AI bot based Service process redesign

AI bot based Service process redesign

AI bot based redesign of service processes is a structured and repeatable process that opens up many possibilities through AI (artificial intelligence) and continuously trained content.

The industry discovers artificial intelligence and bots as a means of redesigning service processes. Clever action delivers impressive results.

This post describes the necessary steps to design virtual agents (chat bots, smart bots, interaction bots), implement them technically and introduce them in an organization.

The method

The following steps are carried out iteratively for each release:

Smart Bot Design Prozess

After the first release it is usually possible and also probable that steps can be skipped, parallelized or executed in a different order, if in the previous release appropriate knowledge or preliminary work has already been done in the development of the backlog, UX or UI design or even content topics.

Understanding service requirements

Analysis and understanding

At the beginning of each release cycle there is a phase in which a common understanding of the expectations and needs of stakeholders and future users is created. At the same time, the given boundary conditions for content, technologies and operation are determined.

User or customer journey analysis

Bots imitate the question and answer or logic behavior of humans, so that in many cases processes previously represented by humans are to be intelligently virtualized. Alternatively, new processes are established that have not yet been mapped interactively.

Bots imitate the question and answer or logic behavior of humans, so that in many cases processes previously represented by humans are to be intelligently virtualized. Alternatively, new processes are established that have not yet been mapped interactively.

Content analysis

For interactive bots it is crucial to understand the intended persona types with their potential questions and to be able to deliver adequate content as an answer. In general, content is available, but rarely in the required form or quality.

During the content analysis it is therefore determined to what extent existing content can be used and where content must be generated to meet expectations.

Backlog structure

The Back Log is used for the structured recording of all functional, non-functional, process-related and organizational measures identified during the analysis phase.

The backlog is prioritized cyclically and evaluated in terms of maturity. In subsequent phases, it serves as the basis for deciding on sprint or release content.

AI Bot interaction design

In the design phase, the prioritized findings of the analysis phase are transformed into concepts that form the basis for later implementation. The design process is iterative and approaches from the abstract blue print to implementable details.

UX-Design for AI Bots

In UX (User Experience) Design we design variants of interaction processes based on the preceding analyses and optimize them with representative representatives of the previously identified persona types.

The working method is based on processes, not on design, and is supported by various methods and tools that make it easy for users to imagine new processes visually.

UI-Design 

Once the UX design has reached a sufficient level of maturity, we or an appropriate agency create the visual design to increase the acceptance of the UX-optimized processes.

In this context, it is important that an optimal UX process with optimal UI design also represents a transformation of your existing processes from the user’s point of view. Change explicitly requires change management in the organization.

Content-Creation for the customer journey

In this context, it is important that an optimal UX process with optimal UI design also represents a transformation of your existing processes from the user’s point of view. Change explicitly requires change management in the organization.

This means both the creation of question and answer learning patterns for the Microsoft Q&A Maker and the optimization or creation of content as a result of understood questions.

Example:  

Today, content for a question is available in a system or in a form that is only accessible to limited user groups due to technical limitations or content complexity.

A bot should overcome borders and deliver content in a simple, understandable and personalized way. This requires work that is started in this phase.

Operations concept

In the operations concept, the later DevOps and operations procedure as well as the technical stages are conceptualized and described.

Implementation of an AI Bot

The implementation needs the least description. This is where the technical realization of the bot, the design and the bot operation is carried out. For this we use Microsoft Power virtual Agent.

In parallel, the previously created or provided content is refined to acceptance in several test and improvement loops.

AI Bot Sprint Loops 

From the backlog, planned implementation content is bundled into so-called sprints, which are then processed and approved according to a fixed procedure. Content can be functional or non-functional.

Depending on the complexity, it may be necessary to run through several sprints in order to achieve an acceptable delivery result.

Bot-Training > Content-Acceptance-Loop 

In these loops, questions and answers are tested with bot logic and improved until potential users classify the results as useful or satisfactory

Bot-Entwicklung -> UX/UI-Loop 

In these loops the design drafts are technically implemented and improved. Here we can either use standard frameworks like Power virtual agent or custom development.

This process is similar to the trade fair prototype process in industry: In the design phase, unrealizable extremes are presented, which are adapted to the feasible realities in line production.

This phase can be critical if the design drafts lose value from the user’s point of view rather than gain in value through fine tuning.

Content-Performance Monitoring / App-Usage-Monitoring 

In order to be able to analyze the performance of the bot in later operation, mechanisms must be defined and implemented that allow for continuous improvement in operation.

Bot performance is evaluated based on the ability to answer questions from users in a qualified and satisfactory manner. This requires not only the ability to measure the questions and answers, but also requires business logic and technical metrics to support the analysis semantically.

The performance measurement process is basically similar to the patterns known from search engine optimization – only that the complexity is higher for bots.

Operations : Infrastructure as Code 

When setting up the development environment, it makes sense to script the development environment so that the cloud stages and DevOps processes required later can also be generated by code when they are needed.

Pilot tests of the AI Bots

A pilot then serves to test the interim results with selected persona types in a pilot environment in order to approach a minimum viable product (MVP), which can also be productively rolled out in a limited target group.

Rollout of the Bot supported service processes

AI Bot MVP 

Once a level of maturity has been reached that meets the minimum requirements of a target group, this stand is made available to real users in order to learn in practice which functions are useful from the user’s point of view, which are identified as additionally required and which are identified as dispensable.

At this stage at the latest, it makes sense to introduce instruments such as user voice to facilitate the prioritization of feature requests.

AI Bot Operations  

At the latest when the bot is made available as an MVP, the operation of the bot must be transferred from development to a DevOps-capable staging environment and operations must begin. In the best case, Infrastructure and DevOps As Code can be used here.

Content-Performance Monitoring  

When the system goes live, the Content Performance Monitoring prepared during implementation is also activated.

At this point in time, the ongoing analysis and improvement loops with the specialist department based on the monitoring should also be established.

AI Bot -Usage-Monitoring 

In addition to measuring the success of content, it also measures how actively the services are used by which user groups in order to determine the rollout status and potential marketing, communication or training needs.

Measurement of success

Each bot release benefits from marketing and feedback from bot users to your satisfaction with the customer journey, user experience, UI design and usefulness of the bot.

Success measurement as an integral part of the release process leads to bots as a tool being perceived faster than “normality” – but also becoming visible to those who only sporadically come into contact with bot-based services.

Release planning

Backlog-Review 

At the end and beginning of a release, the backlog has to be restructured in order to be able to focus on important topics with measurable demand and value creation with the help of the collected knowledge.

It is common that a backlog continuously grows faster than the ability to handle the requirements. And it is a healthy process to implement only fractions of the ideas and requirements of a backlog.

User-Voice Review 

If user voice was used, the feedback from users is used for better prioritization and to supplement or reduce the backlog.

Release decision

After the backlog review, it is usually useful to assess whether further releases are useful and which core topics they should contain.