Large Language Models (LLMs) and Actionable Insights Enable Knowledge Curation for KCS

GenerativeAI is unlocking new opportunities to ensure quality content in both the Solve and the Evolve Loop, helping Knowledge Workers, Coaches, KDEs, and Leadership to do their jobs without adding extra steps.

As part of our KCS Aligned Tool series, Brian Corcoran and Lynette Ledoux demonstrated how SearchUnify’s Knowbler tool and analytics suite are accelerating success for their customers.

The presentation covers how the integration of AI and analytics enhances KCS. It focuses on capturing case knowledge, differentiating article types, centralizing coach activities, visualizing content gaps, prioritizing content creation, and recognizing contributions of Knowledge Workers.

Recording

Watch the session recordingaccessible to Consortium Members via login. If you are a Member and need access, please contact support@serviceinnovation.org.

Resources Shared

Chat Q&A

  • Jessica Peralta (she/her): We are using copilot for summarizing zoom transcripts.
    • Elliot Acosta (he/him): I’m curious, why copilot over the native AI Zoom tools?
    • Jessica Peralta (she/her): We have copilot in a secure envirnemtn for a few people and have not enabled the very public zoom AI
  • Eli Dornor: Is Knowbler capable of working with private LLMs?
    • Taranjeet Singh | SearchUnify: Yes it can, if you are hosting LLMs on your own infra like mistral/azure or using third party api like converge.ai, it is pretty flexible.
    • Garima: We follow a Bring your Own LLM (BYOL) approach and give you the flexibility to integrate Knowbler with Public, Private or Partner Provisioned LLMs as per your use case.
  • Doug Brockbank: Can you give a few beginning steps on how to get started and working and training an LLM (e.g. Identify an LLM tool to use, giving instructions, etc.)
    • Taranjeet Singh | SearchUnify: The safest and easiest way would be to create an azure account, They have a pretty intuitive interface, drop the content you want to train on in their UI and then start playing with it.
    • Doug Brockbank: Is it free to use Azure? And do you have step-by-step instructions on how to go from creating the Azure account, to dropping content to train on it, and then testing it?
    • Taranjeet Singh | SearchUnify: Not free, https://learn.microsoft.com/en-us/azure/machine-learning/how-to-use-retrieval-augmented-generation?view=azureml-api-2. This should help.
  • Michael (Kaiser): We inherited very unstructured and complex content that doesn’t follow a KCS standard. How does Knowbler or any LLM for that matter handle that in its training set? Is there data preprocessing we need to complete from knowledge management prior to ingestion and training these models?
    • Victor Feyen | TOPdesk | Belgium: That could be tricky, since you’re asking the LLM to transform non KCS content into KCS content. It would have to rely on other training data to understand and interpret your existing content, and transform it into something else. Would still be cool if that would be possible
    • Russ Brookes, Avaya: I would think you could use few-shot prompting to pass it an unstructured article, and a manually created article as an example (or a few examples), then provide it a new unstructured article and ask it to structure it. Then build from there.
    • Michael (Kaiser): That’s my thought too. I just don’t know how well that would work while avoiding hallucinations because the steps are very business specific.
  • Kristin Hunter | Strata: I’d love to see some examples of the evolve loop content recommendations as well as VoC gaps.
  • Michael Fisher: We’re getting ready to implement a tool called Observa.ai, which listens to a Genesys agent phone conversation and then provide an interaction summary at the end. When we implement KCS, I see the opportunity to leverage those call summaries as the basis of new KCS articles. Am I hallucinating?
    • Heather Ausmus: Do your support agents open tickets when they are working with a customer on the phone to track their productivity/provide case history to the customer? Do you also require they summarize the solution in the ticket they created? If yes, on both, I think you’re covered and don’t need to capture the call summaries for articles.
    • Michael Fisher: many of our agents are offshore/ESL and don’t express themselves in American English. AI assistance could help here.
  • Eli Dornor: Can you use Knowbler with another cognitive search solution like Coveo, IBM or Microsoft?
    • Lynette Ledoux | SearchUnify | New Orleans: Yes, you can!
  • Greg Weinert: Great feature to have the ‘helpful/reference article’ categorization – helps in the progression of the case handling – even if not a direct solution (and doesn’t confuse with it being the actual resolution article)
  • Eli Dornor: Is the SU analytics Case section part of Knowbler? Additionally, what’s the ease of reporting with Knowbler.
    • Tanya Dhar | SearchUnify: Currently SU and Knowbler analytics are different entities. We have a dedicated adoption dashboard for Knowbler that we will showcase today.We also plan to combine the sharing analytics of SU and Knowbler in future.
  • Heather Ausmus: Is the Knowbler in app experience similar in Zendesk too?
    • Mani (SU): Yes
  • Eli Dornor: Do agents have access to the feedback from the Content Standard Checklist? Within the context of Knowledge creation – What fields are leveraged from the case to create the draft article? And are the case comments also leveraged in the draft article?
    • Tanya Dhar | SearchUnify: Agents will have access to the feedback comments that are put in while reviewing the comments. As for the fields used to create article it is totally dependent on the configuration done by the admins , any custom , standard field or comments can be mapped to generate the article.
  • Nasir Mehdi- HCL-India: Do we have an option to do a PAR (Process adherence review too to evaluate Contribution index?
    • Tanya Dhar | SearchUnify: We have this planned in the roadmap.
    • Eli Dornor: is your roadmap available to the public?
    • Garima: Yes, Lynette will present it right at the end of this presentation 🙂
  • Doug Brockbank: Do you have a way of measuring ‘Knowledge Contribution’ of agents?
    • Mani (SU): Yes, we have an analytics report on that
  • Arun Vats: Which KCS Compliant tools is knowbler compatible with? Does it work with BMC Helix solution?
    • Tanya Dhar | SearchUnify: We have support for Salesforce, Zendesk, MS Dynamics, and are currently building the integration with Khoros.
    • Jon Chellis: ServiceNow?
    • Tanya Dhar | SearchUnify: We are planning on ServiceNow and Confluence integration after we make Khoros integration live.
  • Michael (Kaiser): How well does Knowbler handle articles that contain complex CSS code within the articles? Like an article with steps in a HTML table and screenshots?
    • Jason Swinamer: How are screen captures and videos handled in Knowbler
    • Mani (SU): Knowbler can support rich text field and with custom editor that supports media as well
  • John Coles: Can you talk about how it harvests content by looking across Cases that do not have an Article Attached?
  • Brad Smith: Do you have a slack integration?
    • Mani (SU): Currently not, but R&D team is working on it.
  • Paul Stejskal: Are there ways to filter on some of the views like the low article visibility for different verticals? We at NetApp use KCS across multiple product lines and specialities within our support center, so we need per speciality/vertical reports, as just one example.
  • John Coles: How does it interact with Field Service that does not use a HelpDesk Console?
    • Sara Feldman | Consortium | Las Vegas: Yes – we have Members exploring KCS in Sales and HR (what Brad just said)
    • Jessica Peralta (she/her): We are doing KCS with engineering processes. Steering committee + KCS global council.
    • 01:04:43 Kelly Murray | Consortium | Seattle: I think there’s an important difference between “centralizing content” and inviting everyone in the organization to participate in knowledge creation & sharing.
  • Traverso, Alyce: I’m interested in hearing more about how LLMs, Generative AI and SearchUnify can use case notes to create drafts of Knowledge articles.
    • Mani (SU): Yes we use AI capabilities to redraft/refine it
  • sarahmass: Beyond the ability to surface existing knowledge and detect duplicate content, could this potentially take it a step further and help identify specific updates that need to be made, and/or content that’s no longer accurate? For example, highlighting a piece of content that we should revisit and maybe update.
    • Tanya Dhar | SearchUnify: We are building a feature wherein we can input the release notes/any change that might impact the KB and based on that input showcase a list of articles that might need an update. This is still in R&D though.
  • Eli Dornor: What are the risks that concern Search Unify regarding Knowbler and the trajectory of the industry, and what excites Search Unify about what’s to come?
  • Eli Dornor: Does Knowbler prompt agents to adjust behavior if they don’t follow Process Adherence?
    • Victor Feyen | TOPdesk | Belgium: First gut reaction: isn’t that what you have KCS coaches for?
    • Heather Ausmus: @Victor Feyen | TOPdesk | Belgium I thought the same thing. 🙂
    • Eli Dornor: A refinement to my question is that can Knowbler prompt someone who seems not to be searching or not attempting to capture or create a KB?
    • Eli Dornor: KCS Coaches should assist in prompting the right behaviour however, in process correction goes a long way. In our adoption journey, we have some knowledge workers that almost immediately forget what they were told a week earlier.
    • Holly Palmer – Veeam Software: Yes – that’s what I was thinking…
    • Stejskal: This tool would just become coaching by e-mail but a computer doing it instead of a human…
    • Paul Stejskal: At the end of the day the coach should be having 1:1s with a knowledge worker under him, and if they don’t show up it becomes a management problem.
    • Eli Dornor: I don’t believe it can replace the human component that coaching brings.
    • Eli Dornor: Wouldn’t it violate the principle that all knowledge workers are volunteers?
    • Paul Stejskal: Do you want KCS to be an optional thing though? We have it required for our knowledge workers and have high management buy-in at high and low levels.
  • Travis Myers JLG Industries PA: What about using transcriptions with the case data to create KB content?
  • Justin Loera: what is the tech stack for the LLM?
  • Harm: Can Knowbler handle multiple languages as far as customer context and analyst case notes, resulting in content that is all in English?
  • Jose Escobar – Sysdig Support: Is there any way to tell what changes were made by the agent on top of the draft provided by Knowbler? Just to see who is just saving the KB as it is vs who is making an effort to improve the suggestion.
    • Tanya Dhar | SearchUnify: Not currently , but this is something that can be worked upon for our future releases. Thank you!
  • Denisha Jennings: Is there a way to track changes on an existing article vs reviewing two separate versions?

Kuddos

  • Doug Brockbank: Terrific presentation! Amazing product!
  • Michael (Kaiser): I really appreciate it! Thank you all
  • Julieta.Estades: Great topic! Thanks!
  • Heather Ausmus: Thank you!
  • Erin McGarry: this was great – thanks all!
  • Cheryl: Thank you all! Great!
  • Beth Coleman | CATALYNK: Thank you!

Leave a Reply

Your email address will not be published. Required fields are marked *