"The illusion of an answer is more dangerous than no answer at all."
On chat-with-data & semantic layers
WIP · v0.18 · 09.03.26
"The illusion of an answer is more dangerous than no answer at all."
On chat-with-data & semantic layers
"Obsessed with rigorously connecting what I'm building to business value, by working backwards from understanding the business."
Most data work asks 'is this elegant?' — Nick asks 'is this the right question?'
"I don't buy into the logical fallacies of the experts or the majority in the slightest."
First principles win
"A data pipeline's time-to-value was never prolonged by writing joins, but rather by not understanding the business."
Where data value actually comes from
"I don't have grumpy days at the office, or days I let my standards down."
Consistency as a differentiator
"So much human capital wasted moving numbers around to make a board deck sound better, regardless of true value to the business."
On where analytics effort goes to die
"Not everyone is capable of being a data analyst. Nor does everyone want to be. And nor should they."
— Nick, on the consumption layer. see source
The rule it setsHand people a tool and they use it the way it fits, not the way you meant. If every shape fits in the square hole, that's where it's gonna go. Know who's really using it before you ship.
Quickest read first, deepest last. Jump anywhere; each section stands alone.
In Nick's words — lightly edited for clarity. The full transcripts live in the Q&A.
I'm at heart a positive, optimistic, energetic, extroverted person who gets energy from being around people and solving problems. Hard to talk about myself without mentioning real ADHD-hyperfocus tendencies. I take a new problem and work at it until I find a solution. Serious about the mission, but myself: mixing in positive energy and awkward jokes that are often more confusing than funny.
Where I separate myself in data and analytics: I'm obsessed with rigorously connecting what I'm building to business value, by working backwards from understanding the business. A lot of smart people build elegant solutions stakeholders love; that doesn't mean they're providing real business value.
On the negative side, I tend to over-engineer, to build for the abstraction too much. Things that aren't complex to me can be overly complex to others, and I don't always predict that well. Sometimes there's a simpler path to value.
And I don't care much what the trends or "consensus" best practices are. Not overtly contrarian, but I don't buy into the logical fallacies of the experts or the majority in the slightest.
Three years architecting, leading, and managing the data modernization of a 300+ location automotive collision-repair operator across eight workstreams (Manager at SDG). Client anonymized, outcomes as delivered.
Cloud data platform
On-prem warehouse → cloud-native. Near-live, under 15 minutes, for nearly every flow.
Azure · Snowflake · dbt · Qlik
Cash Application automation
~97% of electronic payments auto-matched. No more manual allocation.
UiPath · Snowflake External Functions · dbt · Qlik
NetSuite accounting integration
Replaced a packaged connector with a pipeline finance owns.
Snowflake External Access · NetSuite · SOAP · Azure Data Factory
UKG HRMS event-driven pipeline
Turned a system with no push API into near-live HR events.
Azure Durable Functions · Snowflake · UKG · FreshService
Central Review estimate-scoring API
Every estimate scored in under 15 seconds, before a carrier sees it.
Azure Functions · Snowflake · Streamlit · Qlik
Next-Best-Action conversational analytics
Operators choose where to explore, quality data analysis built in.
Snowflake · ML anomaly + trend detection · multi-agent LLM · Qlik
Insights Hub & KPI analytics
Your top 3, bottom 3, and where you rank in one screen.
Qlik · Snowflake · normalized KPI store
Estimate Recommendation Engine
Learns from history to fix estimates before they're sent.
Python · association rules · anomaly detection · Snowflake
Three working side projects where I'm actively exploring vibe coding and AI-assisted development flows.
A local-first recurring-event tracker built to test one idea: that lists, habits, tasks, and timers are all the same primitive viewed differently. Everything is one event-sourced items row; behaviors drive logic, labels drive display. Less a finished app than an exploration of how far a few primitives stretch before complexity is forced.
A local-first desktop tool to capture work items across Azure DevOps and Jira in seconds: hotkey, voice, or paste. A half-sentence becomes a fully-fielded ticket via a Claude pipeline with per-field confidence. The deeper experiment: increasing both the speed and quality of vibe coding through detailed documentation of requirements and architecture decisions.
Turns Claude Code from a coding tool into a stateful project manager. One bootstrap command and 11 questions render a project skeleton: up to 19 custom slash-command skills, a versioned doc spine, and hooks enforcing consistency every turn. Not "can an AI write code?" but "can an AI run the project?"
Six industry trends, reacted to from first principles — the short version. Key point: the illusion of an answer is more dangerous than no answer at all.
Pulled from
"Not everyone is a data analyst, wants to be a data analyst, or is capable of being a data analyst."
Not built (yet). Five architectures I've worked through end to end, with the business case already made. Three stack into one data platform: owned ingestion, an agile scorecard, a best-/worst-case ROI engine. Leaner bets on PR-review throughput and cross-system workflow sit beside them. Open any for the full thinking.
Change a key table and you can’t see the downstream blast radius — a known unknown.
Column-level lineage makes the before/after impact analysis tractable — one AI-assisted pass.
Review pipelines you’re only adjacent to, with confidence — not chasing every unknown.
iPaaS connectors don’t fit odd sources, so custom pipelines drift into a dozen flavors.
A portable dlt + field-level CDC pattern standardizes the raw→staging layer.
Agents and humans both build fast on one consistent base — capacity stops being the bottleneck.
Multi-site scorecards are always in flux, so the data team plays whack-a-mole with logic.
Marts derive from a field-level change log — definitions and anchor dates become config.
Re-base a metric and restate years of history in one run, not a pipeline project.
Roadmaps chase line items that feel urgent but can’t actually move the P&L.
Ride the KPI tree to estimate each bet’s best-/worst-case P&L impact — days, not months.
A fast ROI range before you commit teams to the wrong KPI.
Approvals hop systems and stall at the seams; the process is capped by the vendor.
One activity schema on the lakehouse orchestrates human-, automation-, and agent-owned steps.
One surface for enterprise workflow — stitch the cross-system processes vendors won’t.
The full interviews, by theme. Direct quotes that support the page content. Every chatbot citation links straight here.
A chat trained on the site content, answers referencing the Q&A. Always in the top bar.