Missed the webinar? Watch it on demand.

Our "is your data ready for AI?" webinar explores how data quality can impact AI outcomes and the steps organisations can take to build a stronger, more trusted foundation for AI.

Whether you're already experimenting with AI or you're still working out where to begin, this session will help you understand what to look at first.

Watch the webinar on demand

Simply fill in your details and we'll send the recording straight to your inbox.

The Q&A time:

1. Can’t AI just monitor our data for us?

It's a fair question! But AI doesn't know what the good data looks like.

AI can play a role in identifying issues, but before you ask AI whether your data is good enough, you need to establish what “good” actually means for your organisation.

That could include accuracy, completeness, consistency and valid values. It also means understanding what individual attributes represent and which source should be considered authoritative when different systems contain conflicting information.

A monitoring tool like Binary View gives you a structured way of continually measuring those rules rather than relying solely on AI to interpret them. And we all know that, at present, AI can make mistakes whilst confidently stating it is correct!!

2. What system(s) does Binary View (Data Quality Tool by Binary10) work on?

Binary View isn't tied to one particular application or system. Binary View can work with data sources where Binary10 can obtain the data and bring that data into SQL Server, even converting from other types of data sources, e.g. Data Warehouses, Oracle dbms, as part of the implementation. It can also support data from multiple systems within one instance, including checks that compare information across systems.

That flexibility is particularly useful because many data quality problems don't exist within one system alone. You might have information about the same customer, student, employee or supplier held differently across several systems.

3. You talked about data governance and data policy. Aren’t these just documents we pay a fortune for that no one ever reads?

They can be, if governance stops at producing the document.

The point Steve and James made during the webinar was that policies need to be short, practical and actively used. Crucially, there should also be clear ownership of the data and responsibility for monitoring and resolving problems.

Good governance should help answer practical questions:

• What does good data look like?

• Who owns it?

• Who monitors it?

• Who takes action when something goes wrong?

• What happens when a new system or process introduces new data?

If those questions aren't being answered in practice, having a policy document by itself isn't going to solve the problem. One way to ensure the DQ Documents and Policies don't end up being ignored is to instigate regular DQ Working Group meetings to review progress.

4. Does Binary View cleanse the data too?

No. Binary View identifies data quality issues for cleansing, rather than automatically changing the underlying data.

Once you've identified an issue, the right response will depend on what caused it. It could require cleansing individual records, changing a process, applying a business rule or deciding that a particular issue doesn't need correcting. The results identified through Binary View can also be exported to support remediation activity.

The important thing is being able to see where the problems are, how significant they are and what needs attention.

Binary View allows you to categorise data quality issues, so that you can focus in on the issues that definitely require upfront data cleansing, and avoid the noise from "nice-to-have" data cleanse fixes.

5. What’s the biggest benefit from having data constantly AI ready?

Confidence in the data your AI is relying on. Getting your data into good shape once doesn't mean it will stay that way.

During the webinar, Steve explained that process changes, staff training issues, new systems and new workflows can continue introducing data quality problems. That's why continuous monitoring becomes increasingly important when AI is regularly consuming organisational data.

Rather than discovering a problem after it has affected an AI output, report or decision, ongoing monitoring gives you visibility of how your underlying data quality is changing, allowing you to make data-driven decisions you can trust.

6. This all sounds great but a lot of work. In practice, does this just add huge workload to companies regularly monitoring data quality?

There is some work upfront. There's no magic button that suddenly makes organisational data perfect.

You need to decide what matters, define what good looks like, establish the relevant checks and make sure somebody owns the process. Steve acknowledged that upfront investment during the webinar.

But continuous monitoring is intended to shift the organisation away from repeatedly discovering and fixing the same problems. If the same data issue keeps appearing, the longer-term question should be: what is causing it?

James discussed tackling recurring problems at source, including looking at the processes and controls creating them, rather than continually fixing the resulting records.

So the aim isn't to create more manual work. It's to make data quality more visible, controlled and manageable.s.

It's short-term pain for long-term gain. Put in the effort upfront to resolve your key data quality issues and you will reap the rewards later on for years to come.

7. What are the most common data quality issues that prevent AI from delivering reliable results?

During the webinar, Steve and James highlighted several recurring problems:

• Duplicate records

• Incorrect values

• Missing information

• Information stored in the wrong fields

• Invalid values such as email addresses or postcodes

• Unnecessary or outdated records

• Inconsistent reference data

• Different versions of the same information across multiple systems

But there's another important issue: meaning.

Even technically clean data can be difficult for AI to use reliably if it's unclear what an attribute represents, which duplicate field is authoritative, or how particular values should be interpreted. Data quality therefore isn't simply about filling in blanks. It's also about making sure the data is consistent and understandable.

8. How would you go about testing the quality of the underlying data or the structure of your prompt to improve the reliability of AI outputs?

A useful approach is to look at these as two separate things.

First, test the underlying data against your agreed quality requirements. Is it accurate? Is it complete? Are values consistent? Are different systems telling you different things?

Then look at the AI output itself.

During the webinar, James and Steve stressed that AI-generated outputs need to be challenged and reconciled rather than simply accepted. They discussed how AI may present an incorrect conclusion confidently and that its answer may change when challenged or prompted differently.

Ultimately, improving the wording of your prompt won't fix unreliable underlying data. A better prompt can help AI understand what you're asking. Better data helps ensure the answer it's working towards can actually be trusted.

9. Bringing in a data quality tool just for a data migration project is often very costly and can take months to get approvals from client organisations. For the same cost, you can hire 2 or more extra people to help handle the data quality issues during the project. The DQ tools are often more cost effective when used in the longer term on organisation's operational systems. Is this just my experience?

This is a fair thing to say, but the answer will depend on the organisation, the scale of the project and the DQ tool in question. A well priced DQ tool, such as Binary-View, would cost less than 2 senior data analysts, with specific DQ experience and SQL skills, over the course of a year, and such expenditure on extra resource is likely to be subject to the same budget approvals, and may be take longer due to the recruitment process.

For a one-off project, adding more people can appear to be the simpler option. But people still need a way to identify where the issues are, how widespread they are, what needs prioritising and whether things are actually improving.

There’s also a bigger question about what happens once the migration is finished. One of the themes we discussed throughout the webinar was that data quality is no longer necessarily a one-off exercise. New systems, changing processes and everyday data entry can introduce new issues over time. If that data is feeding AI and other business-critical processes, organisations need a way of understanding whether its quality is being maintained.

That’s where the business case for a data quality tool starts to look different. Rather than comparing the cost of a tool with the cost of additional resource for one project, it becomes a question of how you monitor and manage data quality over the longer term.

The strongest approach may not necessarily be people or technology. People still need to make decisions, understand the business context and resolve the underlying causes of issues. Technology can give those people greater visibility of where attention is needed.

So, no, we don’t think this is just your experience. For a single migration, the cost-benefit calculation needs to make sense. The greater value comes when the monitoring established during the project continues to support the organisation afterwards.

Want to explore this in more detail?

In our "is your data ready for AI?" webinar, we explored where to start, which data quality issues to prioritise and why keeping your data AI-ready needs to become an ongoing process rather than a one-off exercise! 💚

Fill in your details and we'll send the recording straight to your inbox. 📥