4 ways companies that value their customers should use AI (and 4 to avoid!)
The capabilities of large language models (LLMs) continue to improve and offer huge potential for companies to streamline their operations. But consumers are becoming wise to the perils of “AI Slop” - cheaply generated AI content that’s fluffy, wordy, and corporate but provides little valuable information. Companies looking to project a premium brand, one that place a high value on customer or user experience, must carefully think how to use AI: avoiding the use of AI will leave you in the dust compared to competitors but using AI in the wrong way will alienate customers by delivering a sub-par experience.
Note that this advice is not applicable in all cases, and part of your AI strategy should consider what constitutes the best overall experience – some businesses need to prioritise cost or speed to the consumer over the quality of the experience, in which case a very different AI strategy is required.
In our work with companies, we’ve come across 4 ways we recommend using AI when the quality of the customer experience is the top priority:
Ways to use AI
1. AI to summarise
LLMs are excellent at taking large swathes of text, be it internal documents and policies, customer interaction history or external information, and producing accurate summaries of the information (sometimes with citations linking to the source documents).
Using AI to surface quick summaries to human operators, for example in customer support situations, can improve the customer experience by giving the operator more information in less time and effort than without AI. Providing the tools are designed to allow the operator to dig down into the summary and find the source information, the human operator can avoid the issue of incorrect information being produced by the AI. By using the summary to quickly find the detail required, they can quickly provide the response required.
2. AI to triage/classify
Closely linked to summarisation, using AI to classify incoming requests or triage them into priority queues can represent great efficiency gains and still maintain the customer experience.
LLMs are able to understand requests more effectively than any other AI technique to date, and can easily parse free text to extract key information. Companies can use this to prioritise requests ready for a human to look at. We believe this is the best way to utilise AI for premium customer service – give your staff a prioritised list of requests allows them to service the most important customers and focus on addressing the underlying issue rather than triaging the request themselves. AI is an efficiency gain, but the customer experience is preserved.
Unlike completely automated decision making (which we don’t recommend in this case), prioritisation doesn’t completely reject requests, meaning that even if the AI is wrong the customer is still serviced.
We’d recommend allowing exception workflows and giving human operators some flexibility to ignore the prioritisation to spread the risk of missing high-priority requests that have been incorrectly identified as low priority.
3. AI to brainstorm
It’s well known that sometimes the only way to progress an idea is to bounce ideas off of someone (in software development we call it “rubber duck debugging”, making use of a toy rather than a person), and LLMs are perfectly situated to provide ideas and lists of starting points. It’s important to remember that LLMs often represent the “average” of the knowledge they’ve collected and are unlikely to provide completely novel ideas, but in most cases they’re a good jumping off point and can help break a “writer’s block” situation.
Closely related to brainstorming from scratch, we believe some of the best AI tools in business integrate with customer data (e.g., contact history, purchase history) and use this to prompt human operators with new ideas based on a personalised view specific to an individual customer (e.g. prioritising up-sell based on a transcript of the last call that mentions details).
4. AI to edit
While we don’t recommend using AI to write content directly, we do believe LLMs are an excellent editor. Asking AI to rewrite certain tricky sentences, make sure content is on-brand with the company voice or meets customer messaging guidelines are all very successful methods of using AI to improve communications while avoiding the pitfall of directly drafting content. It is important to not apply AI to every sentence – AI has a distinctively average, corp-speak style voice than can come through if every sentence is heavily edited by AI.
AI Use Cases to Avoid
There are many ways to use AI that, while popular and offer compelling efficiency increases, we recommend avoiding when the customer experience is paramount. Here are the top 4 ways to avoid using AI in high-touch or premium customer experience cases:
1. Writing content directly
The converse of using AI to edit, using AI to directly write content comes with many pitfalls that represent disaster when the relationship with the customer or end user is highly valued. AI’s distinctive tone of voice and long-winded nature means this content is often a poor experience to read, and many end users are becoming wise to this tone (~50% in one estimate).
LLMs are able to, and frequently do, “hallucinate” information, which requires careful review to catch. We recommend extreme caution when sending the output of LLMs directly to customers, and in almost all cases where high-value customers are involved, we recommend keeping a human in the driving seat. Instead, use AI to suggest topics and edit the content for tone of voice while keeping a human touch.
2. Chatbots
There are some circumstances where chatbots do have a use – ~31% of customers don’t attempt to self serve before contacting customer support, so using well-built AI chatbots to assist these customers can deflect from more expensive channels. However, care should be taken using chatbots in place of humans, especially when you’re looking to provide a high-quality experience.
The issues of “hallucination” from LLMs apply to chatbots, and it’s likely you will be held liable for any misinformation that your chatbot gives out – Air Canada was held liable for misinformation from its chatbot by a civil-resolutions tribunal.
Additionally, LLMs tone of voice and long-winded nature make interactions with AI powered chat bots annoying.
When you’re looking to keep a great relationship with customers, the frustrations of a bad chatbot and resulting impact on customers can easily cost more than the chatbot saves. Overall, we recommend extreme caution using chatbots where accuracy or quality of experience is top of your list of priorities.
3. Unsupervised decision making
The problems of using AI to write content directly also apply to automated decision-making processes, and often apply more severely. Again, the issue of LLMs getting information wrong applies, and numerical AI models (Machine Learning, in old-school terminology) all come with false positive and negative rates. While careful training of the AI on a wide variety of data will mitigate most issues, bias can come from surprising places – subtle differences in the dialect people use can cause bias in LLMs.
Additionally, customers in the EU have a right to not be subject to 100% automated decision making, so at minimum a manual review process is required by the GDPR. Where a high-value experience for the end user is essential, we recommend using AI to triage and advise while leaving the final decision to a human.
4. Replace all customer interaction / human flexibility
When you’re looking to differentiate yourself based on customer experience, removing all flexibility from your staff is the easiest way to fail. Reducing the customer experience to a fully automated process with no escape routes will leave customers frustrated – inevitably not all needs will be captured by the automated process, and many customers won’t have the skill to navigate it effectively.
Even if some human interaction is left, it’s demoralising for staff to be forced into rigid processes dictated by the AI, leading to disillusionment and poor performance by staff, which impacts their interactions with customers. Instead, use AI to empower human staff by giving them access to information and recommendations, but leave the final decision in their hands.
Our Recommendation
We boil our recommendation down to one simple rule: When you’re looking to provide a high-quality experience to customers or end users, AI should be supporting skilled human operators, not driving interaction unsupervised. When AI is used to provide humans with enhanced tools and information, the human connection that’s required for exceptional customer experience is preserved while still taking advantages of the efficiencies and capabilities of AI.
Making a clear decision on how you will and won’t use AI is an important part of an AI strategy – check out our full series of articles on defining an AI strategy for more on how to build out your path to successful use of AI.