The 25-Year System Trap: Why AI Exposes Weak ATS and CRM Governance
By Tony Kvatch
Why ATS and CRM data becomes unreliable: competing team inputs, unnecessary fields and systems teams without authority to challenge requests. How AI magnifies the problem.
AI can make system data more useful, but it cannot define the business rules that give the data meaning. Illustration generated with AI for FutureHero.
I’ve spent more than 25 years in recruitment, including over 15 years focused specifically on CRM, Data, AI and Marketing Automation recruitment.
I’ve also helped shape the ATS and CRM systems used in the organisations I’ve worked for. My contribution has included strategic input on data, governance and workflows, as well as creating Salesforce flows, fields and dashboards. Across that experience, one problem keeps returning: dirty data is rarely just a technical problem. It is often the result of how an organisation makes decisions about its systems.
How the data problem is created
A business invests in a capable ATS or CRM. Then different teams ask for fields and processes that suit their own work. One team wants to capture information one way; another wants a separate field or status. Each request can seem reasonable on its own.
The problem is what happens when nobody assesses the requests together. Definitions are not agreed. Similar information is entered in different places or formats. New fields accumulate, including free-text fields that let everyone record the same thing differently.
Meanwhile, the CRM or systems team is treated as an order taker. It is told what to add, but may not have the authority or expectation to ask whether the field is needed, whether the information already exists, or what the change will do to reporting across the business.
That is how a capable system gradually fills with inconsistent, duplicated or hard-to-use information. The dirty data is the visible symptom. The internal process that keeps producing it is the root cause.
When reports tell the wrong story
Consider a revenue report built without agreed lifecycle and source-of-truth rules. It may combine revenue already earned, forecasts for unfilled roles and historic placements. The report can run as designed, yet the number it presents is misleading.
The same pattern affects recruitment data. A “top candidate” feature cannot reliably surface the strongest people if records are duplicated, tags are inconsistent or important details are buried in free text.
When these problems become visible, leadership can blame the software and respond with another dashboard, another field or a major clean-up. But if the same internal practices continue, the data problem returns.
AI does not fix the cause
AI can summarise records, identify patterns and help standardise information. It cannot decide what “active”, “qualified” or “revenue” means to the organisation when teams have never agreed.
When data is incomplete or contradictory, AI can produce a confident forecast from unreliable figures or return a flawed list based on inconsistent records. It did not create the underlying problem. It can make the consequences faster and harder to spot.
I wrote separately about the data foundations organisations need before making an AI hire in AI Won’t Fix Messy Data. This article looks further upstream: why ATS and CRM systems keep accumulating the same data problems.
Give systems teams a mandate to govern
The answer is not to reject every request or make systems harder to use. It is to assess requests before building them. What decision will this information support? Is it already captured? What should the field mean? Who owns the definition? How will the change affect other teams and reports?
Systems teams need the remit to ask these questions and push back when a request creates more problems than it solves. Organisations also need agreed definitions and owners, controlled inputs such as picklists and validation rules where appropriate, and clear source-of-truth and lifecycle rules for records and stage changes.
After more than 25 years in recruitment, I don’t think the answer is simply a more powerful ATS or CRM. The same data problem persists because the internal behaviours that create it have not changed.
AI did not create this 25-year system trap. It makes the cost of leaving it unresolved harder to ignore.