Insights

Field notes from the studio.

Practical writing on data infrastructure, BI, and the operating systems behind compounding revenue.

AI-ready data

We scored every target account for AI maturity. Here’s what the scores got wrong.

A three-dimension scoring model that changed how a field team prioritized — and the four ways it was systematically wrong.

Ian Kessler · Oct 19, 2026 · 3 min read

Read article

Budgeted per run, billed per attempt: one agent run budgeted at ten cents climbs to a dollar fifty over five retries — illustrative

AI-ready data

The agent bill nobody budgeted: retries

Agent budgets are written per run. The bill arrives per attempt. Here’s where the gap comes from, and the three guardrails that close it.

Oct 16, 2026 · 3 min read

Read article

Leaving well

How to leave a client without breaking production

This summer I wound down a long client engagement on a GTM data platform. Nothing broke after I left. That wasn’t luck. It was eighteen days of handing it over one topic at a time.

Oct 14, 2026 · 6 min read

Read article

Two coupled version pins drifting apart: the local and CI runtimes stay above the minimum version a syntax needs while the production runtime drops below it, and the coupled config flag stays on

Leaving well

Two version pins drifted, and the nightly build went dark

A production data pipeline stopped compiling code that had worked for months. Nobody had touched it. Two settings that had to move together had quietly stopped agreeing.

Oct 12, 2026 · 2 min read

Read article

Same hours, different date: a report filtered on the ticket's opened date drops this week's logged hours, while filtering on the work date keeps them all

GTM data you can trust

The report filtered by the wrong date

Logged hours kept “going missing” from a weekly report. Nothing was missing. One date slicer was answering a different question than the one people were asking — and it took a week to notice.

Oct 9, 2026 · 2 min read

Read article

AI-ready data

Before you buy AI agents, fix your context layer

On a client engagement I built an AI-readiness score for an outbound team: 42 signals, a 0–100 score, a tier for every account. Picking the signals was the easy part. What made the score worth acting on was the boring layer underneath it.

Oct 7, 2026 · 6 min read

Read article

The easy twenty minutes: a stage definition is encoded in five places — raw ingestion, models and snapshots, monitoring, the CRM, and historical comparison — drawn as five rings around it

GTM data you can trust

How to change a funnel stage without breaking every metric you have

Adding one stage to the front of a sales funnel touches ingestion, models, alerting, the CRM and every historical comparison. Here’s the order that works.

Oct 5, 2026 · 3 min read

Read article

Before: Stage 1 meant a picklist moved and the leaderboard counted it anyway. After: Stage 1 means a meeting happened, checked against warehouse data, and 129 deals moved back to S0

GTM data you can trust

Your sales leaderboard is lying to you

We moved 129 “Stage 1” deals back a stage once we checked whether the meeting behind each one actually happened. The fix took a script. The hard part was the definition.

Oct 2, 2026 · 2 min read

Read article

One grant, three lanes: a warehouse feeds one least-privilege role, which feeds BI workbooks, reporting apps and an AI agent; revoke the role once and every lane closes

AI-ready data

Your AI agent just walked around your dashboard permissions

I ran a managed BI tool and our own git-backed reporting apps side by side on a client engagement. The BI tool won some rounds outright. The round that decided it was the one nobody plans for: AI agents.

Sep 30, 2026 · 6 min read

Read article

A dependency with good uptime versus a written-down asset: questions routed through one person on the left, the team connected to written references on the right

Enablement & handover

Your data platform shouldn’t need you in the room

Every model shipped, every number reconciled — and the engagement still could have left the client worse off. What handing over a data platform actually takes, including teaching Git to people who aren’t engineers.

Sep 23, 2026 · 4 min read

Read article

Same report, two paths: the governed path keeps all 1,000 records, the shortcut path silently drops 388 and still looks fine

RevOps & data quality

Green checks, wrong number: the bug your tests can’t see

The worst data bugs don’t throw errors. They return fewer rows than they should — and everything downstream looks perfectly healthy.

Sep 16, 2026 · 4 min read

Read article

Propose freely, execute never — the agent's allowed actions, a human review gate, and the actions that require a human to apply

AI & data governance

My AI agent fixed the error by altering the production table

It was told to stop on errors. It hit one — and resolved it with a schema change against production. Correctly, helpfully, and with no human in the loop. The fix is about capability, not instructions.

Sep 9, 2026 · 4 min read

Read article

Bars showing planning time staying flat as warehouse size grows, while the purpose-built model runs in one second

Data engineering

You can't buy your way out of a bad data model

A dashboard took twenty seconds to load. The warehouse was already oversized. Here is the fifteen-minute check that tells you when more compute will never help.

Sep 2, 2026 · 4 min read

Read article

Architecture

Propose Freely. Merge Never.

The architecture that makes an AI agent safe to run against production revenue data — and the three unglamorous decisions that make it work.

Aug 26, 2026 · 9 min read

Read article

Newsletter

New insights, in your inbox.

One short note when a new piece goes live. No spam, unsubscribe anytime.