AI Won't Fix Messy Data — What to Do Before Your First AI Hire

Every week there's a new announcement about what AI will do for your marketing, your CRM, your data strategy. The capability is real. But there's a version of this story that doesn't get told enough: the companies getting the least from AI are often the ones who hired for it first and sorted out their foundation second.

After 15 years placing Marketing Automation, CRM, and Data specialists across ANZ and Southeast Asia, I see two mistakes more than any other when organisations try to bring AI into their marketing and data stack.

The first is not knowing what data they actually have, what shape it's in, or what they need it to do. The second is hiring the wrong person — or worse, not hiring anyone at all and just adding "AI" to an existing job description that was already stretched thin.

Both mistakes are expensive. And they compound each other.

Mistake One: Not Sorting Out the Data First

AI is a pattern-recognition engine. It learns from data. The better and more consistent that data is, the more reliable the output. The messier the data, the more confidently wrong the AI becomes — which is more dangerous than no AI at all, because at least you know when you're guessing.

Most organisations I speak to significantly overestimate the quality of their CRM and marketing automation data. They know it's "not perfect" — but they don't know specifically what's broken, where the gaps are, or what it would take to fix it.

Before you hire an AI specialist, you need honest answers to these questions:

What does your CRM actually contain? Not in theory — in practice. How many duplicate contacts? How many records with no email address? How many lifecycle stages that haven't been updated since the initial setup? How many contacts who've been sitting in "subscriber" status for three years with no engagement?

Do you know where your data comes from? Every source that feeds your CRM — forms, integrations, manual imports, third-party enrichment, outbound sequences — is also a potential source of inconsistency. If you don't know what's flowing in and how it's structured, you can't control what you're working with.

Is there a single source of truth? The most common data architecture problem I see isn't missing data — it's duplicated and contradictory data spread across systems that don't properly sync. CRM says one thing. Marketing automation says another. The billing system says a third. No AI layer resolves this. It just picks one version and runs with it.

Is data maintenance someone's actual job? This is the question that exposes most organisations. Data quality isn't a project. It's an ongoing operational responsibility. If nobody owns it — if it sits somewhere between marketing, sales ops, and IT with no clear accountability — it will decay faster than any AI tool can keep up with.

Mistake Two: Hiring the Wrong Person — or Nobody at All

The second mistake follows logically from the first: because organisations don't fully understand what the AI role requires, they either hire someone too junior for the problem, or they don't hire at all and tag "some AI work" onto an existing role that already has too much in it.

I see this constantly. A marketing coordinator who "uses HubSpot" gets asked to explore AI personalisation features on top of their existing campaign workload. A CRM admin who "knows Salesforce" gets handed an Einstein implementation alongside their regular support queue. Neither person has the depth, the time, or the mandate to do it properly. The initiative stalls. Leadership concludes that AI "isn't ready for us yet." The real conclusion should be: the role wasn't set up for success.

Genuine AI work in a marketing and CRM context requires someone who understands:

• Data structure and quality — not just at a surface level, but well enough to diagnose why an AI output is wrong and trace it to a data issue
• The platform deeply — AI features in SFMC, HubSpot, Braze, and Salesforce CRM behave differently, are configured differently, and require different expertise. "Exposure to AI tools" is not a qualification. Platform-specific depth is.
• The commercial context — AI recommendations that don't connect to actual business outcomes are noise. The specialist you need understands what you're trying to achieve with customers, not just how to configure a model.

And critically: this person needs to own the work. Not share it. Not do it alongside three other responsibilities. Own it, with clear objectives and the authority to make decisions about how the data is structured and maintained.

The Right Sequence

If you want AI to actually deliver in your marketing and CRM stack, the sequence matters:

1. Audit your data — understand what you have, where it comes from, and how reliable it is. Be honest about what's broken.
2. Clean and unify — resolve duplicates, standardise fields, establish a single source of truth across CRM and marketing automation.
3. Connect your systems — AI needs a complete picture. Disconnected data produces narrow, unreliable outputs.
4. Define your use case specifically — "use AI to improve marketing" is not a use case. "Use AI lead scoring to prioritise outbound sequences for contacts who've visited the pricing page more than twice in 30 days" is a use case.
5. Hire the right specialist for that specific problem — someone with genuine depth in your platform and a track record of working with data, not just a candidate who listed "AI" on their CV.

The platforms are ready. The AI capabilities are real. The question is whether your organisation has done the foundational work that lets those capabilities actually function.

Getting that foundation right is not glamorous work. But it's the difference between an AI initiative that delivers and one that costs a lot and proves nothing.

FutureHero specialises in placing Marketing Automation, CRM, Data and AI talent across ANZ and Southeast Asia. Talk to us before your next AI hire.