A Head of Capital Markets sat across from me a few months ago, frustrated that his team was trying to manage billions in investor relationships with spreadsheets and email.
“We manage relationships with dozens of institutional investors,” he said. “But when the CEO asks a basic question about an investor’s exposure, I spend 30 minutes digging through emails and spreadsheets.”
Investor lists in one spreadsheet. Deal participation in another. Meeting notes in OneNote. Relationship history buried in email.
“We looked at Salesforce,” he said. “Six-month implementation. $200K. And we’d still be forcing investor relationship management into a sales pipeline that doesn’t match how we work.”
Then he said something I’ve been hearing more often: “What if we just built our own?”
The Signal That’s Hard to Ignore
Earlier I posted about this trend: companies building their own CRM systems using AI tools instead of buying enterprise platforms.
The post generated over 100 comments. Not theoretical discussion but real examples. Founders who had already done it. Teams moving business units off Salesforce. Consultants describing clients replacing expensive CRM licenses with lightweight systems.
The comments that stuck weren’t from the builders. They were from the skeptics:
“The CRM tool will work great till the vibe coder quits!”
“Building a CRM with AI sounds smart until 6 months later when nobody knows how it works.”
“If you can’t handle a prebuilt CRM, building one will be worse.”
CRM Drift: The Pattern Behind the Rebuild Impulse
When companies start rebuilding core systems from scratch, something else has usually broken first.
Not the technology. The operational reality underneath it.
Here’s the pattern:
- Year 1-2: CRM implementation. Clean data. Teams actually use it.
- Year 2-3: Customizations accumulate. “Just add this field.” “We need a workflow for this edge case.” Nobody removes anything.
- Year 4-5: The CRM no longer reflects how the team works. Too many unused fields. Workflows built for projects that ended years ago.
- Year 5: Leadership asks: “Why are we paying $100K a year for a system nobody trusts?”
This is CRM Drift: the slow accumulation of complexity when nobody owns ongoing system health. Every addition made sense at the time, but the system only grows. Eventually the gap between how the CRM expects work to happen and how work actually happens becomes unbridgeable.
Teams don’t abandon the CRM first. They work around it. Spreadsheets for the real pipeline. Email threads for customer context. Slack channels for deal intelligence.
The CRM becomes what I call a Shadow CRM situation. The official system exists for Compliance Theater. Update it once a week so leadership sees activity in dashboards. But the actual work happens elsewhere.
And now, AI tools have made acting on that frustration trivially easy.
Why AI Changes Everything (and Nothing)
Five years ago, building a custom CRM meant hiring developers, defining requirements, managing a months-long project, and owning ongoing maintenance forever. The cost and risk weren’t worth it for most companies.
AI coding tools changed that equation. You can now describe what you want in plain language and get a working prototype in hours. The barrier to experimentation dropped to near zero.
But here’s what the enthusiastic comments missed:
AI removed the hardest part of building software. It did not remove the hardest part of running systems.
Who owns data quality when there’s no CRM admin? Who enforces process discipline when there are no validation rules? Who maintains the system when the person who built it leaves?
If your team struggled with Salesforce adoption because of governance failures, poor data quality, and Shadow CRM behavior, a vibe-coded system doesn’t solve those problems. It removes the guardrails that were barely holding the chaos in check.
AI doesn’t fix weak foundations. It exposes them.
Systems Built for Reporting, Not Execution
There’s a deeper issue driving this trend that most people aren’t naming.
Enterprise CRMs are built as systems of record, not systems of execution.
They’re designed to capture what happened so leadership can report on it. Fields to fill. Stages to update. Activities to log. The system tracks work. It doesn’t help you do the work.
This is why Shadow CRMs emerge. When the official system only serves reporting needs and doesn’t support the actual workflow, teams naturally build their real process somewhere else.
Enterprise CRMs like Salesforce and Dynamics are built around transactional sales pipelines: Lead, Opportunity, Proposal, Closed. That model works for transactional sales. It doesn’t work for relationship-led businesses, capital markets, professional services, or nonprofits managing donor relationships.
When your workflow doesn’t map to the CRM’s assumptions, teams build workarounds.
And eventually, someone asks: “Why are we paying for a system that doesn’t match how we work?”
The Case That Changed How I Think About This
The capital markets firm didn’t vibe code a CRM from scratch. They took a different path.
We used an enterprise platform with AI-assisted application building. A modern work platform that provided the infrastructure (database, security, audit trails, role-based access, integrations) while letting us configure custom workflows matching how their team actually operates.
We built two dashboards: one for executives showing investor exposure and capital allocation. One for the capital markets team showing deal status and relationship history.
Delivered in 60 days.
- Before: 30 minutes digging through email for basic investor information. “I think they were in that deal…” Relationship context siloed with individuals.
- After: 30 seconds to pull a complete investor view. “$40M participation at 8.5%” instantly available. Entire team aligned on every relationship.
But here’s what made it work:
- Clear ownership. The Head of Capital Markets owned the system. Not IT. The business leader whose success depended on whether the team actually used it.
- Enterprise infrastructure. Security, compliance, and audit trails came with the platform. In financial services, non-negotiable.
- Governance from day one. Data quality standards. Required fields. Review cadences. We built the operational discipline layer, not just software.
- Designed around actual workflows. We didn’t force investor relationship management into a sales pipeline model.
If any of those had been missing, the system would have failed within six months.
When Weak Foundations Get Exposed
This is where most vibe-coded CRMs fail. Not because the AI writes bad code, but because the conditions required to run a system well are rare.
- You fail when there’s no clear data owner. CRM Drift happens faster in custom systems because there are no forcing functions to slow it down.
- You fail when the team struggled with adoption before. Shadow CRM behavior doesn’t disappear because you rebuilt the system. It gets worse.
- You fail when security and compliance requirements exceed your capability. Vibe-coded tools don’t come with SOC 2 compliance or audit trails. You’re inheriting responsibility for all of that.
- You fail when you’re optimizing for cost instead of operational fit. The license isn’t the real cost. The real cost is low adoption, Shadow CRM behavior, and decisions made on incomplete data.
AI doesn’t fix these problems. It exposes them faster.
The Governance Layer Nobody Talks About
The debate about vibe-coded CRMs focuses on technology: Can AI write production-quality code? Can weekend prototypes scale?
These questions miss the point.
Technology was never the constraint. Governance always was.
Enterprise CRMs provide forcing functions. Admins who enforce data standards. Validation rules that prevent bad data. Workflows that ensure consistency. Dashboards that create accountability.
When you remove those guardrails, you inherit responsibility for building your own.
The companies succeeding with custom CRMs aren’t succeeding because they have better AI tools. They’re succeeding because they built operational discipline first.
Clear ownership. Data quality standards. Review cadences. Process enforcement.
These are the things that make systems work. Not the code. The governance layer.
If you can’t maintain that discipline in Salesforce, you won’t maintain it in a custom system. You’ll just trade licensing costs for technical debt that compounds faster than SaaS subscriptions.
What This Actually Means
When companies start rebuilding core systems from scratch, it’s usually a symptom, not a solution.
The symptom: the system stopped reflecting how work actually happens.
The real problem: nobody owned the system. Nobody pruned unused fields. Nobody removed workflows built for strategies the company abandoned.
AI tools make rebuilding faster and cheaper than ever. But speed and cost aren’t the variables that determine whether a CRM works.
Governance is. Ownership is. Operational discipline is.
The rebuild impulse isn’t about technology. It’s about CRM Drift, Shadow CRM behavior, and systems built for reporting instead of execution.
AI doesn’t solve those problems. It just makes the consequences of ignoring them arrive faster.
If your team could design a CRM from scratch today, how different would it look from what you’re using now?
If the answer is “completely different,” what does that say about the gap between how your business operates and how your systems expect it to operate?
The question isn’t whether you should rebuild.
The question is whether you’re willing to do the unglamorous work required to make any system, custom or enterprise, actually work.
Because that gap you’ve been ignoring? AI just made it impossible to hide.
