Your AI Tools Are Failing Because Your Data Is Broken

Before you hire another prompt engineer, fix your CRM. Part 1 of a 5-part series on building AI-GTM infrastructure that actually works.

Everyone’s hiring prompt engineers. Nobody’s fixing their CRM.

This AI gold rush in B2B marketing has created a predictable pattern: companies are spending five-to-six figures on AI tools while their contact database has 47,000 duplicates and their marketing automation platform still can’t talk to their data warehouse.

You can’t build AI on top of a mess. And most B2B SaaS companies have a massive data mess they’re pretending doesn’t exist.

I’ve audited the GTM infrastructure of a dozen companies in the last six months. Every single one wanted to talk about generative AI, predictive scoring, and automated campaigns. Partly because it’s all the rage. It’s all anyone is talking about online these days. But not one of them wanted to talk about why they had three different definitions of “qualified” living in three different systems, or why their sales team didn’t trust any data coming from marketing, or why half their tech stack doesn’t integrate.

The gap between AI ambition and the reality of some of these infrastructures is stunning. It’s costly, yet literally no executive wants to hear you say the word “infrastructure” — they aren’t getting it.

---

👋 Hiii, it’s Kaylee Edmondson and welcome to Looped In, the newsletter exploring demand gen and growth frameworks in B2B SaaS. I write this newsletter every Sunday, and wildly, a few thousand of you read it each week. I’m grateful. If there’s ever anything in particular I can help explore and write about, hit me up! I’d love to chat.

---

Over the next five weeks, I’m breaking down how to build an AI-driven GTM engine (not just bolt AI tools onto your existing chaos). This series will cover the full build:

• Part 1: The Foundation — Data Hygiene and Infrastructure (this article) Before you automate anything, you need clean data and connected systems.

• Part 2: The Inputs — Messaging, ICP, and Intent Signals

  Teaching AI what “qualified” looks like in your GTM motion.

• Part 3: The Engine — Workflow Automation and Decision Loops

  Where human strategy meets machines.

• Part 4: The Layer — Content, Creative, and Conversation

  The customer-facing applications where AI impacts your market.

• Part 5: The Feedback Loop — Measurement, Iteration, and Enablement

  How to measure, refine, and scale your AI-GTM engine.

Most companies are trying to start with Part 4 (the sexy AI content applications) while ignoring Parts 1-3 (the infrastructure that makes AI actually useful). We’re going in order for a reason.

More AI, More Bad Data Problems

Here’s what most people miss about AI in GTM:
AI doesn’t fix bad data, it scales it.

If your CRM has duplicate records, AI will target the same person 6 different ways through 6 different “accounts.” If your attribution is broken, AI will optimize for the wrong signals. If your ICP definition lives in someone’s head instead of in structured data, AI will target everyone and no one.

Three months ago, I audited a demand gen team that had spent $200K on AI tooling in the previous year. Their setup looked impressive on paper:

• Clay for data enrichment

• 6sense for intent signals

• Pocus for product-led scoring

• Custom GPT implementations for content

• Jasper for copy generation

• Multiple AI-powered ad platforms

They were tracking 180+ different signals across these systems. They had AI generating email sequences, ad copy, and landing page variations.

But here’s what I found during that audit…

Their CRM had accounts with 4-7 duplicate entries because different tools were creating records differently. Contact records were split across “Leads” and “Contacts” objects with no clear conversion logic. Lead Status/Contact Status properties in their CRM weren’t being accurately (automatically or manually) updated, leaving no one with a clear view of their status. Their intent signals from 6sense weren’t syncing properly with HubSpot, so sales was working off incomplete data. The product usage scores were in a completely different system than their marketing qualified leads.

When an account showed high intent, nobody could figure out who to reach out to because the contact data was fragmented across three systems. When sales asked “why is this account qualified?” marketing couldn’t answer because the signals were scattered. They had to manually query multiple systems to provide a story.

The VP of Marketing told me: “We have all this data, but we’re less confident in our targeting than we were two years ago when we had a simple scoring model in HubSpot.”

I watched one brand’s AI-powered ad platform spend $15K targeting accounts that were already customers because their CRM didn’t clearly flag which accounts were in which stage. The AI was optimizing beautifully toward completely wasted spend.

Another brand had AI generating hyper-personalized outreach based on intent signals, but 40% of the email addresses were wrong because nobody had cleaned their contact data in 18 months. Their bounce rate made their domain reputation tank, which hurt deliverability across all their campaigns.

You can have the most sophisticated AI models in the world, but if they’re training on garbage data, you get sophisticated garbage.

What “Clean Data” Actually Means for AI-GTM

When I talk about data hygiene with clients, I see eyes glaze over. Everyone knows they should “clean their data,” but nobody wants to do it because it feels like busywork. I can’t stress enough that this is not in fact busywork.

Let me be specific about what matters for AI-driven GTM:

Single source of truth for accounts: Every company in your TAM should have exactly one record in your CRM. Not 3 records with slightly different names. Not separate records for different divisions. One record, with one unique identifier, that every system references. Take Parent/Child relationships seriously, too.

Unified contact data: All contacts associated with an account should be in the same system, with clear relationships to that account. You should be able to pull up an account and see every person, their role, their engagement history, and their current stage.

Consistent taxonomies: Your industry classifications, company size buckets, and account stages need to mean the same thing across every system. If marketing calls something “MQL” and sales calls it “SAL” and they’re tracked in different fields, your AI can’t learn from the pattern.

Connected signal capture: Every touchpoint—website visits, content downloads, product usage, email engagement, ad interactions, sales calls—needs to flow into your source of truth system. Not just captured, but properly attributed to the right account and contact.

Historical context: AI gets better with more training data. That means you need to preserve historical engagement, deal cycles, win/loss reasons, and campaign influence—not just current state.

This isn’t about having perfect data. That doesn’t exist. But you need systematically structured data that AI can actually learn from.

Your Tech Stack Is Probably Working Against You

When I did a tech stack audit for a Series B company. They had:

• Salesforce (CRM)

• HubSpot (Marketing automation)

• Clay (Enrichment)

• Clearbit (More enrichment)

• ZoomInfo (Even more enrichment)

• 6sense (Intent, or something)

• Pocus (Sales intelligence signals)

• Apollo (Outbound, but also enrichment)

• Gong (Call intelligence)

• Chili Piper (Scheduling, Routing)

None of these systems were talking to each other properly. Data was flowing one direction but not back. Different systems had different definitions of the same fields. RevOps was drowning in internal enablement, manual processes. The house was literally always on fire.

They asked me: “What AI tools should we add?”

My answer: “Honestly, none. Your tech stack needs to get smaller and more connected before you can make AI useful.”

The problem with the great unbundling of ABM platforms (which I’ve written about before) is that everyone went from one expensive, bloated system to 15 cheaper, disconnected systems. We traded one problem for another.

Here’s what a connected tech stack for AI-GTM actually looks like:

Core system of record: Your CRM (Salesforce, HubSpot, or Attio, whichever) becomes the single source of truth for all account and contact data. Everything else feeds into this.

Orchestration layer: You need something that can move data between systems and trigger actions based on signals. This might be your marketing automation platform, a lightweight tool like Zapier or Make, or even a dedicated tool like n8n.

Enrichment pipeline: Pick ONE primary enrichment source. Have a clear process for when and how data gets enriched. Stop paying for three different tools that do the same thing and conflict with each other.

Signal aggregation: Choose how you’ll collect and score signals. Maybe that’s native in your marketing automation platform. Maybe it’s Pocus or Common Room or a custom data warehouse setup. But pick one approach and enable the team.

Execution tools: These are your email platforms, ad platforms, sales engagement tools. They should READ from your core system and WRITE back results, but they shouldn’t be creating their own versions of truth.

The RevOps Bottleneck (And Why You Need to Fix It)

Every AI-GTM conversation eventually comes back to RevOps. And for good reason.

RevOps owns the infrastructure that makes AI possible. They manage your CRM architecture, your data flows, your system integrations, your reporting infrastructure. If RevOps is underwater, your AI initiatives are D.O.A.

Here’s what I keep seeing: Companies pile more tools onto their RevOps team, expecting them to maintain integrations, clean data, build reports, AND enable AI, all while supporting a growing sales team, new product launches, shifting GTM strategies. Wild. There’s not a single RevOps team I’ve ever seen that feels properly staffed. PSA: Send a note of gratitude to your RevOps teams today.

If you’re serious about AI-driven GTM, you need to invest in your RevOps capacity before you invest in more AI tools. That might mean:

• Hiring a dedicated marketing operations person who can own the marketing-to-sales data flow

• Bringing in a data engineer to build proper pipelines and automation

• Allocating budget for professional services to properly implement and integrate your core systems

• Saying no to new tools until your existing stack is properly connected

The unsexy infrastructure work pays off. Trust me.

A Data Hygiene Sprint

You can’t fix everything at once, but you can make meaningful progress in 30/60/90 days, whatever it might be. Here’s an overly simplified rough draft for what might need to be included in that sprint for your org:

Audit and Assess

• Document every system that touches customer or account data

• Map how data flows between systems (or doesn’t)

• Count your duplicate accounts and contacts

• Identify your biggest data quality issues

• Assess current RevOps capacity and constraints

Define and Standardize

• Create your single source of truth hierarchy (which system owns what)

• Standardize your field taxonomies (industry, size, stage, etc.)

• Document your defined lifecycle stages (with actual criteria)

• Define your account and contact matching rules

• Build consensus across marketing, sales, and CS on definitions

Dedupe and Connect

• Run your deduplication process on accounts (start with Tier 1)

• Merge duplicate contacts (carefully - this is where you can break things)

• Set up or fix your critical system integrations

• Establish data enrichment rules and workflows

• Build basic data quality monitoring

Test and Document

• Test your data flows end-to-end with sample accounts

• Verify that signals are properly capturing and attributing

• Document your new data standards and processes

• Train your GTM team on the new structure

• Set up ongoing data quality checks

This isn’t a one-time project. You’ll need ongoing maintenance. But this sprint gets you from “our data is unusable” to “we have a foundation we can build on”.

The Hard Truth About AI Readiness

Most marketing teams aren’t ready for more than point solution AI. Not because they lack the budget or the talent, but because their data infrastructure can’t support it.

The companies that are seeing real results from AI in GTM are the ones who spent the last year cleaning up their data model, connecting their systems, and building clear operational processes. The ones who invested in RevOps capacity. The ones who said no to shiny new tools until they fixed their foundation.

You can hire all the prompt engineers you want, but if they’re working with fragmented data across disconnected systems, they’re just making expensive garbage faster.

Start with your data. Everything else follows from there.

Next week in Part 2: We’ll cover how to teach AI what “good” looks like—defining your ICP with precision, mapping your messaging hierarchy, and building the signal taxonomy that makes AI useful. But you need clean data first, or none of that matters.

See ya next week,
Kaylee ✌