My learnings in Knowledge management and AI

This blog post was written by Mattia a 6-month placement student, and Mael (Ansearch’s CEO and editor of this blog post) helped clear out the technical details.

1. Introduction

During my six-month placement as a marketing student at Ansearch, I learned about Knowledge Management (KM) and how organisations handle information.

This blog post will explore the importance of KM in creating, storing, and sharing information within companies, the challenges of keeping knowledge accessible and the tools to overcome these challenges. I'll also discuss Artificial Intelligence and large language models, their limitations and how to incorporate them into knowledge management.

2. What is Knowledge Management?

First, a few definitions:

Knowledge represents all the company's information (it can be know-how, emails, notes, etc). This information can be stored in tools (say a wiki) or in people’s minds (collective intelligence).

Knowledge management represents the various processes around handling information (creation, communication, organisation, usage, and maintenance).

Employees require this knowledge to carry on their work. So the goal of knowledge management is to ensure that knowledge is stored in the right place, easily accessible, and reachable only by the right people.

Good knowledge management practices help employees to be effective and efficient at their jobs. On the opposite, bad (or no) knowledge management practices will impede work, create duplicated work, and even lead to costly mistakes.

Two indicators of bad knowledge management are an overreliance on meetings and repeated questions (through voice, email or chat).

Various tools can help establish good practices or enhance the employee's experience around knowledge management. Some examples of those tools are enterprise search, intranets, document storage, and wikis/handbooks.

3. What are data silos, and how do they affect knowledge management?

Data silos happen when information becomes inaccessible to various departments or teams within a company, generally due to ineffective communication among people, teams, or applications. Many tools created today are ultra-specialised (e.g. CRMs) but not made to effectively sync and share information with other tools. Additionally, the IT department will make it hard or impossible for people to get an account for all the various tools in a company (for a good reason).

This leads to a lot of data being inaccessible (you are not in the right team) or hard to access (you are in the right team but don't know where the information is), lacking synchronisation. In the worst-case scenario, you also have different departments using the same tools but with other versions (e.g., one team using HubSpot, another using Salesforce, and a third using a different version of HubSpot).

Synchronising this data can be challenging: current tools are trying to integrate with each other a little bit more, or you can create glue code (scripts specifically made to sync data between multiple tools) but often this requires a lot of time to create the automation, or developers to build this glue code.

4. What are the limitations of Traditional Knowledge Management?

Traditionally, knowledge management has been difficult to set up and overlooked (not enough time, not enough resources, “no need”, etc). This was mainly because setting good practices takes time (knowing your tools, knowing how to organise them, creating processes, communicating with your team on those processes, etc), specialised tools (wikis, enterprise search, often custom-made or self-hosted), and sometimes dedicated teams (knowledge people/librarians), which put together represent a high entry budget.

To add to the entry budget, many knowledge management tools are sold as is, instead of being software as a service (or worse, custom-built/built in-house), and require servers and IT departments to install and maintain them.

The result is that SMEs and new teams skipped knowledge management entirely, and the issue was tackled only by companies that must have good knowledge management (such as law firms, accountants, consultants, and massive organisations).

5. How can Knowledge Management improve collaboration across departments?

Knowledge management establishes guidelines on where to store information, and how to do so. Since every individual has their own way of thinking and organising themselves, creating those guidelines reduces chaos and creates a more unified way of documenting.

This is where it becomes important: sticking to those guidelines (even if this is not the “best” way to document) ensures other people can trust where the information is. In turn, trusting where this information is allows access to knowledge easily. Being able to access knowledge easily reduces the time taken to do a task, interruptions when asking colleagues, and occurrences of duplicated work. Even better, it reduces knowledge lost when someone leaves the company and prevents costly mistakes.

For example, imagine a company's sales department using a CRM (Customer Relationship Management system) to record every interaction between a salesperson and a customer. Good knowledge practice is to make clear to the whole department that everyone must record every meeting, email and note, and save this in the CRM, and files that cannot be saved in this CRM are stored in a shared online drive with one folder per company and a specific naming scheme. If a salesperson leaves the team or is unavailable, the rest of the team can safely take over the customer: the customer information is easily reachable and up to date.

6. What is the difference between structured data and unstructured data

The best way to visualise what is structured data is to look at a form. Structured data refers to data that is labelled and standardised. Those documents have “metadata”, which provides information about the document content and structure (for example a customer always has an email). It means that no matter how many data points (or documents) you have, they will all have the same fields (for example: every customer will have a different email, but you are sure the email will always be present).

On the other hand, unstructured data is raw and lacks standardisation. It is often uncategorised and does not contain tags or metadata, making it challenging to analyse using traditional computing methods. For example, the text of a book online, with no indication of how the chapters are identified.

Humans are very good with unstructured data (just by looking at a book, you can locate the title, the summary, and the editor by looking at the letters' size, location and if it is bold or not). Computers are not good with unstructured data (if you don’t specifically tell the computer where the title is, it will not be able to locate it). However, humans are slow to gather and use information, while computers can process millions of records in less than a second.

This is why the field of AI has been getting more and more traction in the last 10 years: AI enables computers to process unstructured data.

7.  What is AI?

If you are not in tech, a quick thing to know: AI has existed since the mid-90s and started to gain popularity with techies around 2015.

AI, or more accurately, machine learning/deep learning, involves feeding vast amounts of data into a mathematical model to enable it to recognise patterns. For instance, in image recognition, millions of images are presented to a model with their associated label (ex: this image has a dog). Over many trainings, machine learning algorithms refine their capabilities, and the model will start recognising patterns of pixels (ex: many brown pixels next to black pixels is usually a dog). Once the model has learned those patterns, it can be used with unknown images to "predict" what is in the image.

As we saw in the previous section, AI enables computers to process unstructured data: by looking at patterns AIs can categorise and sort data, for further use.

8. Limitations of AI

AI implementation is complex and demanding, leading many projects to fail due to inappropriate tools, data or scoping. Often, tried and tested “non-AI” solutions already exist, are simpler to implement, and can be a good starting point rather than going straight to AI solutions.

To illustrate, a gardener needs to plant a flower. Using an excavator to plant a single flower in a garden is excessive, a simple shovel would be easier and faster to use, faster to deploy and much cheaper.

Now this same gardener has more customers and needs to do a large botanical garden. By learning how to use a shovel, you also learned how to solve the issue at scale with the right team and the right machinery (a plough is more adapted than an excavator).

In other words, AI can be exceptionally effective, but you need a plan of how to use it, where, and how you will build it.

9. What are LLMs?

Artificial intelligence, particularly in the field of Natural Language Processing (NLP), has seen significant progress with the development of Large Language Models (LLMs). 

These models are trained to comprehend and analyse human language allowing them to either understand complex queries (intelligent agents) or simulate reasoning (the chats).

Don't get fooled, LLMs are no smarter than your regular software. They are just really good with language patterns.

Companies are just starting to adopt LLMs, it is still unclear exactly how much potential they have, how to use them effectively, and they come with a high price. But everyone agrees they have significant potential for disruption, close to the early age of the internet or smartphones.

9. How can AI be implemented into Knowledge Management?

AI can be implemented into knowledge management in many ways, bridging gaps and addressing challenges faced by traditional methods, such as handling unstructured data and reducing strain on existing knowledge teams:

• During creation, by helping to create documentation.

• During research, by helping search for information and digest it for the user

• During maintenance, by finding duplicates and stale information

This list is exhaustive: LLMs are a new technology, and their potential is still unexplored, meaning more solutions and applications are yet to be created.

11.  Are LLMs secure?

The security of LLMs in terms of data breaches depends on several factors.

Firstly, fine-tuning an LLM is often used to make an LLM excel in a specific field. However, the model will learn all the data it is presented with. This causes a big issue when said data is sensitive: once learned, it will be saved in the “memory” of the LLM. This model will then answer anyone with this knowledge. Any individual with access to the LLM can ask for and get access to this sensitive data since the model cannot distinguish whether the user should have access to certain data leading to unauthorised disclosures. Worse, this creates a red carpet for extracting sensitive data (ex: executive-only information used during training can be accessed by a simple placement student who is a prime target for phishing).

However, a general LLM or a LLM fine-tuned without sensitive data will not have those issues.

But this is only looking at the issue from one side, the other side is who is hosting the service, and how they secure it. To create relevant text with an LLM, you will need relevant data. And it is extremely important to secure access and storage of this data, just like any other application. Unfortunately, this side is often overlooked, and SaaS products have basic, lacklustre, security practices (not at Ansearch though!).

12.  How I used LLMs for knowledge management during my placement

During my marketing placement at Ansearch, I discovered that LLMs can significantly improve Knowledge Management processes. Here are some practical examples of how I used LLMs related to Knowledge Management:

• When I had to deal with lengthy articles, research papers, and reports, I discovered that LLMs offered a solution for extracting essential information. By using LLMs, I could efficiently understand key insights without spending hours reading extensive texts. This feature can also be applied to Knowledge Management for internal documents, saving time and improving productivity.

• LLMs also helped as a proofreading tool to identify grammar and syntax mistakes. This functionality can also be applied in Knowledge Management when communicating and creating content for stakeholders using company data.

• Finally, as a marketing student, I struggled to understand complex software-related concepts or industry jargon. LLMs helped me in comprehension, breaking down complex concepts into clearer explanations. This use of LLMs can also be useful in companies, particularly when onboarding new employees, enabling them to quickly find what they need, improve learning, and become more productive in less time.

13. Ansearch as an AI Knowledge Management Tool

At Ansearch, we are aiming to help businesses with their knowledge management. Our goal is to make your information high quality and available immediately when you need it, as well as make knowledge management easy to implement or improve.

We are starting with our core feature: a multi-application search engine (it’s in the name Answer + Search = Ansearch).

With only a few clicks, you can connect Ansearch to all the tools you are using for your day-to-day job (emails, calendars, CRMs, chats, etc) and make them searchable. This saves a lot of time, especially when you “know you have a document, but can’t recall where it is” and reduces repeated questions from colleagues.

This search is just the beginning, we are working hard on bringing new features to help enhance knowledge management at every step, with AI (and non-AI) features such as summarising your documents, asking questions against your knowledge, providing an overview of what your business knows, and helping you track duplicates.

14. Conclusion

In this blog post, I discussed what I learned about knowledge management in improving productivity within organisations, particularly focusing on the role of AI in this process. I talked about challenges in maintaining accessible knowledge, touching on the concepts of data silos, the difference between structured and unstructured data, and the limitations of traditional knowledge management systems. I then explored how AI, including LLMs, can be implemented to enhance knowledge management processes and improve search efforts, and I shared some practical examples of how I used LLMs related to Knowledge Management.

If you are curious, our product is free as long as we are in beta (just don’t mind the occasional bugs), you just have to click on sign up at the top of the page

You can also follow our adventures on Linkedin or reach out! (we love to hear from you)

Contact us today to see how Ansearch can help your team. We'd also love to hear your thoughts on knowledge management challenges – let's start a conversation!