Generative AI For Customer Service:
8 Best Practices For Success

ChatGPT has reignited interest in the broader domains of generative AI, AI and knowledge management (KM). In this article, we will focus on how generative AI can be harnessed for customer service automation and contact center agent augmentation. Here are eight best practices for success.

Deploy Prudently

Instead of taking an all-or-nothing approach to deploying generative AI, we recommend using an activation framework based on risk and value. “Risk” could mean risk for the business (e.g., compliance) and/or the consumer (e.g., medical question), and “value” could mean customer value (e.g., value of a customer’s deposits or net worth) and/or the transactional value (e.g., shopping cart value).

  • Low risk, low value: An example is a cross-sell recommendation when a shopper is buying a commoditized product. High automation, enabled with generative AI and with no human supervision, can be deployed here.
  • Low risk, high value: An example is when a shopper is looking for advice on which laptop to buy. Since the risk is low, high automation can be applied to this scenario as well, but since the customer (e.g., B2B client) or transactional value (laptops can even cost thousands of dollars) is high, we recommend at least a “minimal” level of supervision—for validation of generated answers/advice.
  • High risk, low value: Even though the value is low, this scenario still warrants fairly high supervision since the risk is high. Since the value is low, businesses in a low-regulation environment can perhaps get away with some automation. An example is selling consumers a new account in a bank. The deposit may be minimal, but the cost of noncompliance can be high.
  • High risk, high value: Examples are medical advice sought by a consumer for a non-trivial illness or product advice for a complex product like medical equipment. We suggest going with no automation and human involvement before gaining more experience.

Identify Trusted Content

Enterprise content tends to fall into three groups: Curated, documented and in-process/developing. The first step is to group content assets into these categories.

  • Curated content: Curated content and know-how is correct and compliant with regulations and organizational best practices and consumable in the flow of work.
  • Documented content: Documented content is correct and compliant but difficult to consume. An example is multipage, procedural documents. While they may have been approved, it is onerous for agents to find the answer that may be buried somewhere in that document.
  • Developing content: Developing or in-progress content may be good draft versions of content but not 100% correct and compliant since it is still a work in progress.

Remember GIGO

While generative AI technologies have already shown great potential to enhance business productivity and elevate user experiences, they are also known to make mistakes or make up stuff (i.e., hallucinate). The GIGO (garbage in garbage out) concept applies here as well. It is therefore important to train bots on your trusted data and content (i.e., curated and documented content), which elevates the importance of a robust KM system in your organization.

Integrate Into A Knowledge Hub

Generative AI should be an integral part of your organization’s knowledge; otherwise, you risk creating another silo of inconsistent knowledge, which is a big deterrent to good customer service, according to many consumer surveys, a Forrester survey on CX sponsored by our company among them. An unsiloed approach is to create a knowledge hub that unifies and orchestrates all the building blocks needed for modern KM—content management, profiled content access, natural language processing, intent inference, personalization, search methods, conversational assistance, process guidance, and knowledge analytics, powered by conversational AI, generative AI, and ML, all in one place.

Avoid Vendor Lock-In

Make sure your KM software provider offers a BYOB (bring your own bot) architecture that allows you to plug in any LLM or bot (ChatGPT and Google Bard, for example) onto your AI tech stack.

Pick Sweet-Spot Use Cases First

It’s best to get your feet wet with sweet-spot use cases that best help you address your goals before experimenting with uncommon ones. Among them are:

  • Identifying likely questions from specific audiences or on specific topics. This will help scope the KM and generative AI project.
  • Repurposing curated or documented content to fit your brand values, brand voice, and the target persona.
  • Generating draft-curated content from documented content. This will ensure that the curated content is not only correct and compliant but also consumable.
  • Simplifying feedback on answers to make them more easily actionable. This will help optimize knowledge-base effectiveness faster without having to go through reams of raw feedback data.
  • Automate customer interactions. Answers can be generated from the customer’s question and content retrieved from an approved knowledge base and repurposed for persona, customer sentiment, and brand voice. It can then be evaluated for semantic match and if the match exceeds a certain confidence level, it can be surfaced to the customer without human intervention.
  • Agents spend a considerable amount of time creating call notes at the end of a call or before a transfer. Generative AI can be used to automate the creation of such notes.

Analyze For Continuous Improvement

Ongoing analysis is critical to optimizing your deployment. Analytics will help you answer questions like “What is the effectiveness of a generated article in improving metric X?” “What query topics are missing?” “What is the productivity of knowledge author A versus B?” and so on.

Mitigate Risk

Partner with a proven solution provider in the AI and KM domains. Ask for a risk-free, cost-free pilot, where you can experience the solution in a production environment, to see if the vendor will put real skin in the game. Best practices and client success at scale are equally important. Vendors’ domain experience in AI and KM can make the difference between boom and doom.


Adding generative AI as an integral part of your overall KM strategy is a winning approach that is bound to add transformational value to your business when combined with best practices.

This article was first published on
Skip to content