements, which  an iPaaS can effectively address thanks to its rich set of functionality and the high productivity and short learning-curve of its development toolset.

Of course the megatrend that is taking the market by storm is AI, and Gen AI in specific. This is great news for the iPaaS market because widespread AI adoption is favoring  the emergence of a number of new, AI-centric automation and integration scenarios, which CIOs and IT leaders will have to tackle as they venture in the AI world. Chances are that specialized forms of automation and integration technology will emerge to address these scenarios. Moreover, various established platform types - for example, low-code application platforms, business process automation tools and robotic process automation - will evolve to support AI-centric automation and integration. However, we expect iPaaS will confirm as the platform of choice because of its versatility, ease-of-use, widespread adoption, support for IT and non-IT users, and popularity among both midsize and large organizations.

What are these emerging AI-centric scenarios and what is the iPaaS role in enabling organizations to address them effectively? The matter is in flux as AI platforms are still rapidly evolving, both functionally and architecturally. However, we have so far identified four new AI-centric scenarios that organizations should be able to support now or in the not-too-distant future.

Three of these scenarios can be supported by the most advanced iPaaS already now. We expect that the leading platforms will enable the fourth one - AI-driven dynamic process composition - over the next 12 to 18 months.
 

1 - In-process AI functions.

This is the scenario where organizations improve their business processes by injecting AI capabilities to perform tasks such as OCR; text translation; drafting documents and emails; text categorization, analysis and summarization; chatbots activation; decision making support via ML-based advanced analytics; and more (see figure 1).

Figure 1 - In-Process AI Functions

This is a “low hanging fruit” scenario, because, technically speaking, it is relatively simple to implement via an iPaaS. It’s just a matter of engaging with LLMs, chatbots, NLP systems and advanced analytics tools leveraging the API and event channels they expose. The main challenges are:

• Navigating across the myriad of tools and APIs to identify those relevant to the particular business process
• Designing the process so that “humans are the loop”, to validate, govern and control AI outputs and decisions, when appropriate or needed.

2 - Retrieval Augmented Generation (RAG).

Gen AI Chatbots (for example, ChatGPT, Google Gemini, Bing Chat Enterprise, YouChat, Tidio, Jasper and Freshchat) provide answers to natural language prompts (queries) on the basis of an underlying LLM. The LLM was typically trained on static data, therefore the answer cannot include fresh, up-to-date information such as, for example, accounting data stored in the organization ERP system. Many AI vendors (for example, Microsoft, IBM, OpenAI and Meta) have implemented an architecture called RAG (retrieval augmented generation), which is enables a Gan AI chatbot to generate an answer based on both the query to the LLM and data retrieved from external sources (see figure 2).


Figure 2 - Retrieval Augmented Generation (RAG)

The power of RAG is to make it possible for business users to leverage Gen AI chatbots to access enterprise data via queries expressed in natural language. For example, OpenAI recently announced GPTs, a technology that enables the implementation of customized versions of ChatGPT via basic RAG capabilities.

Through RAG, business users, even those with minimal or no IT skills, can retrieve the data they need, when they need it, according to the democratized IT delivery model. 

However, although specific frameworks are emerging (for example, LangChain, LlamaIndex, and Haystack), RAG is more a conceptual architecture than a well defined set of standard technology building blocks. Therefore aggregating a RAG stack is still a complex exercise of making multiple components (vector databases, LLMs, knowledge bases and other tools) work together, which is something an iPaaS can easily do. Nonetheless, putting together a RAG stack requires skills and expertise that only the most technically savvy organizations have at their disposal, but iPaaS templates meant to help mainstream organizations implement RAG are emerging (see the Knowledge Workbot Accelerator for an example on how these components can be orchestrated via an iPaaS).  Moreover, getting access to the external sources may not be that easy as they are trapped into legacy systems, hard to integrate with and wrapped in layers and layers of technical debt.

3 - Composite AI.

This scenario refers to the orchestration of multiple AI techniques and tools to address a particular business issue. For most people AI is just a synonym for generative AI, however, this is only the tip of the iceberg. The somewhat vague (and to some extent unfortunate) term “artificial intelligence” groups together a variety of techniques and technologies including machine learning (ML), natural language processing (NLP), rule-based systems, OCR, optimization systems, knowledge graphs, autonomous agents and other disciplines.  

Therefore, AI developers might need to use multiple of these techniques, including in some cases multiple LLMs, to address a particular business problem. For example, a ML model to classify an incoming customer email, a business rule system to decide which business process must be triggered to deal with that particular email (e.g., order management, customer support ticket opening or management escalation) and Gen AI to create a customized response to the incoming email.  The tools and cloud services implementing these different AI capabilities are, in general, highly specialized, may come from different providers and are usually not integrated with each other. In this scenario, an iPaaS can provide the orchestration features needed to “make different AI capabilities work together” by leveraging the APIs and event channels they typically expose (see figure 3). 


Figure 3 - Composite AI

In this scenario too, typically humans must be part of the process to validate, govern and control the composite AI outcomes.

4 - AI-driven Dynamic Process Composition.

This scenario showcases the potential power of AI when it comes to automating business processes. It actually potentially changes their nature from a static set of predefined actions, tasks and decision rules to a much more dynamic, business goal-oriented and AI-driven way of responding to business events.

In AI-driven dynamic process composition, AI technology identifies, classifies, and extracts data from incoming “business events” (for example, a customer email or a message from an application system or from an  IoT device) represented by various forms of unstructured data. Then it “devises” what the end-to-end process to deal with each particular event looks like, and oversees its execution. AI dynamically plans the composition of predefined actions (we can call them “building blocks”) needed to achieve the desired outcome. Based on its understanding of the incoming event and of the overall business goal, AI chooses the right building blocks to run, specify their input parameters and orchestrates their execution flow.

Domain-specific knowledge and feedback from each building block execution make it also possible for AI to adapt its behavior in real-time to improve the business process effectiveness. Different incoming events may determine the execution of different actions orchestrated according to different processes (see figure 4).

Figure 4 - AI-enabled Dynamic Process Composition

In this scenario, a “classic” iPaaS can be used to implement the predefined building blocks as well as to include human supervision and approvals in the process. AI technology is responsible for the selection and orchestration of those building blocks needed to react to a particular business event.

This is the most complex, but also forward looking opportunity for CIOs and IT leaders to revolutionize the way they address the problem of “making independently designed systems work together” to achieve a business goal. iPaaS and other platforms are emerging to simplify the implementation of this particular scenario. Moreover, we expect that the emergent, yet quite immature, AI agents technology (also known as “intelligent agents” or “autonomous agents”) will play a pivotal role in this context.

5 - How Will CIOs and IT Leaders Invest to Tackle The Challenge?

CIOs and IT Leaders worldwide are investing considerable amounts of time, efforts and money to take advantage of AI and Gen AI in specific. However, the risk for them is to repeat the same mistake many of them made when they embarked on ERP, CRM, mobile, APIs and digital transformation initiatives: addressing the associated automation and integration challenges in an opportunistic, tactical way. The lack of an automation and integration strategy led many organizations to high costs, long time to value, duplication of technologies, skills and efforts, and mounting technical debt.

Many CIOs and IT Leaders recovered from the mistake by tackling the automation and integration challenge holistically and strategically. This implied:

• Selecting an appropriate set of technologies, more often than not centered on an iPaaS foundation,
• Setting up a dedicated team of solution architects and engineers
• Establishing a well defined operating model, whether centralized, democratized or hybrid.

However, AI experimentation is often carried out in business teams, which are not under the supervision of the central IT team. They may or may not be aware of the enterprise strategy and decide to leverage ad hoc tools, open source frameworks, custom development and all kinds of other integration technologies, which do not comply with the strategy thus leading to the high costs, duplication of efforts and technical debt issue mentioned above.

In part this is inevitable and in the nature of a new technology adoption process. However, CIOs and IT Leaders can avoid, or at least mitigate, some of these risks by proactively educating the business teams and, if needed, investing to extend their automation and integration strategy with the capabilities needed to support AI-centric scenarios.

In specific they will have to put efforts and money in:

• Brainstorming with business leaders about how infusing AI in their business processes can help the organization build competitive differentiation, improve business agility and increase productivity.
• Assessing which of the four AI-centric scenarios they will likely have to support in the short (6-18 months), medium (18-36 months) and long (36+ months) term
• Identifying a small number of “low hanging fruit” opportunities, which they can leverage to both help their teams climb the learning curve and validate the potential business benefits of AI-infused processes.
• Implementing these low hanging fruits by investing in an AI-capable iPaaS, thus giving them the opportunity to qualify and quantify its benefits in terms of builder productivity and operating costs reduction.  
• Developing a plan to leverage AI-capable iPaaS as the cornerstone of their automation and integration strategy, extending its use to a larger and larger number of scenarios.

CIOs and IT Leaders should consider undertaking these investments with a sense of urgency. AI initiatives are spreading like wildfire in large and not-so-large organizations. If they procrastinate for too long they may end-up in a hard to recover situation of different teams using different technologies to sort out their AI-centric automation and integration requirements, which inevitably, at some point, central IT will have to take on and deal with.

As they say “prevention is better than cure”. CIOs and IT Leaders able to “prevent” by proactively addressing the AI-centric automation and integration, for example, by investing in an AI-capable iPaaS. This will help their organizations take advantage of the AI revolution by minimizing the associated costs, reducing time-to-value for AI initiatives, and quickly building differentiation by combining AI and traditional systems to deliver new products and services to their customers.