Foundation First
Not in a warehouse. Not in a pipeline. In an inbox.
That's the data infrastructure underneath a lot of companies running AI initiatives right now. Most don't have an AI problem. They have a data problem they're hoping AI will solve. Foundation First is the practice of fixing that before you build anything on top of it.
Decisions get made on calls and leave no record. Institutional knowledge walks out when people do. The report everyone trusts is one person's spreadsheet, rebuilt by hand every Monday. Every AI program that stalls stalls here, and the starting point is always the same work: mapping the decisions, tracing what data each one needs, building capture points that never existed.
Nobody puts that work in the press release. It's not interesting until it works. But without it, AI doesn't give you better answers. It gives you wrong answers you can't argue with, generated faster.
Five of the Eight Domains
Domain 01
Name the job before you go shopping for a tool.
Leadership wants AI, but there's no target workflow. Just a mandate to "explore."
One named workflow, measured today, that AI would clearly speed up or replace.
Domain 02
AI can only use what's actually captured.
Your team pulls spreadsheets before every meeting because the systems don't talk to each other.
The data needed for most decisions is captured automatically. No exports required.
Domain 03
Data can be sitting right there and still be wrong.
Someone always has to fix it up first, and it is always the same someone.
Data holds its shape from source to decision. The team trusts it without checking behind it.
Domain 06
If decisions happen in meetings, AI has nowhere to plug in.
Key operational calls happen in someone's head or a hallway conversation, not in a system.
Decisions run on defined inputs and documented logic, which is something a model can actually plug into.
Domain 07
The tool is almost never the thing that failed.
New systems get used for three weeks, then quietly abandoned. Change management is a recurring post-mortem item.
The team keeps using new tools because the tools solve something, not because adoption is being tracked.
Score Yours
Five questions covering five of the eight domains, about two minutes. The auto-reply tells you what's blocking you and what to fix first.
Take the Assessment →The Foundation-First Sequence
Stage 01
Map every key operational decision against the eight domains. Find what data it needs, what's missing, and what's never been captured at all.
Stage 02
Design the data flows, capture points, and decision logic that have to exist before anything automated is layered on top.
Stage 03
Build in production, not a prototype. The infrastructure that actually runs the decision, with a named internal owner accountable for it.
Stage 04
Run it without outside support. Anything works while the help is still in the building. What counts is what happens the month after they leave.
Foundation-First Assessment
Five questions, five of the eight domains. Auto-reply delivers your score bracket with a plain-language interpretation and what to fix first. The full assessment covers all eight.
Foundation Gaps
The infrastructure isn't ready. The auto-reply covers what to fix first, and why skipping this stage is how AI initiatives fail.
Partial Foundation
Some pieces in place, specific gaps to close. Usually one domain is bottlenecking everything downstream.
Strong on these five
Solid across these five. The three this doesn't cover (infrastructure, security, ownership) often decide whether a build ships. The full assessment covers all eight.
One question per domain, five of the eight. Scored 1–3. Takes two minutes.
Free, no account needed. Results come by email.
About
Twenty years at Yahoo and Verizon Media, including the team that shipped the first rich-media ad on the Yahoo homepage and, eventually, a global team behind $500M+ in annual digital advertising revenue. Then VP of Engineering at 84.51°, Kroger's data company, running a 150-person org building advertising infrastructure on data and AI.
It started earlier than that. I was online in 1992, back when that meant Gopher and FTP servers and raw text. In early 1993 I got my hands on Mosaic, on an X Terminal in a university computer lab, and taught myself HTML. A few years later I heard about a Dallas startup called AudioNet that was putting sports and radio on the internet. I got a meeting, showed up with a webpage sketched on a notepad, and started in March 1996. AudioNet became broadcast.com. Yahoo bought it for $5.7 billion.
Now VP of Product Solutions, leading AI and digital transformation from the inside.
What I actually build is the foundation that makes AI work: capture points for decisions that currently happen on calls and leave no record, formats that work across systems instead of inside one person's spreadsheet, inputs someone has actually verified before anyone touches a model. The gap between AI that changes how a company operates and AI that lives in a slide deck is almost always in the infrastructure, not the model. I've built that foundation before. I'm building it again now.
I still write code, because it's the fastest way to find out what I don't actually understand yet.