Every organization depends on data it cannot fully trust, and that gap has stopped being a filing inconvenience and become an AI exposure. In this episode of What Counts, Lee Karas and Maura Dunn pick up the two questions they left open last time: how do you determine which version of a record is the most recent, and how do you determine whether your data is reliable? Their answer is that these are the same question, because version is a component of reliability, and neither one gets fixed by searching harder. Maura draws the distinction between universal keyword search, which is designed to return the universe and leaves you to filter it, and purpose-built systems that tell you what actually happened most recently. That distinction is low stakes when you are looking for the nearest gas station and very high stakes when you are looking for the last inspection, leak, or repair record on a pipeline. Listen to that argument with a current ear and it describes the enterprise AI problem exactly. Retrieval-augmented assistants like Microsoft 365 Copilot ground their answers in whatever a user already has permission to open across SharePoint, OneDrive, Teams, and Outlook, which means every stale copy, every unlabeled duplicate, and every wrong version is now a candidate answer delivered with total confidence. Obscurity was never a control, and AI removed it.
The conversation then moves from safety-critical data to the ordinary documents that quietly become emergencies: the certificate of insurance you need in a hurry, the broker email you never followed up on, the process that works perfectly right up until the moment it is critical. And the fix, Lee and Maura argue, is not a better search bar or a better model. It is creating and storing information through a consistent process, so that finding it, trusting it, and grounding an AI system in it all become straightforward downstream. They walk through both the automated path (workflow engines, defined field mappings, source-to-analytics integration rules, threshold triggers) and the manual path most organizations actually need: record process mapping, checkpoints at the moment information becomes a record, clear ownership, and periodic quality assurance sweeps across SharePoint, email, and local drives. Those checkpoints are the same lineage and provenance controls that AI governance frameworks now demand, which means the unglamorous work of the last twenty years has quietly become the prerequisite for the next ten.
The episode closes on a deceptively simple example that lands harder every year: six interviewers, six saved copies of one resume. The copy itself was never the problem. Federal recordkeeping rules require employers to preserve applications and resumes from candidates they did not hire, so one governed copy is exactly what you are supposed to have. The other five are the problem. No owner, no retention clock, no disposition, scattered across inboxes and hard drives, and kept for the most human reason there is: nobody wants to be the person who lost something. Those five ungoverned copies are now five retrievable sources for an AI assistant grounded in whatever a user already has permission to open, and five items you will be explaining after a breach. The through line is Maura’s standing rule, and it has never been more literal: stop making copies, and do the thinking at the beginning.
What Counts is produced by TrailBlazer Consulting, LLC and hosted by Lee Karas and Maura Dunn. Learn more or reach out directly at info@trailblazer.us.com. Explore compliance-ready corporate training programs at the TrailBlazer Learning Academy. Read more from Maura at Maura’s Substack. Music by Jason Blake. Full disclaimer.
Episode Chapters
00:00 TrailBlazer Renewal Tracker
00:26 Welcome to What Counts
00:41 The Two Questions We Left Hanging
01:12 Stop Making Copies: The Bold Statement Revisited
01:52 Technology Is Great at Creating Data, Not Finding It
02:20 Universal Search Is Designed to Bring You the Universe
03:25 Low Stakes Versus High Stakes: Gas Stations and Pipelines
04:19 Purpose-Built Databases and Knowing What Changed Last
05:30 The Documents Nobody Builds a System For
05:58 The Certificate of Insurance Fire Drill
07:17 Everything Is Easy Until It Is Critical
07:58 Your Process Only Works When Everything Is Fine
08:32 Version Is Part of Reliability: The Same Question
09:16 How Does Data Actually Get Into the Report?
10:16 We Are Fixing the Wrong Problem
10:43 Fix How You Create and Store It, Not How You Find It
11:23 No Rules, No Trustworthy Analytics
11:40 But People Still Have to Follow the Process
12:20 The Automated Path: Workflow, Field Mapping, Thresholds
13:25 Writing It Down Is Tedious, and It Is the Investment
14:06 Sizing the Investment to Your Risk
14:29 Checkpoints and Record Process Mapping
15:07 Six Interviewers, Six Copies of the Resume
15:31 Why People Keep Copies: CYA and Fear of Losing Something
16:14 The Copy That Is Not Harmless: EEOC and Privacy Risk
16:26 Designing the Checkpoint: Tell People What to Do
17:00 Send a Link, Not Six Emails
17:40 The Quality Assurance Sweep: SharePoint, Email, Hard Drives
18:22 Checkpoints in a Manual Process
18:32 Ownership End to End: The Recruiting Example
19:27 The Payoff: What Happens When You Get Breached
20:46 Don't Make Copies, and Do Think
21:12 Next Time: Create Data Closest to the Source
21:35 Where to Find Us
22:10 Thanks and Credits