In my first blog post, I introduced the ‘Coblox framework’ – a set of 33 cognitive building blocks to simplify leveraging AI-technologies for process and decision automation. We explored how applying AI and cognitive methodology to the business domain can unlock outstanding automation, sustainable innovation and collaboration benefits for the organization.
In this second installment, we will learn how to apply this AI-methodology to a real-life business process– complaint management.
Complaint management is the process of how organizations handle, manage, respond to and report customer complaints. Systems are put in place to track the data that is captured by complaint management processes. Businesses use complaint management systems to analyze where improvements should be made and use this information to satisfy customers and protect the company from repeated complaints. A basic complaint management process normally includes a timely acknowledgment, a process towards resolution, and a final outcome that benefits everyone.
Challenges & Solutions
Typical challenges in this very labor-intensive process are:
- It has to handle structured and unstructured data (text, images)
- It usually involves the knowledge of several specialized organizational parties
- It has a significant impact on customer experience and loyalty
- It requires root-cause analysis and decision making under partial uncertainty due to its retrospective nature
- The results have to be updated in several back-end systems
These challenges can all be addressed by an AI-centric approach as AI can:
- Integrate structured and unstructured data
- Represent the knowledge of several subject matter experts in a single knowledge base
- Improve customer experience with chatbots and service automation
- Reach conclusions under uncertainty
- Automatically update several back-end applications consistently
How to model mail-based complaints with Coblox
This process variant requires assembling five cognitive building blocks:
Click on the image to view in full-screen mode.
Each block represents a specific cognitive capability, in which it applies human problem-solving skills:
- Information Extraction fills all required slots of a complaint template from an e-mail: customer name, complained product, complaint reason, invoice number etc.
- Classification assigns various categories to the case to automate its subsequent processing: New Case yes/no, CR-agent required yes/no etc.
- Validation uses these categories and looks for a similar case within its knowledge base (case base). The most similar case is adapted to the new situation and automatically prioritized. From the past solution important attributes are re-used (e.g., investigation yes/no, sample yes/no).
- Routine Action Automation 1 creates or updates a new case automatically with all the information previously determined and triggers an investigation or requests a sample (if indicated from the past case).
- Routine Action Automation 2 finally selects the settlement action as defined by the complaint policy, closes the new case and informs the customer automatically (using his preferred channel).
- All Cognitive Blocks automatically involve a human agent when it determines it can no longer handle the case autonomously. The various exceptions are described in the graphic (e.g., when the template cannot be filled from the mail or when no similar case exists).
Complaint Process Variant
Another complaint process variant is the conversational complaint process, which uses very similar building blocks plus a block to conduct frame-based (controlled) conversations:
Click on the image to view in full-screen mode.
The first chatbot conversation collects just enough data from a user to be able to classify the complaint as described in ‘mail-based complaint’. The second chatbot conversation collects all remaining complaint-type-specific data which are required to search for similar complaints in the case base as effectively as possible. The remaining process is exactly the same as for mail-based complaints. Because of this two-step process, the conversational variant is likely to find more similar cases than the mail variant, which will tend to involve an agent more frequently.
Further complaint variants like ‘web-based complaint’ or ‘rating-based complaint’ could be easily modeled with a combination of common cognitive blocks plus one or two channel-specific blocks.
In my next and final article, I will apply the Coblox framework to the beauty care industry, where artificial intelligence is already making a huge impact.
Thanks for reading, and I look forward to your thoughts!
Click here for part 3: AI in beauty care

Harald Gunia
Associate Partner, Infosys Consulting
Dr. Harald Gunia is an enterprise architecture and artificial intelligence expert. He has more than 29 years of experience in all major AI technologies, including, machine learning (ML) and robotic process automation (RPA). He was worked in more than 10 industries and has deep expertise in digital capabilities and large-scale business transformations. Harald holds an M.Sc in computer science and a PhD in artificial intelligence.