AI Analytics Architect
The title is new. The work is not: specify what correct means, build it, test it, publish the evidence.
Open to analytics manager, lead, or principal roles where reporting has outgrown trust, and to conversations about roles that don't exist yet. Seattle or remote.
LinkedInlinkedin.com/in/aaron-robbins-6b25a2136 GitHubgithub.com/RobbinsAnalyticsI have built teams, led programs and stayed hands-on with analysis. The throughline is understanding the decision, defining the numbers and making the system useful.
BI management
ATG Stores (Lowe's)
I proposed the BI function, hired its foundational team and managed six analysts and developers. When accounting needed a partner, I hired for accounting judgment as well as technical skill.
Analysts took ownership of their stakeholder relationships.
Program leadership
Philips Healthcare
Reorganizations brought competing analytics backlogs together. I made the tradeoffs visible: effort, expected results, sites served and cost. Stakeholders could challenge the assumptions, and unresolved priorities went to the sponsor.
The result was one agreed queue for the work.
Hands-on analysis
Philips Healthcare
When quality notifications spiked, I traced most to label misprints caught and corrected on the line. I challenged what the headline measure was telling leadership.
Management accepted the recommendation to focus improvement work on failures with meaningful cost or risk.
AI Analytics Architect
Cascadia
For Cascadia Revenue Assurance, a portfolio build using synthetic data, I set the rules and expected answers before directing the AI build.
Its published results are checked through two separately implemented calculations. Responsibility for what those numbers mean remains mine.
The job has three moves. None of them is writing the code.
Business rules
Turn a business need into a precise spec: the data model, the measure definitions, the matching rules, and the invariants that must hold.
Correct is defined before anything is built, so there is something to check the build against.
Delivery
AI tools or an engineer do the building; either works. The job is to aim the build at the spec, not to pour the concrete.
That leverage is why one person now covers ground that used to take a team.
Evidence
Nothing ships on trust. Every number is re-derived a second, independent way and reconciled against source, and a check that disagrees stops the publish.
Trust what you can test.
Separate requests often point back to the same underlying system. A sales report, a service dashboard and a finance review may all describe the same customers, products, locations and dates.
Define those shared building blocks once, including what each means and how records connect, and each report can use the same foundation.
The architectural work is recognizing that overlap, deciding what should be shared and where differences matter, then giving the build a structure that can be checked.
I do not hand-write the code, and that is a choice. There are plenty of people who write code well. Specifying and verifying AI-built analytics inside a business domain is the rare skill, and it is the one this site is built to demonstrate.
01
Built the Open Quality Notification reporting that eight factories ran on. I built the original myself, before there was a development team, sourcing from SAP directly so nothing was copied between workbooks. It scaled from one site to 8, cut about 32 hours a week of manual work, brought average closure time down 5 days, and reduced the notifications aging past 30 days by 20%. Refresh went from weekly by hand to hourly during business hours, on a cadence set by how the work actually runs rather than by how often we could run it.
02
Gave three plants one set of numbers. The monthly operations report pulled from Murrysville, Böblingen and Shenzhen, each calculating the same 12 KPIs differently. I ran the initiative end to end, surfaced 6 definition misalignments and reconciled 4 of them myself, and delivered one automated Power BI report on aligned definitions. It removed about 36 hours of manual compilation a month and moved the reporting cycle from the 15th to the 5th.
03
Governed data at enterprise scale. I standardized 200+ Critical Data Elements and common data model definitions across a global manufacturing platform, enforcing semantic consistency across systems that had none.
04
Eliminated 100+ hours a week of manual reporting. I moved disconnected spreadsheets and point-in-time extracts into automated, governed dashboards.
05
Took $80K a year out of the reporting estate. I cataloged every pipeline under the Operations umbrella, traced each to the reports it fed, measured what people actually used, and retired the dead pipelines and over-frequent refreshes. No report was moved or deleted.
06
Founded a BI function. I proposed the department at Lowe's/ATG Stores, hired its foundational team, and managed six analysts and developers through the build of the data warehouse, the KPI definitions and the stewardship model.
The split below reflects the human and AI roles described in In Practice.
Let’s talk
Comfortable as the person building it, or as the person who stands up the function and hires the team. The standard doesn’t change either way.