A trends briefing for CEOs and revenue leaders trying to work out why their AI investments are not delivering what the demos promised, written by Chris Watson-James.
Most AI projects are not failing because the technology is broken. They are failing because the data underneath is not in a state that lets the technology do anything useful. Gartner predicts that through 2026, 60% of AI projects will be abandoned because they are not supported by AI-ready data. A Gartner survey of data and analytics leaders also found that only 39% of technology leaders are confident their AI investments will have a positive impact on financial performance. Six in ten are not.
RAND’s research on more than 2,400 enterprise AI initiatives puts the overall failure rate at 80%+, twice the rate of non-AI IT projects. Almost none of those failures are about the technology. The models work. What is going wrong is sitting in the layer underneath them.
The question isn’t which AI platform to choose. It’s whether your CRM would let any platform succeed.
Clean data is accurate, deduplicated and well-formatted. Structured data is connected, with records linked to each other in a way that reflects how the business actually operates. AI agents need both, but structure is what most CRMs are missing.
This is the part most leadership teams get wrong, and the part vendors do not really want to talk about. When people say “data quality” they usually mean hygiene. Are the records accurate, are they de-duplicated, are the fields filled in, is the formatting consistent. This matters, and it’s what AI can actually help you fix.
Structure is a harder problem to see, and a harder one to fix. Where hygiene asks whether your data is accurate, structure asks whether it is connected. An employee record that knows which company that employee works for. A company record that knows which other company records belong to the same parent. A contact record that knows there is already an open deal in progress with that person. A deal record that knows which marketing campaigns the contact engaged with before they ever spoke to a salesperson.
Clean data without structure gives you a tidy CRM and an agent that cannot do anything useful with it. Structured data is what feeds first-party context into the AI tools you are building, and first-party context is what separates an agent that works from an agent that hallucinates.
An AI agent sees only the records and associations it has direct access to. If two records are not linked in your CRM, the agent does not know they are related. If a contact has no parent company, the agent treats them as a standalone individual. The agent operates on the structure of your data, not on what you, as a human, know about your customers.
The easiest way to explain this is to think about what an agent is actually being asked to do. If you ask someone to paint a scene of your leadership team, and you only show them one person’s face, they are going to come back with a lovely picture of that one person. They are not going to come back with a picture of the team, because they never saw the team.
That is essentially what is happening every time you ask an agent to do something against a CRM that is full of disconnected entities and missing associations. The classic example is an employee record that is not associated with a company record. Your agent goes off to do some research before reaching out, and it has the person’s name and email, but nothing about the business they work for. It either pulls back a generic, useless piece of research, or it goes off and tries to figure it out from the email domain – which sounds fine until you remember that LinkedIn profiles are usually matched to personal email addresses, that brand names and domains often do not match, and that Companies House does not use domains at all, it uses company numbers.
Or take a business like Nike. Most large companies have multiple domains. A .com, a .co.uk, a .org, regional sub-brands, acquired subsidiaries. If your CRM has those as separate, unlinked company records, your agent has no way of knowing they are all the same parent business. It will treat them as three different prospects. It will run three different outreach sequences. It will report three different sets of engagement stats back to your team. And nobody will spot the problem until the customer does.
This is what people mean when they say first-party data is the most valuable input for good AI agents. It is not just having the data. It is feeding the structure of the data into the agent, so it can see the whole picture rather than the three corners you happen to have records for.
The fix is not really an AI problem. It is a revenue operations problem. Rev ops as a discipline has been around for a while, and the core idea is straightforward. You build a single source of truth that spans the full customer life cycle. First touch, marketing engagement, sales conversation, deal, onboarding, customer success, expansion. One end-to-end view, accessible to everyone in the business who needs it.
The challenge is that most businesses are running in exactly the opposite direction. ChiefMarTech’s 2025 State of Martech report lists 15,384 marketing technology solutions, up 100x since 2011. WalkMe puts the average large enterprise at 625 applications in use, with 43% of stacks more complex than they were three years ago. Every new tool is another potential silo. And every silo is another piece of context your AI agents will not have access to when you ask them to do something useful.
Source: ChiefMarTech State of Martech Report, 2025
The reason this matters for AI specifically is that an agent can only operate on what it can see. If your sales team and your marketing team are working off different systems, your agent is also working off different systems. If your customer success data is sitting in a tool that does not talk to your CRM, your agent does not know whether the contact it is reaching out to is a happy customer, a churning customer, or a former employee who left two years ago.
The companies getting AI to succeed are the companies that have done the rev ops work first. Not because rev ops is glamorous, it is not. But because rev ops is what gives your AI agents the connected first-party data they need to actually be useful.
The pattern in the data is fairly consistent. The businesses spending the most on AI are not, in our experience, the ones getting the most from it. The ones getting genuine return are the ones who recognised that the AI investment was, fundamentally, a CRM and data investment first. They got the rev ops foundations sorted, they made sure their systems were properly integrated, they built a single source of truth, and then they layered AI on top.
The ones writing off projects in 2026 are mostly the ones who did it the other way around. They bought the AI platform, deployed it against a fragmented CRM, watched it produce inconsistent or embarrassing output, and concluded the technology was not ready. The technology was ready. The data underneath it was not.
If you are a CEO or a revenue leader thinking about where to put the next chunk of AI budget, the most useful question is probably not about which agent to build first. It is about whether your CRM, in its current state, would let you build any agent successfully. If the honest answer is “we are not sure”, that is the place to start.
The questions below come up consistently when we work with leadership teams on this topic. They are intended as a reference, definitions and diagnostics you can return to rather than a recap of the argument above.
AI-ready data is data that is accurate, governed, and structurally connected in a way that lets AI agents see the full context of a record. According to Gartner, it requires data aligned to specific use cases, actively governed at the asset level, supported by automated pipelines with quality gates, and continuously quality-assured. Traditional data management practices were built for reporting, not for AI, which is why most CRMs are not AI-ready by default.
Most AI projects fail because of the data and systems underneath them, not the AI itself. RAND’s research puts the enterprise AI failure rate at over 80%, twice the rate of non-AI IT projects. Gartner predicts that 60% of AI projects unsupported by AI-ready data will be abandoned through 2026. The technology works. The foundation it is being asked to run on usually does not.
Read our full guide ‘What to Do If You Want Your AI Project to Fail: A Leader’s Guide’.
Clean data is about hygiene: accurate fields, no duplicates, consistent formatting. Structured data is about connections: whether your CRM records are linked together in a way that reflects how your business actually operates. An employee record linked to a company. A contact linked to an active deal. A parent company linked to its subsidiaries. AI agents need both. Most CRMs are reasonably clean but structurally fragmented, which is why AI projects built on top of them underperform.
Start by checking three things:
1. Are contacts properly associated with the companies they work for?
2. Do your sales and marketing teams work off the same connected data, or two separate systems?
3. Can someone in your business see the full customer journey for any given account, from first touch through to current status, in one place?
If the answer to any of those is no, your CRM is not yet in a position to feed AI agents the context they need.
If the patterns above feel familiar, there are two useful next steps depending on where you are.
If you want a day out of the business to work through this properly, away from the inbox, with the right people in the room, our AI Workshop for Leaders is built for exactly this. A full day working through the AI readiness of your data, systems and processes, and identifying the AI projects most likely to deliver early commercial impact in your specific business
If you would rather have a short conversation about your specific situation first, you can book a call with me directly and we can talk through what good looks like for your business and where the foundations need work.