The data quality excuse (and why it’s holding you back)

Here’s why the “perfect data” myth is more dangerous than messy data.

Insurance
Lloyds of London
AI
ML
Data Quality
Author
Affiliation
Published

18 June 2025

“Our data isn’t good enough for AI yet.”

I heard this phrase 23 times during our recent survey of Lloyd’s market leaders. It’s become the most sophisticated way to avoid making a decision about artificial intelligence - and it’s costing companies competitive advantage.

Here’s why the “perfect data” myth is more dangerous than messy data.

The Problem: Waiting for Data Perfection

Our survey found that insurance professionals consistently name “data quality” as their number one concern about AI implementation. On the surface, this sounds prudent. Responsible. Professional.

But here’s what’s really happening: companies are using data quality as a sophisticated excuse to delay decisions they’re uncomfortable making.

Meanwhile, their competitors are moving forward with imperfect data and learning faster.

The Insight: Humans Work with Messy Data Too

Here’s the insight that changed everything for one Lloyd’s managing agent we surveyed: “If human resource can be trained to cope with and adapt to those data vagaries and shortcomings, then so can AI – in time, probably better and faster.”

Think about how your best underwriters make decisions. Do they wait for perfect information? Of course not. They work with incomplete, inconsistent, sometimes contradictory data and still make profitable decisions.

The question isn’t whether your data is perfect. The question is whether AI can work with imperfect data better than your current processes.

The question isn’t whether your data is perfect. The question is whether AI can work with imperfect data better than your current processes.

The Action: The 80% Rule

Instead of asking “Is our data good enough for AI?” ask these three questions:

  1. What decisions are we making with current data quality? If humans can make profitable decisions with your existing data, AI probably can too.

  2. Where would 80% accuracy beat current performance? You don’t need perfection – you need improvement.

  3. What would we learn by trying? The fastest way to improve data quality is to start using it systematically.

What can you do today with 80% of the data you wish you had instead of waiting for perfection? What can you do with 60%?

The Impact: Learning While Competitors Wait

One survey respondent described using AI for classifying claims information and identifying trends in pricing and reserving work. Not because their data was perfect, but because even imperfect pattern recognition was better than manual processes.

Result? They’re not just getting operational benefits – they’re discovering data quality issues faster and fixing them systematically. Their data is actually getting better because they’re using AI, not despite using imperfect data.

Meanwhile, their competitors are still in “data preparation mode.”

The Bottom Line: Perfection is failure

Data will never be perfect. But waiting for perfection is a perfect way to fall behind.

What decision could you improve with 80% of the data you wish you had?


Methodology

This analysis is based on research conducted by the Management Decision Analytics and Insurance Consulting teams at Barnett Waddingham in partnership with the Lloyd’s Market Association. The full survey represents approximately 55% of Lloyd’s market capacity.

Download the full report here: CROs and actuaries: the transformative impact of AI and ML,