KnowledgeBase Questions from Product Roadmap Update

[Gerry]

@Emily Patrick

@BobbyB

Hi, 

Following on from yesterday's meeting, you both asked questions in relation to KnowledgeBase which I did not get to answer live, so I thought I would post here and mention you both.  First of all, thank you for attending the session, I hope you found it of use. 

• "Any news on a knowledge base?"
• "Within SM, is there going to be a dedicated knowledge base? that can be used instead of the current FAQ's? And if so, is there a way to seamlessly move what is currently set up as a FAQ into that knowledge base?"

Thank you for your questions. In terms of the KnowledgeBase, this is very much at the heart of what I referred to as the "call logging experience" and in particular, the way in which technicians and customers should be able to interact with knowledge. I would put is at a 3 out of 10 of what should be possible today and I believe we can get this to 9 or 10 out of 10, maybe even above and beyond that. The key here is, presentation, access to, and purpose of knowledge as well as desired outcomes, and how specifically these things fit into the process of raising requests, or indeed your customers self-servicing/self-resolution. This is very much on my agenda and I will go out on a limb to say there is good news because there is scope for a lot of improvement and innovation in this area, and I will be ensuring we deliver on that. 

In response to: "will there be an easy migration path from FAQ's to whatever comes next", and the answer to that is yes, we will take care of that. 

Please watch this space, and thank you for your feedback and interest. 

Gerry

 

[Stephen.whittle]

@Gerry

 

Are you able to share any indicative high-level timelines around the delivery of the knowledge base? 

[Gerry]

@Stephen.whittle

I want to try and avoid calling it "the knowledge base" as if its a new icon or something like that.  The work around knowledge is all to do with creating a much better experience and integration of knowledge during the call logging and analyst support workflows, a very big part of which will be better access to knowledge, with actionable outcomes, better, more contextual search and so on.  The foundational work around this is already underway and so you will start to see evolutionary improvements in FAQs for example in the coming weeks, with a bigger foundation of other changes following that. Timescales are variable of course because we have many commitments and much to do, but broadly in line what  I had set out in the three step plan, the first of which (the admin stuff, the new ITOM app) are complete now, we are currently working on some deliverable improvements around FAQ's and have just started the next major Stage 2 work around the service portfolio/catalog/request configuration changes, this is going to be a big change, a lot of work and possibly some breaking changes we will have to manage along the way too.  This work in the portfolio gives is a solid foundation to get to the final stage which is the portals, the employee and customer experiences, data models and of course that all important knowledge integration with these processes.   So work in progress, lots to and you will see various incremental improvements along the way during Q2 and Q3. 

Gerry

[Stephen.whittle]

Thanks Gerry, it is useful for us as an org to be able to frame our own internal service development plans around the timelines for new functionality. I completely agree with your intended approach that the traditional sense knowledge base simply won't suffice. It needs be contexualised and align with the various customer journeys for consuming knowledge IE reporting a request, using the portal search bar or potentially invoked during a request lifecycle. My biggest frustration at the moment is around the knowledge space. We have a 20k+ staff base (customers) and delivering content on that scale without an easy to manage central knowledge base with capabilties to be able to plug in search bars, chat bots or direct integration into our intelligent captures makes it difficult to direct our staff to self-help. 

I look forward to seieng the changes in the coming weeks. 

 

@Adrian Simpkins& @SJEaton

[samwoo]

@Gerry - remember I asked the question about AI during the Insights event? I wonder if this might be a suitable time to start having a look at what it could do for the knowledgebase in the future?

The AI would use actual knowledge in the system and build upon that. The AI would be able to enhance user experience and provide the right guidance to the right people with machine-learning.

I am going to admit that I don't have a great understanding of how AI works - but this is my take on it: it involves a lot of work in the short term, for long term gain. It involves machine learning as the AI attempts to learn and categorise as much of the data in the system as possible, so when a user asks a question (specific or otherwise), it can provide the most relevant results through the data we hold on to the system. It can be "taught" to provide responses and ask additional questions if further information is needed. 

It could be used to signpost users to the best Service for an issue they have or provide a link to the FAQ that will help the user solve their issue or provide the right advice, or even log a ticket on behalf of the user if further investigation is needed and there is nothing the AI can provided.

Just a thought.

[Gerry]

@samwoo

Ok so you are going to love my rather "unexciting" view on this subject... "The AI would use actual knowledge in the system and build upon that. The AI would be able to enhance user experience and provide the right guidance to the right people with machine-learning." that is basically an index. Think about what an index is... every time you create a piece of relevant information it is indexed, that could be said that "the machine has learned that new piece of information", such that the next time you look for that thing or something like that the machine (your index) will know about that and will return you that information, with some form of relevance in relation to either the thing you searched for, or, the other things that were also returned. 

"I am going to admit that I don't have a great understanding of how AI works" you are not alone here, AI has been so over-hyped in the market that you think its some magic, but its really not.  There are companies, even in our very own service management tooling space that literally include AI in their product name, why, because it sells... its the sexy new thing that companies invest in.  I have seen so many products essentially call their indexing or their ProCap equivalent or even their business process tools that can make decisions based on user input - "AI", but its not, at least not what my understanding of AI is. 

There is far too much to say on this subject to cover AI as a topic in this post, but I would say this. 

I believe that everything you have suggested in your post above we will achieve, the system will help users get to the right service, and provide the most relevant answers, and may even lead to a self-resolution outcome, this will require quite a lot of up-front work for customers to set up properly , but once set up it will appear like magic.  If we archived that, and then called it "AI" would you consider that requirement met?  Going back to your statement "I am going to admit that I don't have a great understanding of how AI works". if we make something that looks like it works well, and we call it "AI" even though under the hood there is no "AI Technologies" at play, would that be good enough? would you just take our word for it that there as "AI" involved, or would you go and learn enough about AI technologies to refute our claim about "AI" if the solution we said was AI did not work too well?

Here is the thing, I am, very much to my own detriment, "honest" and I do have a basic grasp of how AI/ML works and how it can be applied, and for me to take something that "looks like AI" and call it "AI" when I know deep down, its does not meet that criteria, I just can't bring myself to do it, I simply can't be dishonest like that, I do not want to mislead people, or make claims that are not true, or indeed take advantage of people who, like yourself, "admit that I don't have a great understanding of how AI works" which means we don't get to sell fantasies or tick the AI box because my own personal pride and professionalism gets in the way.  

Now do not get me wrong here, I am not saying AI is not a real thing, it is, and there are some really extraordinary use-cases for AI/ML in use today, but they are very limited in their scope, its best said like this...

AI is not a general technology that applies to any software and any situation to make it better, rather,AI/ML is a general set of computing ideas that have very specific applications, that applied right can be infinitely more effective than a human intelligence. 

The general principle of AI/ML is this.  You feed the learning model known good, tagged, data, that in the data sample has a high degree of repetition of characteristics that can be classified.  For example, in image recognition, you feed in images, let's say of animals, into the model, telling the model what each animal is, each time it gets a new image, the model does its adjustments looking for best accuracy of results.  After a decent enough sample of tagged data, the model is now able to recognise animals in other images that it has not previously seen in the training phase, "with some degree of statistical accuracy".   The output is a statistical likelihood of the image being a cat or a dog etc but its not 100% precise...  thats a textbook example of AI that is one of the earliest use cases. 

Refinements on this might be multiple models, say for facial recognition, where the first model might generally identify characteristics of a face, for example, gender, race, hair style/type, eye colour and many other generally identifiable, then using those identified characteristics, a second model might be used to take a much smaller subset of individuals from a national database to try and identify the individual person. 

Yet another example is, cancer screening, feed an ML model with tagged images of chest x-rays for all patients over the last 20 years that were screened for lung cancer, telling the model which ones had a positive and negative diagnosis, once the model has seen enough data, you can now feed in new x-ray images and the model can help screen and identify potential positive diagnosis.  This is a great example that is used in real life today and it works, it saves lives for sure, but its not foolproof and there is probably at least 20 years to trials now to see if, and I know that this sounds harsh, but to see if the false negatives are a low enough percentage to be acceptable, but until then, while the ML models are getting it right 80-90% of the time, thats well below the threshold of it being an *alternative* to a qualified medical expert looking at each x-ray.  Of course people also makes mistakes, so its reasonable to think that this technology should get good enough to be on average, better at diagnosing this kind of thing than the human equivalent. 

There are so many other AI use cases, but they all fit this pattern, speech recognition, image recognition, big data pattern recognition is large scale retail/consumer information, moving image recognition and many related technologies around joining the dots between expected and differences, for example, moving vehicles on roads in self-drive and driving assistance systems. 

So back to your typical service desk environment, its much harder to find a use-case that will actually make a material difference, and that is certainly true in relation to the cost of truly realising AI. I don't know of any servicedesk deployments that I have ever seen that have anything like the amounts of relevant long-lived data needed to train a model with anything meaningful, and, in IT especially, things change so very quickly, stuff thats relevant today will not be relevant in 12 months time.  Think about it like this, if you wanted to train a model on Windows 10 errors and make it really smart so customers could help themselves, you could do that, but by the time you have collected enough data, trained the model and deployed into use, its likely that Windows 11 will already be replacing Windows 10, but in my example above, screening for lung cancer for example, those x-ray images and the characteristics that identify lung cancer will only change over many 100's of years, so the ML model has time and enough data to learn from and get better, it will not suddenly be the case that everyone now has Lungs 2.0 that are fundamentally different to Lungs 1.0. 

The bottom line is, for most companies "peddling" AI, it is more marketing than it is of real value for most use cases. There is a lot of snake-oil selling of stuff with the "AI" badge precisely because buyers are  "going to admit that I don't have a great understanding of how AI works", and are going to buy into the idea of AI because it looks good, so, if they see something that looks like AI because it works well, and its called AI, then to them its of course it is AI... but for me thats really dishonest marketing and not something I have ever been able to bring myself or Hornbill to do. Of course that does mean that when we get those tender documents and they say "you mush have AI" we generally say no here.. so there is continued commercial pressure on us to play the same marketing gams around AI ... sadly... and that can cost us customers, but for me, our professional integrity is more important. 

Gerry

 

 

 

[samwoo]

2 hours ago, Gerry said:

Ok so you are going to love my rather "unexciting" view on this subject...

This is a fantastic view and makes a lot of sense when I think about it.

I don't have a mindset for a lot of things going in the industry at this sort of level, as I don't have a great deal time to read into and understand it all with the work that I do in the role that I'm in along with family life outside of work, so having some form of AI is kind of like a dream in improving what are often mundane tasks or providing exact instructions required the moment you need it (as you say, we think of it like magic), but when you bring your points across, along with your knowledge and expertise of the industry, I can really understand and see what you mean.

It makes sense when you say AI is often dishonest marketing (except as you say for certain situations such as image recognition, text recognition, diagnosis etc.) In fact, I feel like fool as I really should've known that, as I've already in fact have seen some of these improvements to the search functionality in all parts of Hornbill and existing implementation of the knowledgebase that pops up when logging requests... and nowhere have you called it AI... so, it just goes to show how much education is required to help people like myself to understand, even at a basic level of explanation like you did in your post, which you didn't have to do, but you did.

I thank thee for furthering my understanding on this topic, and for your time in providing a detailed explanation.

I am very much looking forward to the improvements you and the rest of Hornbill are bringing to the table, keep up the excellent work!

[Gerry]

@samwoo

"I am very much looking forward to the improvements you and the rest of Hornbill are bringing to the table, keep up the excellent work!" thank you, I am very happy to be called back to this subject if what we deliver is not up to the promises I have made above. We have a lot to do but we also have a lot being worked on, so there should be q few incremental improvements made available along the way too. 

I will probably turn the above diatribe into a blog article at some point

Gerry