Overcoming AI Roadblocks by Fixing the Data Foundations

This article is by George Tziahanas, VP of Compliance at Archive360

The insurance sector has a major opportunity in front of it in the form of artificial intelligence (AI). As AI becomes ever more accessible and finely tuned, it offers a host of benefits to companies across the industry – improving the accuracy of risk forecasting, reducing repetitive, low-value workload on skilled staff, speeding up collaboration, and improving compliance processes.

But those benefits don’t just arrive automatically. AI applications and systems need to be carefully implemented, and businesses often hit serious challenges along the way, keeping them from achieving successful deployments. Right now, poor quality data is one of the biggest such roadblock to AI success in the insurance industry. In fact, a whopping 72% of insurers say it’s holding them back, leaving underwriters unable to accelerate AI adoption and missing out on efficiency gains.

In truth, that figure isn’t really a surprise. Despite the revolutionary ways it’s predicted to change our work, AI is only ever as strong as the data behind it. Poor data leads to poor results. When insurers feed AI models with incomplete, outdated, or inaccurate information, it’s like running an engine on the wrong type of fuel: the shiny, expensive machine is going to suddenly fail, leaving you with a serious headache to tackle.

As a result, insurance companies need to pay as much attention to their data foundations as they do to their front-end AI applications. While it’s certainly a significant piece of work, it doesn’t have to be an insurmountable problem. Rich, real-time data from a modern archive can transform AI from unreliable to invaluable.

Data quality debt

It’s worth exploring in more detail why this ‘data quality debt’ is a major risk in AI development in insurance. In one sense, insurance companies are uniquely placed to benefit from AI’s information-processing abilities. AI’s great strength is in discerning correlations in truly vast datasets and then articulating those correlations in usable ways – whether that’s a GenAI image or, in this case, a finely-detailed risk profile.

In other words, by applying AI’s raw power to the huge amount of risk information an insurer possesses, it becomes possible to create highly reliable, highly customised profiles that can then feed into policies, pricing, and organisational strategy.

But the flipside is also true. If AI is using poor quality data – whether it’s out-of-date, incomplete, or simply incorrect – the results could be catastrophic, both in terms of regular business outcomes and regulatory penalties. Skewed or biased AI decision-making can lead to damaging inaccuracies. And AI usage is subject to increasingly stringent regulatory oversight, particularly in regulated industries like insurance. Insurers need transparency in how an AI model has produced the recommendations it has, and assure the regulator that the process uses accurate, appropriate data that’s been stored securely and processed legally. Poorly managed data sources heighten the risk of failing these controls and making the company subject to expensive fines.

Five steps to a stronger data foundation

As a result, it’s highly important that insurers can feed their AI systems compliant, current, and contextually relevant data. To achieve that goal, there are a number of practical steps companies can take, tackling the data challenge with modern data archives and so supporting AI adoption.

There are five key areas that need to be addressed in the creation of a reliable data foundation. Insurers need to make sure their systems are rock solid in each of these areas, deploying high-capacity, intelligent archiving that doesn’t just use a ‘store it and forget’ approach, but treats data as a living, valuable asset.

Firstly, to strengthen the foundations of their AI systems, insurers need to build complete visibility and control over their data. That begins with understanding where information comes from and how it changes over time. Every dataset has a lineage: its origin, ownership, and transformation history. Keeping a clear record of this chain gives companies confidence in the reliability of what they are feeding into AI. Maintaining detailed metadata and preserving the original source documents ensures nothing is lost or misrepresented along the way.

Authenticity is equally critical. Data must remain exactly as it was first captured, with an unbroken chain of custody to prove it. Storing objects in their native formats, supported by hashing and a full audit trail, helps confirm that no alteration has taken place. This is especially important when regulators or auditors ask for evidence showing that a decision-making process was built on genuine, unmodified data.

Once authenticity is protected, the next task is to classify data correctly. Insurers deal with vast volumes of structured, semi-structured, and unstructured information, from policy databases to call recordings and scanned documents. Each type requires a different approach to storage and governance. Systems that can apply a relevant schema to each class of data allow insurers to manage diversity without forcing everything into a single rigid model.

Consistency then becomes the goal. Normalising data (applying common definitions, formats, and metadata) makes analytics and AI outputs more dependable. Without that standardisation, two systems analysing the same information could produce conflicting results. Tools that can transform or map datasets into a unified view help create the stability that AI depends on.

Finally, access must be tightly controlled. Every piece of data should have clear entitlements defining who can see it, and under what circumstances. This can go down to individual objects or fields when necessary. For example, a claims adjuster might need access to customer history, but not to sensitive financial identifiers. By assigning access in this way, insurers can protect confidentiality while still giving authorised staff and systems what they need to work effectively.

Together, these practices turn data from a passive record into a living, reliable asset. When insurers invest in this kind of intelligent archiving and control, they give their AI projects the solid ground they need to deliver genuine results.

Looking ahead

This fivefold approach will help insurers build a data foundation they can rely on, ensuring the right information is fed to the right system at the right time, and excluding potentially dangerous inaccuracies. With a strong data archive in place, AI projects will stand the best chance of succeeding, allowing insurers to seize the opportunities before them.