Garbage In, Gospel Out

Human being are already vulnerable to believing data before we question it. AI adds the wrinkle of being absolutely sure of its answers, confidence The team had fallen into a familiar trap. They forgot that an AI presents its answers with such confidence that the ‘garbage’ is mistaken for ‘gospel.’ This single mistake undermined trust and gave skeptics the ammo they needed to walk away from similar solutions.

A new data leader saw this and took a different approach. She knew she could never get budget for a six-month data governance project as leadership would see that as pure cost.

Instead, she went to the Head of Sales with a different pitch. She said, ‘I can build a tool that answers one question with 100% accuracy: ‘Which of your customers are actually eligible for an upsell right now?’‘

The sales leader, burned by the last tool, was doubtful but intrigued. To answer that one question, the data leader’s team had to get buy-in. They had to pull and govern only the specific data required: support ticket history (from Zendesk), contract terms (from Salesforce) and product usage (from Pendo).

They built a ‘single source of truth’ for that one question. Two weeks later, the ‘Upsell Agent’ went live. It didn’t do 50 things. It did one. And it was never wrong. The sales team learned to trust it and it generated new revenue. The data leader had successfully funded the ‘boring’ foundational work by tying it to a high-value, high-precision outcome.

A Framework for Funding the Foundation

This approach is how you get leadership to invest in data prep.

1. Stop Pitching ‘Data Cleaning’ Few executives will sign off on a data governance initiative unless it’s attached to a material gain. Pitch a ‘Precision Upsell Engine’ or a ‘Churn Risk Detector’ instead. The boring work of cleaning, classifying, and governing data becomes the necessary, funded plumbing required to build the high-value asset.

2. Pick One High-Value Question Get your department heads in a room and find the one question they all need answered but can’t trust the data on. ‘Which customers are a churn risk?’ ‘Who are our most profitable customers?’ ‘Which product features drive renewals?’ This single, high-stakes question becomes your pilot.

3. Define the ‘Source of Truth’ Map out the minimum viable dataset needed to answer your one question. This forces a cross-functional conversation. To know ‘churn risk,’ you need data from Support (Are they filing tickets?), Product (Are they logging in?), and Finance (Are they paying on time?). This small, governed ‘slice’ of data becomes your first trusted source.

4. Build Your First ‘Precision Agent’ Build your AI tool on only that clean, governed, trusted slice of data. When it works flawlessly, it does two things: it delivers immediate business value and it becomes the unmissable proof that this ‘boring’ foundational work is the only path to trustworthy AI.

Action: Find the one dashboard or report in your company that everyone ‘knows is wrong’ but still uses. Find out why it’s wrong (e.g., it pulls from two different systems, it’s not up-to-date). The business case for fixing that single, high-visibility report is your budget for your first data-governance pilot.

For the research behind this, see how data quality is shaping UK AI infrastructure decisions.

Takeaway

Your data may be a disaster, but leadership rarely pays to ‘clean the house.’ They will pay you to build a high-value tool that can’t be built without cleaning one room first. The potential of AI may be the tipping point for finally doing the foundational data work you’ve ignored for a decade.