Ariel Palones
Every supply chain leader I talk to says the same thing within the first five minutes: "We know we have a problem. But our data isn't ready."
The data isn't clean. The systems aren't integrated. The processes aren't documented - they live in someone's head. And until all of that gets sorted out, they believe, any AI initiative is premature.
I've heard this from companies doing $500M in revenue. I've heard it from companies doing $5B. The language is almost identical every time. And every time, it's the wrong framing.
The Myth of "Data Ready"
Here's what "our data isn't ready" actually means when you unpack it:
Our system of record has inaccurate inventory data because nobody reconciles it in real time. Our purchase orders change after submission and the changes live in email threads, not in the system. Our 3PL sends CSV files that don't match our item master. Our allocation logic lives in a spreadsheet that one person maintains. Our tribal knowledge - the rules about which supplier tends to short-ship, which customer has a special pricing agreement, which warehouse has a labeling quirk - lives in the heads of people who are five years from retirement.
None of that is a data readiness problem. That's a description of the execution gap itself. The data isn't dirty because someone made a mistake. The data is dirty because there's no system managing the flow of changes between what happens in the real world and what gets recorded in the system of record.
Waiting until the data is "clean" before deploying an execution layer is like waiting until you're healthy before going to the doctor.
Why "Clean First, Automate Later" Fails
The conventional approach sounds logical: invest in a data quality initiative, standardize your master data, integrate your systems, and then layer intelligence on top of clean foundations.
In practice, this is a multi-year project that rarely reaches completion. The data keeps changing because the business keeps operating. New suppliers onboard. New 3PLs come online. Product lines launch and retire. The moment you finish cleaning one dataset, three others have drifted.
Industry research backs this up. A major 2026 review of AI deployment outcomes found that the vast majority of initiatives that failed did so not because the AI wasn't capable - but because of fragmented data, siloed systems, and undocumented workflows. The organizations that waited for perfect data never got there. The ones that deployed targeted automation on specific, painful processes saw returns within weeks.
BCG's supply chain planning report found the same pattern: organizations that try to leapfrog maturity by addressing everything at once tend to struggle, while those that layer capabilities deliberately onto existing foundations see durable gains.
The Tribal Knowledge Ticking Clock
There's an urgency dimension that makes "wait until we're ready" even more dangerous.
Across the supply chain workforce, roughly 25% of workers are 55 or older. Industry analysts estimate that 10,000 baby boomers retire every single day. When they leave, they take decades of undocumented, institution-specific knowledge with them - the vendor relationship nuances, the exception-handling shortcuts, the allocation rules that never made it into any SOP.
Every month you delay capturing that knowledge in a system - even imperfectly - is a month closer to permanent loss. A spreadsheet maintained by one person is a single point of failure. A process that lives in someone's head is a ticking clock.
The answer isn't to document everything first and then build systems around it. The answer is to deploy agentic workflows that learn from how your team actually operates - capturing their decisions, encoding their rules, and preserving institutional knowledge as a natural byproduct of execution.
What "Deploy Anyway" Actually Looks Like
This isn't about throwing AI at messy data and hoping for the best. It's about choosing the right starting point.
Start with one specific, painful, high-frequency process. Not "transform our supply chain." Not "build an AI strategy." Pick the process that burns the most hours and has the clearest financial impact. For most brands, that's one of: PO change management, inventory reconciliation, or allocation optimization.
Read from existing data sources as they are. Your 3PL sends a CSV by email? Read it. Your supplier sends ETA changes in the body of a reply-all thread? Parse it. Your system of record has custom fields that nobody documented? Map them. The execution layer meets your data where it lives - it doesn't demand you restructure everything first.
Surface the gap before trying to close it. In one recent deployment, a single inventory reconciliation workflow - reading from 3PL feeds and comparing against the system of record - revealed that more than a third of SKUs had mismatches between what the warehouse reported and what the business believed. Over $1M in inventory uncertainty, invisible until the workflow surfaced it. The data wasn't clean. The workflow made that visible - and then started closing the gap by writing corrections back.
Encode tribal knowledge as business rules. When the ops person says "this supplier always shorts by 5-8%," that becomes a rule. When the allocation manager says "this customer gets priority because of their contractual minimum," that becomes a workflow gate. Every decision your team makes and the system records becomes institutional knowledge that doesn't retire when they do.
The Integration Objection
"But our systems aren't integrated."
They don't need to be - not in the traditional sense. The execution layer doesn't require a rip-and-replace integration project. It reads from your existing data sources: system of record APIs, email inboxes, CSV files, 3PL portals. It writes validated corrections back to the system of record. It operates on top of your existing infrastructure, not instead of it.
The deployment timeline for a first workflow is measured in weeks, not quarters. No disruption to existing configurations. No migration project. No change management crisis.
The Real Risk
The risk isn't deploying with imperfect data. The risk is waiting.
Every week you wait, your system of record drifts further from reality. Every month, more tribal knowledge gets closer to walking out the door. Every quarter, the cost of manual execution compounds - in chargebacks, in write-offs, in working capital tied up in ghost inventory, in customers who ordered products you couldn't actually fulfill.
Your data isn't ready. It never will be - not by itself. The act of deploying an execution layer is what makes it ready. The system reads the messy data, surfaces the discrepancies, routes the decisions, writes the corrections, and learns from every cycle. Data quality improves as a result of execution, not as a prerequisite for it.
Stop waiting. Start with one workflow. See what it finds.
Ariel Palones is CEO and Co-Founder of Emerix, the AI Operating System for Supply Chain Execution.