It’s hard to attend a meeting today without AI taking centre stage. Everybody is talking about what’s next for AI, what benefits it will provide, which tools it is going to be available within, and how it is going to make your life easier, or even worse, take your job!!

In the corporate world, AI is now becoming so widely used to help generate ideas, develop draft documents, provide strategies and suggested next steps. Many software suppliers are even including it within their software to help users gain deeper insights from the data stored within these solutions, i.e., “provide me a summary of how we have responded to this type of customer question in the past”, or “what is our latest policy on bereavement leave”.

The technology has become very advanced, and it is almost like having your own personal secretary to help you with your daily tasks. But… what happens when you have large amounts of records and data in your systems? Are you really getting the accurate results that you thought you were? Is AI working with you or against you?

I live in sunny Queensland, and in my backyard is an extremely large and almost oversized mango tree. This summer was a crazy mango season for our tree, and we had more mangoes that we could possibly handle.

We ate as many of them as we could. We gave away more kilograms of them then I can count. It reached a point where we had to get creative to ensure no mango was wasted. With a little help from Google, we made sorbet, chutneys, jams, cakes… if you can think of a way to cook or bake a mango, I guarantee we tried it.

It made me think how much this tree relates to what we hold in our corporate systems and what we are trying to achieve with the “magic” of AI without any real structure, plans or proper management.

There was so much data (fruit in this case) - if I had used our standard AI approach and uploaded a photo to AI and asked for help, it would have provided me some good results (recipes, ideas etc.), but it would have also provided some terrible results too.

At a glance, you can see all the heathy mangoes on the tree, but the data would also show you that there were also many mangoes on the ground. However, it would not be smart enough or have the human insight to know that they are not usable, they are bruised, smashed or half eaten by the possums.

It would tell me that I had excess mangoes on the ground and what I could make from them, it may also have discovered the dog ball in the mix and thought it was an orange and start to give me information about use of oranges as well.

AI without governance: a fast track to risk

Many organisations have made in AI available without much thought, because it was touted as the future and something everyone should embrace. However, many of these organisations also saw the instantaneous downside of AI.

Within weeks of trials, they discovered that AI could find EVERYTHING that they didn’t realise was even still there, or that didn’t have appropriate security permissions.

Adding AI or even enterprise searching to your Microsoft worlds, Google spaces, and your records and information management systems is a great idea, and a great way to improve decision making; however simply adding them on top of a system with no pre-planning will provide your organisation with unfavourable results.

As records and information management specialists, I highly encourage you to utilise these technologies but be prepared to take a seat at the table to define how they are used, what your data looks like, how to protect the data and how to get the most out of the AI / searching tools.

My key tips

Here are some of my key tips as records and information professionals of how to prepare your systems for this technology.

• Retention and Disposal - Use your retention and disposal tools and correctly - dispose of records and information when they are due for disposition. The old saying used to be, storage is cheap and we can always buy more storage. However, having all that information within your AI’s reach will provide you with a lot of out of date, incorrect or even harmful results. There have been many cases in the press about AI agents providing back results to the public or to staff on out-of-date procedures, policies etc. The last thing you want is the AI engine telling someone that the procedure from 1991 contains the most useful information on seatbelts…

In a recent conversation about AI and removing old information, there was a comment of ‘You have to clean your house before you can have a party’, which can easily be translated for how we prepare our corporate data for the AI and enterprise search party!

• Reduce your data pool – Just because you have 40 million records stored within your repositories, that doesn’t mean you will get better results out of AI. In fact, the more data you point the AI at, the more skewed your results will become, and the longer it will take to provide you the correct information. Often, you only require results relevant to the last 2-3 years of information or pertaining to particular topics. By reducing the data, you allow the AI or search engine to index smaller parameters that will provide you with much more accurate results.
• Storing high value records in context – Appropriately storing records within structures and in context is one of the most effective way to reduce AI hallucinations, whilst also improving your overall records management. Applying additional metadata against records provides greater context to the AI solution so that it can better understand the meaning / purpose of the information instead of just words on a page. I once heard someone say “Context turns data into information, where that information can then be turned into knowledge"
• Security frameworks – Open security is a very powerful and popular security framework and is certainly better than locking everything down to yourself so that no one can ever find it. However, when you are opening / allowing access to AI or other search engines with much deeper reach than human searching, you need to ensure that there are appropriate security measures in place to protect records that require privileged access. Perhaps the better principle is ‘open by default but secured if required’.
• Be very specific in your requests – While you would assume from the title of the solution, i.e. Artificial Intelligence, it sounds like is should know everything, in reality it is very much like a toddler that needs to be provided with step-by-step details and instructions. There was a video going around a few years back where the father asked their child to write instructions on how to make peanut butter toast, the father purposely took the instructions as literal as he could to prove a point, i.e., if the instructions said place the peanut butter on the bread, he quite literally placed the peanut butter jar on the bread and served it back to his child, and repeated similar activities until the child learnt to be very specific in how to make the sandwich. Much like this, AI works best where you provide detailed prompts and very granular instructions on how it should search, what persona it should be searching under, what are the parameters, and in what format you would like the results.

If you can take the time to properly define and setup the rules of how your organisation utilises AI, and you work to ensure your data is appropriately managed, then both you and your organisation will be able to enable AI to work for you and not against you.

With over 20 years of experience in the documents and records management industry, iCognition’s Nicholas Fripp brings deep expertise in eDRMS solutions — including Micro Focus Content Manager — along with a strong track record spanning customer-side and consulting roles across diverse industries.