Poor data foundations are killing your ServiceNow AI investment

It's completely fair to acknowledge that among all the spark and excitement of AI, data is not exactly the topic that sets pulses racing.

But the success of what you’re spending on AI today will only pay off if it is built on strong data foundations.

The saying goes: “rubbish in, rubbish out”. AI will act on what it's given – and won’t pause to question whether the data is stale, incomplete, or missing context. Poor foundations hold you back and lead AI to produce confident mistakes, which, in themselves, are a business risk, especially in an agentic landscape where AI reasons and makes decisions based on them.

These solid foundations provide a grounding that AI models can use, it informs them of what good looks like for your organisation – rather than relying on the training data behind a large language model or the open web.

However, building this cannot be treated as a one-off hygiene task. In this blog, I’ll explain why strong data foundations matter, and more importantly, how you should be approaching your data to get a real return on your AI investment.

Why data is the deciding factor for AI success

Without grounding your data, AI output can become noise and loses the precision and accuracy needed for business-critical decisions.

Grounding data means AI reasons from information that represents the reality of your business at any given moment. Outputs otherwise rely on AI training data and the open internet, leaving them prone to hallucination and incorrect decisions because they lack the necessary context.

While data has always been important, the technological shift towards AI has made it an imperative. No longer considered an IT housekeeping task, the success of AI programmes hinges on maintaining strong data quality management across any source where AI is in the loop.

A data quality framework for AI-ready data

If you consider an area like IT Service Management (ITSM), there are several key data sources that are frequently used in utilising AI solutions. Knowledge articles can contain information such as policies, processes and solutions to issues that have been experienced before. If these are out of date, then so will the generated output produced by an LLM.

The Configuration Management Database (CMDB) is the repository that stores comprehensive data about your infrastructure. The data in that CMDB must be clean, accurate, current, consistent, and representative to be considered AI-ready.

Sitting on top is a CSDM (Common Service Data Model), which organises data into business-aligned services, helping AI understand the context and relationships across different areas of the organisation.

To understand how they work together, imagine the map of the London Underground. The CMDB, which shows every asset, system, and dependency, is represented by the stations. The CSDM shows you the routes between those points on the map – translating raw infrastructure into named services, clear ownership, and the relationships AI needs to reason from.

Governance over both domains is not a one-and-done job but an ongoing process which glues it all together. According to ServiceNow’s whitepaper on CSDM 5, data: “provides little value without governance and effective process to manage the veracity of the data.”

In my experience, the biggest challenge is stale or inaccurate data within the CMDB and CSDM that doesn't represent the business, and this is the failure AI exposes the fastest.

Four principles to get your CMDB AI-ready

Now that we’ve covered what underpins the foundations, here are four principles to get your CMDB working with AI:

1. Start narrow: Focus on the high-value use cases and where AI will interact first. For example, your highest-volume service catalogue items or top business services. Trying to fix everything before you start is how organisations end up with no AI, an exhausted team, and a project that runs out of patience before it delivers.
2. Fix relationships: Help AI understand the connections in the business, looking at dependencies and who owns what. It's tempting to focus on tidying up individual records, ensuring that every asset has the right identifier or classification. But that's not where you will find value. AI's ability to spot patterns depends on the connections: which service depends on which, which team owns what.
3. Build the habit: Assign clear owners for maintaining the data and establish a governance system to prevent AI output quality from degrading over time. It’s the principle that gets the least attention, and it's arguably the most important. Getting your CMDB into good shape is not a “one-and-done” job – without clear owners, quality decays over time, and AI outputs degrade with it.
4. Measure as you go: To understand that data in your CMDB is trustworthy before it informs AI decisions. Treat the CMDB like any other business-critical process: track completeness, correctness, and compliance as ongoing metrics. You need confidence in the data before you can trust AI agents to act on it.

Where data foundations go wrong with AI programmes

An estimated 40% of agentic AI projects will be cancelled by the end of 2027 according to Gartner. The reason behind that statistic is that organisations struggle with governance, escalating costs, and an unclear business value.

Finding ROI from AI remains a genuine blocker. The most common root cause is starting without a clear view of the real use cases for AI. On top of this, AI isn’t always the answer, sometimes automation is a better fit for solving certain challenges.

Without genuine use cases, the focus is on the wrong areas, and organisations don’t achieve  the expected outcomes.

Ungoverned data is another common failure point. When AI operates on stale, inaccurate, or unowned data, it produces incorrect decisions at speed and with authority.

Building CMDB best practice using ServiceNow

Firstly, focusing on the high-ROI use cases can define the scope of the data you fix, rather than letting data define everything else. By starting here, you can build momentum once an initial use case delivers a return.

ServiceNow recommends following a five-step process that acknowledges everyone will be at a different stage in this journey and not to tackle everything at once. These steps are:

1. Foundation: the organisational structure, locations, products, and configuration item (CI) lifecycle status are often thought about too late in the process. This layer underpins reporting, workflows and AI agents.
2. Crawl: focuses on your application inventory, application services, and the discoverable infrastructure supporting them. This is the minimum CMDB requirement to support incident, problem, and change management.
3. Walk: adds technology management services and offerings. Establish who owns and manages the applications and infrastructure CIs, with clear support structures and ownership across the estate.
4. Run: connects the technology layer to business outcomes by introducing business services and service offerings. The layer enables impact assessment in ITSM and lays the groundwork for service portfolio management.
5. Fly: Completes the model with business capabilities and information objects. It’s the strategic layer that allows you to rationalise the whole service and application portfolio and assess whether spend is aligned with the right capabilities.

Governance is the glue that makes this process stick. During our recent Managed Services User Group, we heard how Avanti made its IT community data stewards and put a ‘Change Charter’ around services so data stays current. National Gas implemented governance before scaling – including mandatory CI attributes, and ownership at the right service layer.

ServiceNow makes this simple to manage. The platform shares health metrics on the completeness, correctness, and compliance of your CMDB data, automatically flagging tasks and actions to take when needed.

Let’s get your data AI ready

At UP3, we believe that AI will unlock productivity gains when implemented effectively, allowing people to focus on driving innovation and developing new ways of getting work done.

If you want help fixing your data foundations or identifying your highest-impact AI use cases, get in touch. Through our AI Unlocked Service Offering, we can get your first AI agent live in as little as three weeks, with the right foundations in place.