GenAI for KCS

Leveraging Generative AI in your service organization is a hot topic all over the world. Companies, as well as countless tool vendors, are trying to determine the most appropriate use cases for GenAI. The first wave of use cases is centering around using GenAI as a co-pilot for agents or knowledge workers.

David Kay, Principal at DB Kay and Associates, shared several use cases that he has experimented with in the KCS Solve and Evolve Loop. We had great chat conversations from the participants on what is working well for them and where to be cautious.

Examples include:

  • Structuring KCS content
  • Creating titles, short descriptions, and summaries
  • Creating initial drafts
  • Diagrams and visualizations

Recording

This session was full of actionable information and useful prompt examples to start using right away! See a PDF of the slides, or watch the recording and read chat highlights below.

Resources Shared

Chat Highlights

  • Ryan Mathews | NetApp: What is your ROI on the pumpkin toadlet?
  • Adam Mullen | athenahealth | Remote, Maine, US: I’m so glad I understand that reference now

KCS Workflow

  • Patrick McBride | Oracle | California: Dynamic content creation is only as good as the documentation in the Case/SR. Always have to start with the users documenting their issue effectively somewhere.
  • Brian Gregory: We ran into this issue with people getting paralysis by analysis on selecting the article type. we ended up using a singular template to take away that excuse
  • Ryan Mathews | NetApp: Are you experimenting in a way that these modifications follow your Content Standard in an automated way?
  • Lena Forrest: David, what were the prompts that you used for ChatGPT to bulletize the text and get the template? Is it straightforward  or did you have to play with it to get a good answer?
  • Jakub Adamek: GenAI can be used here on multiple levels 🤔
    • 1. Take unformatted content
    • 2. Pick the right template
    • 3. Format the content based on the template
    • And the more I think about it the more steps can be added like “generate metadata like keywords” etc. Those are interesting ideas.
    • Roberta Miller: I have also used it to generate keywords, you can toss out some, but it has captured some that I have missed.
  • Cyndi: We ingested a simplified version of our content standard gave problem description and resolution from case asked for KCS step by step article w/ open book and closed book referencing…the technical and formatting were great
    • Daniel Miller (Geotab): How did you “Injest” it?
  • John Coles: Is anyone using GenAi to harvest KAs from Case Notes???
    • Koree Mires | Cisco Systems | Dallas: We are experimenting with it.  Nothing in full production.
    • Rick Joslin – Cigna | Pittsburgh, PA: Not yet, but in the backlog.   Generate drafts when knowledge is not utilized can potentially fill knowledge gaps.
  • Patrick McBride | Oracle | California: ChatDBKay
  • Travis Myers: Human in the loop
  • Ryan Mathews | NetApp: If you wait till the end, it’s not KCS
  • Brian Gregory: case note summarization for the agent while working the case is a good use case
  • Jason O’Donnell | KCS Nerd | PDX: So, KCS 101: have the ai do the work IN the workflow not at the end 😉
  • Ashly: From my experience, the best written articles were from the people that documented the case notes and steps as they were working.
  • Koree Mires | Cisco Systems | Dallas: David – I agree with this BUT if GenAI begins monitoring the raw conversations, it may do it in real-time which is even more interesting.
  • Travis Myers: We transcribe our cases and have them attached to the case.
  • Ryan Mathews | NetApp: Did you experiment with GenAI for PAR David? Or CSC?
    • Adam Mullen | athenahealth | Remote, Maine, US: +1 for both of these – given the nuance of assessing PAR opportunities versus actions (relevant actions) if a GenAI system (or other combination of ML + LLM) can do that well, that could be a great time savings for Coaches
  • Kelly Murray | Consortium | Seattle: Schrodinger’s Knowledge Worker! Skilled enough to troubleshoot and talk to customers, not skilled enough to make those answers available to more than one customer at a time!
  • Justin Kelley: What’s fascinating with the trust issue is that no one is reviewing a support agent’s response to a customer on the phone before they give it
    • Paul S: And it can be tough to read the case interactions and see how your agents are communicating with customers.
  • PAWAN KHATAWANE: No, David. Not all leaders are misleading. There are many like you who believe in their people more than a technology’s capability to generate text. 🙂

Use Cases and Examples

  • Daniel Goin: In my own experimentation, I have had great success using GenAi to generate article summaries.  We often have had poor summaries because they get written before most of the article, and they end up less than descriptive.  Just asking it to summarize in a single or two sentences and giving it the article content worked wonders.
  • Patrick McBride | Oracle | California: I’m interested in how GenAi could help prompt the user with better questions to diagnose
    • Lynette Ledoux | SearchUnify: TOTALLY!
  • Ryan Mathews | NetApp: I concur….good for writer’s block.  Especially that first sentence.
  • Christina Roosen | Akamai | California: Also great for authors who aren’t confident in their writing or language skills.
  • Jakub Adamek: + using genAI for translation into multiple languages as its miles ahead to tools like Google Translate
    • Koree Mires | Cisco Systems | Dallas: We do this too.
    • Linde Jilesen: what language model do you use? how do you make your translations are secured from a data privacy point of view
  • Koree Mires | Cisco Systems | Dallas:
    • We have 49 initiatives across CX.  7 in production, 8 in development, 34 in exploration.
    • Jeff Elser | Oracle | Bozeman: That’s pretty funny. Oracle is doing the same thing. I have no idea of the count, but I’d bet it’s in the same neighborhood. It’s a real challenge for us to create common architecture and UX experiences with everyone moving so fast.
    • Koree Mires | Cisco Systems | Dallas: We created a CX GenAI Council which my team leads.  We catalog all use cases and are looking to “steer” the architecture and common use cases.
    • Jeff Elser | Oracle | Bozeman: Same at Oracle. Although slowing down to create common architecture is a real challenge. Hard to balance the perfect long-term solution vs. iterating and learning some hard lessons. I kind of favor the latter at the moment since everyone is so new to Gen AI.
    • Adam Marks (he/his/him): @Koree Mires | Cisco Systems | Dallas I’d love to hear more about this. Perhaps through CS
    • Veronica Armstrong | Brother International: Hi @Jeff Elser | Oracle | Bozeman! It’s been a minute since I’ve had to tap into your vast knowledge you share with our team :)I?
    • Koree Mires | Cisco Systems | Dallas: Happy to talk about it with CSI.  I hope to be at the member summit.  waiting for budget approval. 🙂
  • PAWAN KHATAWANE: We have our internal GPT solution built into Azure OpenAI.
    • For articles we have created a prompt book that covers various situations, e.g. 
    • Draft a short description of a “How to” article.
    • Draft a short description of a “Troubleshooting” article.
    • Summarize the resolution text in numbered steps. 
    • Analyze information from a table.
    • Summarize information from a table.
    • Summarize text in bullet points.
    • Summarize text in simple English.
    • Summarize a table of information in simple English and using bullet points.
    • Convert text into table.
    • Summarize Q and A in bullet points.
    • Translation of article
    • Proof-reading and correct the text.
  • David Larsen: Claude AI (similar to ChatGPT) will accept text files so you export the email to PDF and include it in your prompt.
    • J Taylor: Seems like Claude is better at writing prose for stories.
  • Mir: On the topic of use cases, I was wondering about using AI to keep user profiles – skills- updated.
    • Jacob Watts | ParTech | Cali: that would pair well with Intelligent Swarming.
  • Jessica Flohr: What a good first step for starting out with this?
  • Scott Shelton: Has anyone tried anything like this with a “local” AI model, such as Llama V2 or the like?
    • Koree Mires | Cisco Systems | Dallas: Yes. In some use cases it is just as good.  In others, OpenAI is lightyears ahead.
  • Brian Pahl: Is anyone else seeing pressure to replace conventional search with GenAI? My question is: “How do we know if an article exists to resolve the present case if GenAI is providing only a single generated answer?”
    • Patrick McBride | Oracle | California: I do like how google presents both
    • PAWAN KHATAWANE: the GenAI solution should also return the reference articles it has generated a relevant summary from
    • Brian Pahl: agreed. Coveo also presents both
    • Daniel GOin: We are struggling with that exact thing right now.  We have just launched a customer facing chat version and there is a lot of pressure to give agents similar.
    • Daniel GOin: Requiring it to provide the references, and expecting linkage of the articles references if it resolves the issue is where we are starting
  • Laurel Poertner: On the bright side, we are getting much more engagement from mid-line managers to practice KCS due to the visibility and focus it is getting at the C-level.
  • Sara Feldman | Consortium | Las Vegas: Do you have any thoughts on being polite in your prompts (pleases and thankyous)? I’ve heard some people joke about this but also still say they think it helps, somehow.
    • Tomer Shoshan: in my experience ChatGPT tend to ignore politeness and take the key out of the query 🙂
    • Koree Mires | Cisco Systems | Dallas: You can add prompts for that.  Usually to say “use a professional tone”
    • Sara Feldman | Consortium | Las Vegas: I mean being polite TO ChatGPT, haha. As if it cares or gives better answers that way.
    • Roberta Miller: I have done that – Please create a …….
    • Koree Mires | Cisco Systems | Dallas: OH…lol. I do it too as a habit…. Not that it cares really. 🙂
    • Diana Martinez | Oracle | Zapopan: Lol, me too.
    • Jacob Watts | ParTech | Cali: I’ve heard some pretty interesting stories about how well LLMs perform when treated like a human. e.g.: “take a deep breath, you can do this!”
    • J Taylor: Of course, you have to wonder if we get used to not being polite to AI, will we lose our manners with other humans?
    • Veronica Armstrong | Brother International: I haven’t been too polite in my ChatGPT prompts and felt guilty after, so now I am more polite LOL

Diagrams

  • Johannes Hokamp, Waters: Text baked into images is hard to translate (and unsuitable for the visually impaired)
  • Travis Myers: Our Consumers love images, schematic clips for context and locating parts of the equipment.
  • Kendall Brenneise | F5: The diagram is “sufficient to solve”  😀
  • Paul S: All the created diagrams would need to be updated by a human to be useful, the formats were a bit rough.
    • Lynette Ledoux | SearchUnify: As David said earlier, it’s easier for a human to work from a “first draft” than creating something from nothing.
  • Ryan Mathews | NetApp: On this topic of diagrams, we’ve seen GenAI to be quite compelling in creating videos as Evolve Loop compliments to Solve Loop articles.  Especially when it comes to keeping them up to date over time.  Very powerful.
    • Sarah Eshelman: What tool do you use for this?
    • Ryan Mathews | NetApp: Check out Synthesia: https://www.synthesia.io/home
  • Adam Mullen | athenahealth | Remote, Maine, US:
    • Microsoft Copilot 365 + Visio might do the trick too (now or in the future)

Pros and Cons and Conundrums

  • Koree Mires | Cisco Systems | Dallas: The other interesting discussion is how all of these use cases translate to cost savings vs added expense for GPU, storage, etc.  If you can generate docs, solutions, etc that save 100 cases at $100 each great!  But if it costs $100k to do it in development and GPU….  Hmmm
  • Travis Myers: the conversational AI requires a lot of training.
  • Christina Roosen | Akamai | California: Garbage in / garbage out is just faster, more efficient, and happens out of your view!
  • Jeff Elser | Oracle | Bozeman: I’m seeing that often garbage in leads to hallucinations, although sometimes those hallucinations are actually pretty useful and maybe better than the original garbage.
  • Koree Mires | Cisco Systems | Dallas: Larger issue is looming about how GenAI sources info. Eventually, the docs created are generated by GenAI.  Then other GenAI uses those documents to generate more GenAI answers.  It gets into a fascinating spiral.
  • Matt Seaman | Consortium | Boston: ISSIP (The International Society of Service Innovation Professionals) has been doing some interesting work and discussions on Ethical AI.  By most estimates, ChatGPT uses the same electricity as ~30,000 US Households everyday to process the hundreds of millions of daily queries.  It’s an interesting way to also think about the impact of the transition to AI capabilities will have.

Trust But Verify

  • PAWAN KHATAWANE: Got a question: I hear a few knowledge platforms calling their “content generating” capabilities as “knowledge generating” for the knowledge articles. I am not convinced. Of course, the AI technology can generate the text, but it is not knowledge, and it is not knowledge required to support an enterprise IT product or service effectively. The text is generated based on provided internet data to the LLM. We need the knowledge. It is a problem-solving process, it is a struggle. It is a story. It happens in a technical and non-technical environment. It takes resources. It requires human touch, problem-solving skills, and customer service skills. It is a problem-solving experience. It is knowledge. AI technology is a great enabler. It can enable us to convert our experience of problem-solving knowledge into a great and structured article, but it cannot replace the human problem-solving experience of taking notes or capturing the knowledge as a source.  What is your opinion?
    • Koree Mires | Cisco Systems | Dallas: Good question Pawan.  I also wonder if we will have to tag articles as “review by humans” or “generated by AI” based on customer demand to know.
    • PAWAN KHATAWANE: In our prompt book we have advised users to make such a disclaimer.
    • Kendall Brenneise | F5: @PAWAN KHATAWANE I’ll echo your point.  The industry is jumping on artificial “intelligence” with an assumption that the technology is intelligent enough/wise enough to know when to write or not write a line of text, code, phrase, etc. Trust but verify comes into play in these use cases
  • Stephanie Foor | SAP Concur: We honestly don’t trust AI just yet – so having a ‘quality’ check by a human in this process is what we are leaning towards as a short-term solution.
  • Elena Forrest: I have heard in Oracle’s meetings that all AI generated text needs a human to actually review it and save. So AI us used to help the KM or ticket worker, but not to actually post something without human’s review

Kudos

  • Matt Seaman | Consortium | Boston: Love the simplicity of that
  • Stephanie Foor | SAP Concur: Great 1st example using GenAI in KCS – thanks David!!👏🏼
  • Holly Palmer – Veeam Software: I wish all my customers were like you, David!!
  • Elena Forrest, Oracle, VA: Thank you for this interesting webinar!
  • ekeller: Great presentation and discussion, thank you!
  • Jeff Elser | Oracle | Bozeman: Great presentation, David! Thanks for sharing!
  • PAWAN KHATAWANE: Fantastic webinar. A validation for me that we are on the right track. Excited with our GenAI solution but not misled. 🙂
  • Christina Roosen | Akamai | California: Great practical examples, thank you David! And thanks to all the great info in the chat!
  • Holly Palmer – Veeam Software: Thank you so much, David!  Very helpful information that I can use when we start building a plan for AI integration!
  • Adam Marks (he/his/him): Thanks David and CSI.
  • Mir: Great to “see” so many of you!
  • Melissa Burch: As always David; very thought provoking.  Appreciate you taking the time
  • Paul S: Great conversation and information – thanks All.
  • Moriah Yorkey – Tyler Technologies: Thank you!
  • Callie Cella: Very interesting! Thank you!
  • Jessica Flohr: Thank you!
  • Robert Lee: Thank you very much for putting this together.
  • Marcela Gleixner: Thank you!!
  • Sue van Gelder – Consultant – Charlotte: Thank you for this excellent presentation, David!
  • Justin Kelley: Awesome presentation. Really fascinating.
  • Jakub Adamek: Great presentation! Thank you 🙂
  • Abbey.Middleton: Thank you!
  • Tanya Oplanic: thank you
  • Daniel Le Bars (Alcatel-Lucent Enterprise): Thank you!
  • CACHILCOAT: Thank you so much
  • iPhone: Great information, thank you!
  • Matej Dupal | Oracle, CZ: Thank you, David! Great presentation.👏🏻
  • Laurel Poertner: Great stuff DBKay!
  • Stephanie Foor | SAP Concur: This was great!! Thank you David!!!
  • Brian Gregory: Thank you David! Great use cases and information to navigate this landscape!

Leave a Reply

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