GIGO - Good In, Good Out

[#1]

The more I use systems (especially generative AI ones like Rovo!), the more I am reminded that GIGO is alive and well.

What now?

When I first heard the term “GIGO” it stood for “Garbage-In, Garbage-Out”. Basically, this concept means that if you give a system a bad input, you’ll get a bad output. This concept applies to every system/process/interaction - including ones with other living humans (imagine having to assemble furniture but you didn’t have instructions….). This expresses itself in many different ways, including the inability to locate information, the system “not working” and general frustration.

More recently, I’ve heard a corollary to this - “Good in, Good Out”. This is, essentially, the same idea - when you put something “good” in, you get something “good” out. I’m working on adjusting my habit to use this definition since it’s more positive (and I think we can all, always use more positivity!).

How it impacts Confluence

GIGO impacts EVERY system, and Confluence is absolutely no exception (also looking at you Jira!). Most commonly, I find this pop up in search - someone types in a keyword or phrase and doesn’t find what they’re looking for. This doesn’t mean that the search function is broken, it just means that whatever you’re looking for isn’t set up in a way to find it with your search. In this example, the “garbage” or “good” input could be any combination of:

1. The underlying content - Maybe its title isn’t accurate, or maybe it isn’t labelled properly. Regardless, the search algorithm can’t find it for some reason.
2. Your search - It’s also possible whatever you typed in could use some improvement (e.g. your prompting could be improved).
3. The content simply doesn’t exist - This can be less common; however, it’s also possible the thing you’re looking for simply doesn’t exist.

It’s also basically impossible to know which one of those reasons (or which combination) is what is resulting in your struggle. This makes it… frustrating… at best!

Regardless of why it’s happening, it quickly results in folks complaining that “Confluence is broken”. At best, this frustrates folks since they can’t get what they want; however, it frequently reduces trust in Confluence as a system. This impacts future interactions and easily results in a downward spiral.

What you can do about it

GIGO is resolved by improving either the underlying content or the search terms (prompts) used. This is much easier said than done, since typically you don’t fully control every aspect of the system. For example, in many cases you’re looking for a piece of content that you likely didn’t create. Also, since you’re searching for it you likely can’t find it… so you couldn’t improve it even if you wanted to. Due to this, we’ll look at the solution from two perspectives - personal and organizational.

Personal Solutions

Individuals can help solve the GIGO problem in two ways:

1. Ensuring content YOU create, update, or manage, has clear titles, labels, and other features to improve fundability (more on these below).
2. Thinking through the searches, prompts, and other inputs you use to locate information. This can take the form of trying different keywords, asking folks for advice/help, and iterating on prompts.

Organization Solutions

This one is trickier since it requires people across the org to do things. This mainly boils down to things like:

1. Education. Ensure individuals know how search/prompting is supposed to work. This helps reduce mis-matched expectations (“Well, it works this way in some other system, so it should do the same here”).
2. Standardization. Set standards for things like content titles, labels, keywords, and the like.
3. Active monitoring/maintenance. Get a group of folks to regularly review content to ensure it meets your standards.
4. Feedback. Have clear mechanisms for people to provide feedback on what may, or may not, be working. This gives people who are frustrated an “out” and somewhere to go for help (instead of suffering in silence).

Standardization Guidelines

Every organization will make its own decisions on how to standardize content, but here are some ideas to get you started or add to what you do.

Keywords - Think up a set of keywords that people can expect to be on specific types of content. For example, the word “Project” could be used to always indicate the content is project-related, or the name of the team who works on the content. These keywords could then be used in places like the summary, the content itself, and as labels to help improve search ability.  These keywords should be shared so people know what they are so they can use them.

Maintenance timelines - Set clear expectations on how regularly content will be reviewed. It’s unlikely you’ll have enough time/effort/energy to review EVERYTHING, so instead identify the most critical content (Human Resources / People, Legal, etc.) and review the first. Then, as time allows, review other content in descending order of importance. Encourage teams and individuals to do the same for their content.

Education/Training - Ensure individuals have access to information/training/videos/etc. on your standards, but also how the system operates. This will help reduce frustration and set expectations. Ideally, include this in onboarding so people are exposed to it as they join the company.

Wrap up

GIGO is a simple, but important, concept to keep in mind as we interact with systems… and this is only becoming more apparent as more and more generative and agenic AI is appearing. There are, however, ways to help keep it in check and make things as smooth as possible for everyone. Curious about your thoughts though - what do YOU do to combat GIGO? What works or doesn’t work? Let us know below!

[Stephen_Lugton]

Thanks for your thoughts @Rob Hean , I've got to find a topic for a 'brown bag' session next week, and since we're being pushed to use AI more this is definitely relevant today, so I now have an idea for a topic I can cover in that!

[Rob Hean]

Glad I've helped inspire you @Stephen_Lugton ! I think this topic will become even more prevalent as more AI is used... I've lost count of the number of times people have asked me how to improve Rovo results, or asked "why does it give bad answers"...