GenAI The Right Way

Written by: Stephane

Published: February 1, 2024

Many companies use AI chatbots for customer service, leading to problems like user frustration due to inaccurate or vague responses. This often results in humorous but damaging interactions on social media, undermining the business’s reputation. Users sometimes misuse these bots, causing them to behave inappropriately or even harm the company by leaking data or increasing costs through pointless interactions. There’s also a risk of malicious use, like corrupting data, disrupting IT resources, or launching AI-based attacks, costing businesses heavily. Understanding these issues is crucial for effective implementation and management of AI chatbots.

“This often results in humorous but damaging interactions on social media, undermining the business’s reputation.”

With all this in mind, let’s look at the causes of these challenges and how to address them.

The business challenges caused by GenAI bots

GenAI bots are limited by their training data. If trained on public data, responses might be vague and inaccurate due to diverse data sources. Bots trained on company-specific data can be more accurate but risk leaking sensitive information if not monitored carefully. Businesses should ideally use a local model trained on selected, segregated content to avoid data leaks and privacy issues. Thus, it is essential to implement solid guardrails to prevent any unwanted behaviors.

GenAI bots are often seen as tools for detailed conversations, mimicking human interaction. However, research shows users might prefer simplicity and direct results over a human-like experience. Businesses should consider if letting customers provide open feedback is more beneficial than guiding them through structured journeys. Educational research shows topic-led interactions let customers control the conversation, while task-led interactions guide them through specific steps to achieve an outcome. A task-led approach is advantageous for keeping interactions relevant and within context, avoiding unpredictable customer behavior. This makes task-led interactions a better strategy for digital customer care, setting clear limits on interaction types and content.

“This makes task-led interactions a better strategy for digital customer care, setting clear limits on interaction types and content.”

In the next section, we will discover how Sweepr; the AI-powered Digital Care Platform for Service Providers and Smart Home Providers; has decided to combine both the capabilities of a powerful digital care journey orchestration platform alongside with Large Language Model capabilities to ensure your consumer data are being leveraged to offer resolution-centric, highly personalized experience focused on solving customers’ care problems. 

Sweepr’s approach to harnessing GenAI while protecting your brand and your data 

Based on the above list of challenges, the savvy business will need to find a solution that gives them the ability to safely benefit from the scale gains an LLM can deliver while controlling the data being shared, the type of interaction actually taking place and supporting the ability to personalize the content to deliver best in class CX.

When considering the emergence of Generative AI in 2022, Sweepr decided on a powerful and flexible hybrid approach that will meet all of these criteria.

“Sweepr integrates with any type of LLM…”

For starters, Sweepr integrates with any type of LLM, including any of the mainstream well known vendors in the market today. You can use an existing vendor, have your own model or rely on ours. The recommended use case would be to train a model on a segregated knowledge base that contains only the documents and data relevant to solving your customer care issues. That can include documents, articles, areas of the corporate website…

From there, the Sweepr solution has multiple ways to leverage on LLM in its digital care flows: 

  1. Sweepr can be used as a prompt engine where it will know, based on intent and context, which questions to ask and how to ask them.
    This approach, although powerful, relies on the LLM to deliver the data and passes it through to the end-user after ensuring editorial formatting. This is especially useful when no particular workflows have been created in the system for a specific intent and you want to provide some useful answer anyway.

  2. Sweepr can use the LLM to generate the structure and content for a workflow but making this information available to the public requires review and validation within our authoring environment, Experience Builder (EB). This is the preferred solution as it will allow you to leverage GenAI to create your workflows faster but will still enable your authors to: 
  • Review and adjust the content to avoid hallucinations and incomplete answers, tuning LLM prompts using Sweepr’s low code tooling
  • Add personalization using the platform where appropriate to deliver superior CX
  • Create consistent experiences for all your customers. LLMs do not systematically return the same results for the same query leading to confusion
  • Create outcome-focused interaction to avoid waste and usage that does not fulfill any business goals
  • Treat the LLM as an untrusted user to fully comply with a Zero Trust approach and avoid risking giving unwanted access or information to the public

Finally, the way interactions are designed within Sweepr will give you the best of both worlds when it comes to choosing between topic-led and task-led interactions. Indeed, our approach is to open up the dialog initially and give the opportunity to the end-user to share an utterance of the problem they are experiencing in their own words. From there, our platform infers the intent and leads them through a series of navigational choices that are directly in line with the resolution of the issue. By applying a task-led strategy, Sweepr puts in place the necessary guardrails to ensure the interactions with your customers stay focused on resolving issues rather than becoming a dialog to nowhere with limericks, jokes and riddles along the way.

Bottom line, Sweepr’s technology truly enables superior digital customer care using personalized insights, leading to category-defining level of Digital Resolution.

To find out more, please contact me or visit