Conversational AI offers significant advantages such as high availability and better service quality. The technology is a key element for the contact center of any company. But careful implementation is essential for it to produce its full effect.
Find out in detail which organizational, technical, and personnel requirements a company must meet to achieve a seamless integration of conversational AI and drive digital change with vigor.
Of course, we also discuss the risks and reveal whether introducing conversational AI is worth it at the end of the day.
What benefits does conversational AI bring to a company?
Companies that use conversational AI in their customer service benefit in a very tangible way – because conversational AI increases:
- service availability,
- service quality, and
- efficiency, while reducing costs.
Last but not least, it makes the jobs of contact center employees and customer advisors more attractive.
The most important steps to introducing enterprise conversational AI
Here is a summary of the most important steps and best practices for introducing conversational AI, as we described in detail in the previous blog post:
- Define a vision and align it with the corporate strategy.
- Identify use cases along the customer journey and estimate economic (e.g., higher efficiency) and qualitative added value (e.g., more convenient and faster problem-solving for customers). Create a roadmap.
- Involve stakeholders in the design process and develop needs-based solutions.
- Select an AI platform that is both scalable and can be used to implement individual use cases separately.
- Develop the necessary skills and form a multidisciplinary core team that systematically implements the use cases.
- Develop a clear learning cycle and continuously optimize the solution.
Requirements for conversational AI
Introducing individual conversational AI platforms for enterprises has become easier with the rapid technological development in recent years and the emergence of numerous SaaS solutions. For example, many providers and developers emphasize that a voicebot can be implemented within a few weeks or even days. That is entirely possible. However, the question is whether these solutions are scalable, sustainable, and easily integrated into a company's organizational and technical environment.
Our experience with conversational AI projects in the enterprise context shows that scalability and sustainability are only achieved with a systematic approach to meeting the technical and organizational requirements of conversational AI.
The prerequisites for planning, implementing, operating, and further developing coherent, complementary, and sustainable conversational AI services are illustrated in the following iceberg image and described in the following chapters.
1. Strategy and roadmap, people and processes
A clear vision, a robust strategy, and a reliable roadmap are essential at the start of implementing conversational AI. Considering the current challenges in customer service, i.e., overworked staff, increasing and distributed knowledge, rising costs, and declining customer satisfaction, it is crucial for a company to identify and implement the most important conversational AI use cases and prioritize them based on business cases.
On the one hand, the aim is to fully exploit the potential of automation with such self-service solutions as chatbots and voicebots, for example, for automatic responses or handling transactions. On the other, efficiency and quality of the customer dialog should be enhanced through the support of Co-Pilots that take over such tasks as searching for information and administrative activities. These measures must be orchestrated along a roadmap.
There furthermore needs to be a commitment to the professional development of employees. Companies need to invest in their continuous training and development to ensure their teams have the necessary skills. This applies to all levels of the iceberg. At the same time, creating an AI-friendly corporate culture is an important prerequisite for digital transformation.
This is the reality of today: AI will change traditional roles, create new ones, and set the stage for a dynamic career in contact centers. The use of AI does not mean that people will disappear from contact centers. Instead, it is taking up the challenge of meeting customer needs more efficiently. When companies combine the strengths of their employees with those of AI, they can transform their contact centers into a strategic hub that boosts customer satisfaction and grants a competitive edge.
2. Conversational User Experience (CUX)
How do callers experience discussing a personal matter with a voicebot? How do they perceive such interactions, and do they appreciate them? This so-called conversational user experience (CUX) is a key success factor in conversational AI projects.
As a specialist discipline, CUX design deals with how to make interaction with a digital assistant so intuitive and natural that it resembles a conversation between humans. This includes combining various forms of communication supported by AI, such as chat, voice, e-mails, and mail, in such a way that customer concerns are answered as conveniently and efficiently as possible. A pleasing customer experience pays off in several ways:
- It deepens customer loyalty to the company.
- It promotes customer willingness to buy and pay for products and services.
- It reduces support costs.
CUX design is developing rapidly due to a paradigm shift. This lies in the way we interact with technology. Until now, users had to constantly consider the very limited technical capabilities of computers and change their behavior accordingly. Recent developments in the field of conversational AI now allow humans to communicate with computers in the easiest manner imaginable: with natural, spoken language.
A number of best practices are essential for a positive user experience. This includes the use of so-called contextual data. For example, the voicebot takes matters into account that have been discussed in both the current conversation and in previous exchanges. The design of the interactions between caller and voicebot is consistently geared towards the customer's goals, as are the course of the dialog and the voicebot's announcements and behavior. A particular focus here is on dealing with problems, i.e., how the voicebot reacts to statements that the caller does not make, are incomprehensible, or are misunderstood.
CUX design incorporates the needs and behaviors of everyone involved. Contact center employees are recognized as an independent user group and are involved in the design, development, and optimization of conversational AI solutions. The collected customer data makes it possible to evaluate interactions with a chatbot, voicebot, or Co-Pilot. This mapping of authentic behaviors allows companies to continuously optimize their conversational AI applications and make the customer experience ever more efficient and enjoyable.
3. Conversational AI (CAI) platform
A conversational AI platform offers a wide range of functions for developing, implementing, and managing intelligent and interactive dialog systems. Among the key features are dialog management, integration interfaces, omnichannel capability, and advanced analytics. Providers of these platforms vary from large hyperscalers, to include AWS, Google, and Azure, to specialized providers such as Spitch and Enterprise Bot. Companies often use a best-of-breed approach with a portfolio of niche products that has developed over time.
We expect that companies will consolidate their platforms, as this simplifies supporting a cross-channel customer journey and facilitates the distribution of data and use cases.
| Aspects of selecting and designing an AI platform |
|
|
|
|
|
The following aspects should be considered when selecting and designing an AI platform:
- Supporting intuitive dialog
Utilizing advanced NLP technologies is critical to creating a natural conversational experience. A system that learns from interactions and continuously refines itself contributes significantly to customer satisfaction. - Scalability
An architecture is needed that can scale with increasing user numbers and expanded use cases. It should also enable integration with existing systems and support multiple communication channels, but without neglecting security and data protection requirements. - Choice of infrastructure (on-premises vs. cloud)
This decision depends on the specific needs in terms of flexibility, costs, and data control. Cloud solutions offer scalability and lower initial costs, while on-premises solutions enable complete data control but require more effort to operate. The options for integrating peripheral systems, which may be hosted on premises, can also influence the choice of infrastructure. - Selection of conversational AI technology (hyperscalers vs. specialized platforms)
Large cloud providers offer comprehensive, customizable solutions. Specialized platforms offer rapid implementation but can be limited regarding scalability and functionality. The pricing also differs between cloud offerings, which generally charge on a per-use basis, and specialized platforms, whose pricing models often include basic license fees. - Integration options
The ability to integrate the AI platform into existing business processes is crucial for effective processes and a seamless user experience. If company data is used across departments in compliance with authorizations and data protection guidelines, relevant and fact-based support is possible. Combining enterprise search with generative AI, known as retrieval-augmented generation (RAG), is an important aspect here.
4. Security and data privacy
In conversational AI projects, especially when using cloud solutions and generative AI, the aspects of security and data protection are key. This is because they ensure system integrity, as well as the confidentiality and protection of personal data. With the growing offering of transactional services via conversational AI, which require a direct connection to security-relevant subsystems, the risk of data breaches and cyberattacks is increasing. Transactional services often process sensitive information such as payment data, personal identification numbers, and other private data. If these are compromised, considerable financial and reputational damage can occur. Therefore, the following important aspects must be taken into consideration:
- Identification and authentication
Features similar to those verified by call centers through telephone channels can be checked in order to identify and authenticate users. For example, voicebots or chatbots can ask for a customer number or social security number. Various pieces of information can be requested and combined to authenticate a user, such as the account balance, the total amount of the latest invoice, and details of the client’s product portfolio, for example, whether a partner credit card is included in the account. This information is then compared with the data from the CRM or other third-party systems. This authentication mechanism should limit the number of attempts in order to prevent brute force attacks, in which attackers systematically try out all possible combinations.
Another method for authentication by voicebots is voice recognition, which companies such as PostFinance and Migrosbank have been using successfully for years. Despite the emergence of deepfake technologies, this method is considered sufficient for medium security requirements when combined with other authentication measures.
The most elegant und user-friendly authentication solution is often the one that is already present in existing portals and mobile applications. If there is integration with the conversational service, it can serve as a foundation for chats and even voicebots, ensuring a transparent and seamless experience for customers.
When selecting the authentication method, it is important to note that each additional authentication step can affect user acceptance and increase the abandon rate. Therefore, the most secure method is not always the best choice. The chosen security level should be in relation to the risks and the likelihood of misuse. For example, a lower security level may be sufficient for ordering proof of insurance, as in the worst-case scenario the customer receives an unordered certificate, so the motivation for misuse is low.
An essential element of security is the encryption of data during transmission and storage. Modern encryption algorithms protect data both at rest and in transit. This prevents unauthorized access and ensures that only authorized parties can read the data.
- Data protection
Implementing conversational AI solutions requires compliance with data protection laws such as the GDPR and the FADP, as well as a strict handling and safeguarding of personal data. When designing and implementing conversational AI solutions, it is essential to take data protection into consideration from the very start. One must analyze the entire implemented business process to identify exactly what data is processed, transferred, and stored in what location and in what manner. Data should generally only be stored for as long as necessary. Access to this information must be strictly regulated by technical and organizational measures.
Particular attention must be given to specific personal data, such as voiceprints, health data, and information on sexual orientation. If possible, these should be stored separately from identification features such as telephone numbers to avoid any association between the data and the person concerned. In addition, the explicit consent of a user must be obtained before data is used for quality improvement or training purposes.
It is also advised that information on the data protection of the used voicebots is presented clearly and comprehensibly on the provider's website to ensure transparency (e.g., the data privacy statement of the Canton Aargau voicebot).
It is recommended that customers integrate all the above-mentioned data protection aspects as early as the concept phase, and to do so by means of risk analyses and an information and data protection concept. Once the solution has been implemented, regular penetration tests can be carried out by external IT specialists, depending on the criticality of the solution.
5. MLOps and GenOps
DevOps is an established field of software development. It mainly deals with the provision and operation («operations», ops) of software systems («development», dev). With machine learning (ML) systems, data is another component that poses a particular challenge. This is why the term MLOps was introduced, which was recently supplemented by GenOps.
In general, AI models degenerate more quickly than conventional software systems and therefore need to be evaluated, adapted, and deployed more often. In the case of generative models, such as large language models, the degrees of freedom are so great that additional attention must be paid to such aspects as hallucinations, bias, and data protection.
GenOps and MLOps combine processes, skills, organizational structures, and tools in equal measure. It can be implemented at various maturity levels and expanded continuously. A key challenge is to design development and operational processes in such a way that they take the special features of AI models into account without compromising agility and the pace of innovation.
|
MLOps practices to get the most out of conversational AI |
|
|
|
|
|
|
|
|
|
By implementing GenOps and MLOps practices, companies can take full advantage of conversational AI while ensuring top operational quality. The most important aspects include:
-
Quality assurance
Automated testing and approval processes help teams ensure the quality of their conversational AI applications before changes go live. This includes both functional tests of conversational interfaces and the performance evaluation of underlying models.
-
Operational efficiency
Monitoring and managing the infrastructure in real time is crucial. Operational metrics and telemetry allow teams to monitor the performance of their models and make quick adjustments if necessary.
-
Life cycle management of models
GenOps and MLOps provide version management and traceability practices that are essential for the life cycle management of AI models. This includes the management of data sets, model versions, and their configurations to ensure seamless adaptation to new requirements.
-
Security and compliance
In view of the GDPR and the FADP mentioned above, as well as other data protection regulations, compliance with security and data protection standards is essential. MLOps and GenOps integrate security practices into the development process from the start to ensure that conversational AI applications meet these requirements.
6. Data
The availability and quality of relevant business data are of paramount importance for the introduction of conversational AI systems. This makes it possible in the first place to generate benefits for users. Data is the foundation of every successful implementation and plays a central role in various aspects:
Data platform
A central and solid data platform that eliminates the silos of an organization, and still allows fine-grained access authorization, simplifies the integration of conversational AI systems. Data warehouses, data lakes, or data mesh architectures make it possible to efficiently link and transform data and thus exploit the full value potential. Regardless of whether processing is implemented in real time or as a batch, scalability must be guaranteed with today's data volumes.
Data analytics
Analyzing interaction data makes it possible to understand and respond to customer behavior. Advanced analytics techniques help to identify patterns and gain insights that enable the continuous improvement of conversational AI systems. Up-to-date and interactive reporting combined with appropriate visualization techniques present complex data in an understandable way for decision-making.
Processes
The quality of ML systems depends heavily on the quality of the training data. Careful selection and preparation of training data is therefore crucial. Effective processes for data collection, cleansing, and validation ensure that models are trained on reliable and representative data.
Overall, data is at the heart of conversational AI. A solid data platform, advanced analytics techniques, and effective processes are pivotal to developing high-quality conversational AI systems that meet user needs.
Risks in conversational AI
As a young technology, conversational AI unsurprisingly still bears certain weaknesses and risks to which we devote particular attention in our client projects:
|
Risks in conversational AI |
|
|
|
|
|
|
|
|
|
|
|
- Low user acceptance
User acceptance makes or breaks the success of any conversational AI solution. Even with technological advancements and the implementation of best practices, it is not a given. However, it can be promoted by user-centered UX design and continuous empirical optimization, particularly in error management. The decisive factor is that the service provides users with a clear added value, for example, through convenience, time savings, or resolution of their concerns in terms of first call resolution (FCR). - Overwhelming people and the organization
Introducing a new conversational AI service without proper consideration of its organizational impact can lead to low adoption and utilization, which can negatively impact the customer experience and business case. A comprehensive involvement of all stakeholders in the design of the solution is therefore essential. This is particularly important in service centers, which are often confronted with changes (new products, new rules, new tools, etc.). Successful conversational AI solutions require a coordination of different skills within the organization that need to be systematically developed or mobilized, especially in the operational phase with the continuous optimization of the solution. - Lengthy introduction of subsystems
You can unfold the full potential of conversational AI when it is combined with transactional and personal services – but this requires connection to third-party systems. The complexity of this integration is often underestimated, especially if the required interfaces are not yet available and need to be developed or adapted. Such roadblocks can impact on the business case through higher costs and slow down the implementation plan. This risk can be minimized by having a detailed target architecture and involving the affected system owners within the organization from an early stage. - Insufficient or late consideration of data protection
The requirements regarding data protection have increased in recent years. Although the requirements can also be implemented with cloud and generative AI solutions, this requires the early involvement of data protection officers and IT managers in the analysis and design phase. A late consideration of these aspects can lead to a delayed introduction of the solution. - The business case doesn’t add up
The decision to implement conversational AI for a voicebot or Co-Pilot is based on a business case. This may be the case of improving efficiency, such as a reduction of the handling time by automating the identification of users, or increasing the net promoter score (NPS) with improved customer service thanks to an expanded GenAI-based search function for products and services in a Co-Pilot.
However, the business case may not initially pan out because the envisioned improvements do not materialize for various reasons. It is therefore important to clearly define the KPIs of the business case from the outset and to systematically review how improvements can be achieved through organizational measures (e.g., training), design (e.g. dialogue structure) or technical optimizations (e.g., fine-tuning of the model). It must be taken into account that a significant part of the budget for initial conversational AI projects is used to set up the organization and learn the basic know-how. These expenses at first have a negative impact on the initial business case but can be viewed as an investment.
Is Conversational AI worthwhile for a company?
Yes!
Conversational AI will have a strong and lasting impact on customer service, and this applies to the human, organizational, and technical levels. Knowing how to make optimal use of this technology and how to integrate it into their organization will be a substantial – if not the biggest – challenge for contact center managers in the coming years. They must therefore understand the added value that conversational AI offers for the entire customer journey and incorporate it into a coherent strategy and roadmap.
Admittedly: a long road lies ahead, and it requires mobilizing various skills and functions to ensure sustainable solutions.
But the rewards are worth it: greater efficiency, happier customers, achieving increasingly ambitious economic targets, and a more attractive job for employees.
Take the step into the future
Time for a conversation about how conversational AI is changing contact centers
Conversational AI will be a key element in the first-class customer service of tomorrow.
Has this got you thinking about how to best integrate this technology into your company? In order to develop an individual solution, it would make sense to team up with a company that specializes in the integration of such a solution.
Use the answer to above question as a starting point. You already know the goal you want to achieve: happier customers.
