Show me Your Data & I’ll Show You Your AI

You can slap AI onto SharePoint, but what you’ll get is the equivalent of a toddler meltdown – Chaotic, messy illogical and irrational.

AI is at the tip of everyone’s tongue today. Chatbots here, chatbots there, chatbots everywhere!

Listen, I’m an AI Engineer, I LOVE AI. However, I beg of you, STOP slapping chatbots on disorganized data, and trying to train machine learning models with breadcrumbs.

The quality of any AI is 95% dependent on the quality of its data. What makes excellent models are:

-????????? Smart Cloud Architecture

-??????? ‘Just Right’ Models

-????????? AND BEAUTIFUL DATA

2025 and beyond will be flooded with mediocre quality chatbots, hallucinating AIs, and off-track machine learning models as everyone and their grandmother starts feeding AI random data sets. But you, my friend, will not be one of them.

You are going to learn the power of gorgeous data and help your organization build smart, elegant, and productive models that change the game.

The Toddler vs. The Elegant Lady: Understanding Data Quality

AI is not ‘smart’ by default. It is only as good as the data it is fed. Thus, if you feed it with raw, unstructured, mislabeled, or irrelevant data, it’ll respond in kind with erratic, irrelevant and low-value outputs.

A great and yet frequent example of this is the Copilot agent slapped onto SharePoint. Although at its base, the idea is good, if your organization hasn’t first painstakingly reorganized your SharePoint data, labeled it appropriately vs relying on automatic indexing, it will simply be an overexcited Chiwawa that shoots out random information.

Here's an example:

Prompt: What is our work from home policy?        

The SharePoint Toddler Chiwawa mix: ‘Oooooh here’s a document about HR something I 2017, and here’s a teams conversation between Deborah and Jack about the benefits of working from home, and here’s an email about Jessica saying she would work from home because she was sick 786 days ago’…

Vs.

The Elegant Lady built on elegant data: ‘According to the latest version of the employee handbook v.10 (2025) that can be found here (Insert link), full-time employee’s are allowed to work from home up to 3 days a week without requiring permission from management on any weekday but Mondays due to team standups. If you require permission to work from home more than 3 days per week, you may use this link (link) to submit a request to your manager.’

In our first example, our AI must rely on 20 years of random SharePoint files automatically indexed while the second has access to a beautifully structured SharePoint with a purposeful labeling methodology.

?? Lesson: AI isn’t failing you. Your data is.

What Makes AI "Intelligent"?

AI doesn’t think—it recognizes patterns and makes predictions based on what it has learned from the data it’s been given.

For AI to function like an elegant, well-informed expert, it needs:

1. Clean Data – Free from duplicates, inconsistencies, and outdated information.
2. Structured Data – Organized with labels, categories, and metadata.
3. Relevant Data – Trained on domain-specific and up-to-date knowledge.

If you throw a new employee a bunch of random internal documents and industry jargon expecting him to just ‘figure it out’ you’ll end up with less than desirable results. The same is true of AI.

?

Training AI the Right Way: From Toddler to Expert

?? How to Structure Data for AI Success:

? Metadata & Labels – Ensure every document is tagged with its purpose, status, and relevance.

? Version Control – Prevent AI from retrieving outdated or conflicting data.

? Knowledge Graphs – Connect related data points so AI understands context.

? Continuous Cleaning – Regularly audit and remove irrelevant or duplicate information.

? Make data organization, labeling & connecting the new norm for your organization.

?

Data Perfection: How to Optimize Your Data for AI

Think of it as crafting the perfect diet for your AI—no junk food allowed. Here’s how to get your data in top shape:

A. Define Your Model’s Purpose

• What are you trying to achieve? Predict customer churn? Automate document processing? Generate insights?
• Your AI’s structure should align with a specific outcome, just like an athlete train’s for a specific sport.

B. Structure, Clean, and Normalize Your Data

• Remove noise: Strip irrelevant data, duplicates, and inconsistencies.
• Standardize formats: Ensure uniform data formats (e.g., date/time formats, structured metadata).
• Fix missing data: Use interpolation, backfilling, or other imputation techniques.

C. Label and Categorize Your Data Properly

• AI thrives on context.
• If you feed it thousands of documents with no categorization, it will struggle to extract meaning.
• Example: Instead of dumping raw PDFs, structure them with metadata, labels, and proper taxonomies. You can, and should, create SOPs around this. I usually am not a big lover of documentation but in this context… it’s downright sexy.

D. Balance and Bias Check

• AI learns from patterns, but it can also inherit biases.
• If your data is skewed (e.g., only training a model on urban healthcare data but applying it to rural settings), it will fail in the real world.
• How to fix it: Balance datasets with diverse, representative samples. Make sure that your model understands when to reference one data set vs another, for example reference dataset A when listing heart attack symptoms for women, vs dataset B for those in men. Or in business, procedures for on-site vs remote staff.

?

Keep It Fresh: Data Must Evolve

• AI models degrade over time if they aren’t trained on new, relevant data.
• Set up data pipelines to ensure a continuous flow of clean, updated information.
• Ideally someone on your team should be tasked with reviewing, retraining, and correcting your models weekly.

Author William Golding once said

“I think women are foolish to pretend they are equal to men. They are far superior and always have been. Whatever you give a woman, she will make greater. If you give her sperm, she will give you a baby. If you give her a house, she will give you a home. If you give her groceries, she will give you a meal. If you give her a smile she will give you, her heart. She multiplies and enlarges what is given to her.”

Isn’t that beautiful? Well, he also said this at the end:

“So, if you give her any crap, be ready to receive a ton of shit!"

AI is an elegant lady… she will amplify whatever you give her. So, if you feed her with garbage, she’ll spurt out garbage faster than you can sift through it.

?