The goals and how they were derived
AI provides enormous amounts of value in multiple industries. Because of its high value potential, many companies have been scrambling to implement AI within their organizations. And the projects, when implemented properly, have shown significant returns and improved competitive edge. Business that haven’t started implementing AI may lag behind competitors – Gartner states that organizations that have deployed artificial intelligence in 2019 grew from 4% to 14% and that 15% of all customer service interactions will be handled solely by AI in 2021, a 400% increase from 2017.
Why is there a need to use Artificial Intelligence (AI) within an Integration developer’s tool? Fundamentally, AI is a tool that reduces time for activities, removes drudgery and increases efficiency in systems as it learns the right behaviours to accomplish a task.
To add this capability, several different ideas and potential approaches were put forward however, like any other application, as we progressed it was critical to ensure that the overall perspective was not one dimensional on its development. Hence, we started with IBM Design Thinking and the 3-in-a-box model to bring the unique views of our diverse IBM team together to deliver capability that truly delivers an exemplary user experience and value to the businesses we serve.
Within the realm of integration, a lot of time and effort goes into building the mapping between endpoints. Each endpoint has its own object definition, schema, data types etc. Add to that the complexity of data flow to be captured while connecting multiple endpoints makes it even more complicated. An experienced integration developer who has built many flows obviously can do it faster and more accurately than someone who is a beginner or having average skills. The intent of this mission is to enhance the competitive advantage that IBM App Connect has by making it faster and simpler to build integration flows.
The next step was to derive the right metric as the goal in terms of what benefit would this provide. Several User Research projects were undertaken with customers and practitioners using face to face as well as online methods across multiple geos and based on information collected, it was concluded that doubling the integration efficiency would be the right goal to focus in the very first drop. This would mean that if 50% of mapping is done using AI then the goal would be met.
How does one run an AI project? The skillsets required to successfully complete the AI project were varied and completely different to a typical software project. While there was a need for the usual developer skills and the specialised skill of data scientists, it was obvious very early to everyone that because the intent of the project was to make the user experience simpler and more pleasant, the team needed to have greater representation from UX specialists to shape and guide the implementation. Additionally, App Connect has a history of award-winning Design, it was an imperative for us that we continue to deliver a first-class experience with AI-powered mapping.
Design Thinking practices helped the team settle into a rhythm and really learn from each other during the journey. This was achieved through regular weekly interlocks with all sides pitching ideas, experimentations, achievements as well as failures and frustrations that are very much part and parcel of working on something new. A lot of brilliant ideas were uncovered (some even being filed for patenting) and the teams began to gel extremely well. The bond remained strong despite huge disruption from Covid-19 forcing teams to work remotely.
A plan was created with an assumed squad velocity and burndowns projected for completion with the understanding that since this is a completely new squad, it would take some time to truly get the accurate velocity. Almost like Neo in the Matrix, the squad started realising their strengths and capabilities as well as the complete project challenge and bit by bit were able to negotiate the requirements giving a very clear outlook on projected goals as they moved forward.
The 3-in-a-box model of working ensured the customer goals remained at the heart of all decision making – when challenges arose making the delivery risky, the team quickly and effectively overcame the risks and ensured that there was no erosion to the value to our customers. In the end, this strong focus on user goals turned out to be one of the critical things that helped achieve success.
While competitors also seem to provide some level of auto matching, IBM has bought 20+ years of experience with both cognitive technologies and integration solutions together and combined that with our award winning Design teams to deliver a more intelligent solution. Our teams have gone beyond the basic benefits that AI can provide for mapping and thought holistically about the life of an integration developer.
Since we were going to auto match endpoints, there was a need to experiment with and finalize a Natural Language Processing algorithm as well as the explainability of the algorithm for the end user, which, among other things, also needs to provide and helps users understand the level of matching confidence.
Any project requires rigorous testing in order to ensure its completeness. How does one test an AI project’s success when, in this case, there wasn’t any customer data to work with? IBM corporation has taken a firm stand of not using customer’s data to benefit another customer, so this required either to dig into what we already owned, or we add additional scope in the project plan to build the test data.
App Connect has many templates for pre-matched endpoint fields to help users get an integration project started quickly. Since this was built by IBMers, there was no risk in using this for putting the algorithm to test against integration built by humans to see how well it worked. The result completely met our expectations and the algorithm matched the mapping within templates perfectly as per the confidence level.
Coming back to a holistic customer experience, one additional challenge which needed to be overcome was to find a way to keep the mapping solution as lean as possible in terms of footprint size and resource consumption. AI components are by nature ‘fat’ and resource hungry. A customer deploying on premise, needs their hardware and deployment cost to be minimal even while desiring the efficiency of an AI based system. IBM is mindful of that and it was considered a key part of the overall user experience. Hence while building the complete system, different algorithms and libraries were compared for each of these factors coming up with the most optimized combination. To make it even more customer friendly, a decision was made to make the deployment of this feature optional so the customer can incrementally augment their hardware resource capacity based on how their requirements evolve.
The Mapping Assist feature has provided a much-needed solution to a key challenge for Integration developers of any skill. It has proven benefits not only for our users though… It has also brought AI thinking and AI capabilities into the mainstream of the IBM Integration portfolio and will serve as a blueprint for other planned deliveries in this space in the future, ensuring the market does not look beyond IBM as a partner in their Digital Transformation journey.