To say that businesses are embracing AI at breakneck speed, with so many organisations scrambling to get it embedded into their operations, is probably an understatement. We all appear to be diving head-first into the AI pool.
Technology research leader Gartner notes AI enters organisations in two ways: as a defined initiative, or as an embedded capability. AI-ready data reflects initiative-driven use cases, while data teams increasingly use AI directly in their daily work.
Gartner says this distinction matters because data teams are often trying to apply AI-ready data guidance in situations where it does not fully apply, creating confusion about when that guidance is relevant.
Poor data input into AI, equals poor results, or worse business-damaging outcomes.
What is often missed though, is that the old adage, rubbish-in, rubbish-out, applies very strongly to AI. Poor data quality leads to flawed insights and biased decisions.
Businesses must examine the quality of their data prior to scaling AI projects. In a nutshell, this rule means an AI model gives bad results if it gets bad data or poor prompts. High-quality training and clear instructions are required because machines simply process what they receive.
To put it succinctly, in the context of AI, the quality of output is determined by the quality of the input.
Poor data quality leads to flawed insights and biased decisions.
Many businesses haven't taken the time to even evaluate their data and clean it up. These companies will not get the maximum yield from AI.
All too often, I encounter businesses complaining that the AI tools are wrong, or don't give them what they want. However, it important to understand that the tool is not the issue − it's the data below it.
AI tools, including machine learning models, are only as good as the data and instructions they are provided with. Worse still, in many instances AI adds an additional layer of work for the human instead of removing it.
That is not to say that AI cannot significantly improve many aspects of, for example, customer service − it can. What it cannot do is to independently know which data files need to be fed into the system for accurate analysis. That requires human intervention.
On the ground in the managed services industry
I can cite many incidents in the daily run of a managed services business and liaison with clients. I illustrate one such example as follows: a client logs a ticket for some basic support like installing software. We use AI for these repetitive tasks.
The client's ticket comes in, the AI verifies the identity of the client, and simply put, AI goes out and finds their specific laptop within the management platform. It pinpoints the machine connected to the user profile and then proceeds to install the software on that specific machine.
However, this is sometimes followed by the logging of another ticket from a different user, alarmed that some new software has appeared on their machine.
As soon as the latter happens, a tech, of the human variety, investigates and discovers that four months ago the client bought a new laptop but we didn't update their user profile to reflect the new machine, because we didn't know about it. Also, we didn't update that the old laptop has been passed to a newly employed intern to use − because we didn't know about that either.
The system, of course, still lists the original user of the old laptop and that is the machine we installed the software on. As a result of the data being so inconsistent − inaccurate is possibly a better word − we now have two devices that don't have the correct software on and only human investigation can resolve the issue.
If the data had been correct in the first place, AI would have sorted this request out within minutes. Instead, we have the situation of work down-time and inconvenience for two users, plus an hour of human intervention has been necessary to solve the problem.
So, what should businesses do before they plunge into AI?
The answer to that is simple: sort out the data. Clean it up by removing duplicates, errors and old files before training or feeding files to an AI model. Provide context, instruct the AI on its role, target audience and the style required.
Finish this off by testing and refining the AI model by conversing with it, have a back-and-forth conversation to correct mistakes and guide the output.
Only in this way will you get the most out of the AI revolution.
Share
