AI won't replace data analysts, it replaces the slow parts of the job: writing the same dashboard spec from scratch, manually eyeballing a chart for outliers, building a cohort table in a spreadsheet. It doesn't replace the judgment call on what a number actually means for the business. gtm-skills' Data Analyst pack is 10 free, open source Claude Skills that cover the recurring analyst tasks - KPI design, anomaly detection, benchmarking, cohort tracking, growth prioritization, margin, shipping, inventory, returns, and a weekly readout. None of them can query your live warehouse yet. That's the honest gap, and it's coming.
"Will AI replace data analysts" is one of the most-searched questions about this job right now, and it deserves a straight answer instead of a marketing dodge. Here it is: no, but the job is changing under you whether you opt in or not.
The parts of the job that are genuinely at risk are the parts that were never the hard part. Writing a dashboard spec from a blank page. Staring at a chart trying to decide if that spike is real or noise. Building a cohort table by hand in a spreadsheet for the third time this quarter. Reading 80 support tickets to find the same 4 complaints. AI is already good at all of that, and pretending otherwise wastes your time.
What it's not good at, and won't be soon: deciding whether a benchmark someone found on Google is actually the right comparison for your business. Deciding which of three plausible growth levers is worth pulling given a team that's already stretched. Reading a room and knowing when a number that looks fine on paper is actually a symptom of something else. That judgment is the job. This pack is built to hand you back the hours the first list eats, so you have more of them for the second.
TL;DR
- AI replaces the mechanical parts of analyst work (dashboard specs, outlier scanning, cohort tables), not the judgment calls (what a number means, which lever to pull, whether a benchmark applies).
- gtm-skills' Data Analyst pack is 10 free skills: the-kpi-blueprint, the-anomaly-alert, the-benchmark-check, the-cohort-tracker, the-lever-finder, the-margin-builder, the-shipping-recovery-check, the-inventory-risk-scanner, the-returns-miner, and the-weekly-reporter.
- Install once with npx skills add sidchaudhary/gtm-skills.
- The honest gap: none of these query a live warehouse. Natural-language SQL against your real data is coming, not shipped.
- Intempt's Data Analyst agent already runs this against live event data today, no CSV export required.
How to install these in about 60 seconds
- Run npx skills add sidchaudhary/gtm-skills.
- Create a Claude project per skill - "Anomaly Alert," "Cohort Tracker."
- Paste the skill's SKILL.md into the project instructions.
- Start a chat with your metric data, benchmark source, or research material.
The 10 skills
10 skills covering the recurring analyst jobs that used to eat hours: designing a dashboard before touching live data, deciding whether a chart spike is real, checking a metric against a benchmark honestly, tracking cohorts, prioritizing growth levers with a quarterly plan, building a contribution margin stack, checking shipping recovery, scoring inventory risk, mining return concentration, and producing a weekly operating readout.
| Skill | What it does |
|---|---|
| 1. The KPI Blueprint | Turns "build a dashboard" into metric formulas, chart types, and alert thresholds |
| 2. The Anomaly Alert | Flags which points in a metric's time series are genuinely outside normal range |
| 3. The Benchmark Check | Scores one of your metrics against a stated benchmark, with an honest confidence read |
| 4. The Cohort Tracker | Groups customers by acquisition period and tracks retention or revenue across cohorts |
| 5. The Lever Finder | Assesses maturity and returns the top 3 growth levers plus a quarterly plan |
| 6. The Margin Builder | Builds a per-SKU or per-order contribution margin stack (CM1, CM2, CM3) from raw cost inputs |
| 7. The Shipping Recovery Check | Compares shipping charged against shipping paid, banded by order value and zone |
| 8. The Inventory Risk Scanner | Scores SKU stockout and overstock risk from on-hand units, velocity, and lead time |
| 9. The Returns Miner | Finds which SKUs have real return concentration and which coded reason is the cause |
| 10. The Weekly Reporter | Produces one weekly operating readout instead of scattered dashboards |
1. The KPI Blueprint
Turns a request for "a dashboard" into an actual spec: metric formulas, the right visualization type for each one, alert thresholds, and a section-by-section layout. This is the one to run before you touch live numbers, because a dashboard measuring the wrong thing cleanly is worse than no dashboard at all. Read the skill.
2. The Anomaly Alert
Takes a time series of one metric and flags which recent points are genuinely outside its normal range, using a stated trailing-average method instead of a gut read of a chart. It tells you where to look. It doesn't tell you why the point is there, that's still yours to dig into. Read the skill.
3. The Benchmark Check
Checks one of your metrics against a stated benchmark source and gives you an honest confidence read on how solid the comparison actually is. It won't invent an industry average it doesn't have, and it says plainly when it's leaning on general public knowledge instead of a source you supplied. Read the skill.
4. The Cohort Tracker
Groups customers by acquisition period and tracks retention or revenue across the periods that follow, returned as a cohort table. Use it to see whether newer customers are actually doing better than older ones, instead of a blended average that hides the trend either way. Read the skill.
5. The Lever Finder
Assesses business maturity and returns the top 3 growth levers worth pulling, plus a quarterly plan to act on them. For strategic planning and growth audits, not a metrics readout dressed up as strategy. Read the skill.
6. The Margin Builder
Builds a per-SKU or per-order contribution margin stack - CM1, CM2, CM3 - from raw revenue, cost, fee, and ad spend inputs, so you see which products actually make money after every variable cost instead of one blended margin number. Run it when ROAS looks fine but profit doesn't. It computes margin from numbers you hand over now; if you want the recurring dashboard that tracks margin over time, that's the-kpi-blueprint's job. Read the skill.
7. The Shipping Recovery Check
Compares what you charge for shipping against what shipping actually costs, banded by order value and zone, then tests your free-shipping threshold against your real order distribution. For when fulfillment cost is climbing or that threshold hasn't been recalculated since you set it. It isolates shipping only - COGS, fees, and ad spend stay out of it, which is what makes the number readable. Read the skill.
8. The Inventory Risk Scanner
Scores each SKU's stockout and overstock risk from on-hand units, sales velocity, and lead time, using a stated days-of-cover method you can check rather than an impression that stock "looks low." Use it to decide which SKUs are safe to promote and which to protect from a campaign about to drive demand at them. Read the skill.
9. The Returns Miner
Takes a returns or RMA export with coded reasons tied to SKUs and finds which products have genuine return concentration, plus the likely root cause behind each. For when returns are logged as a cost line and never mined for the fix underneath. Needs structured reason codes rather than free-text notes. Read the skill.
10. The Weekly Reporter
Produces one weekly operating readout from your exported performance, traffic, lifecycle, inventory, and support summaries: what changed, what likely caused it, what to do next. It summarizes across whatever you bring that week - for a deep read on one area, run the matching skill above and feed its output in here as an input. Read the skill.
The honest gap: none of these can query your live data
Every one of these 10 skills is a framework or a spec. You paste in numbers, a transcript, or a benchmark source, and Claude structures the analysis around it. None of them connect to a live warehouse and none of them can run a query for you. If your actual bottleneck is "I need someone to write the SQL," this pack doesn't solve that today.
Natural-language SQL querying against a live warehouse is the single missing piece, and it's the next thing we're building toward, not something we're going to oversell as already here. Until then, pair these skills with tools that already work against real inputs: report-generator for turning raw output into a shareable report, website-grader for grading a live site (Performance mode) or its AEO citation-readiness (AEO mode) instead of a pasted screenshot.
This is exactly the gap Intempt Analyze is built to close: plain-language questions answered against your actual event data, not a CSV someone remembered to export.
The order to build them in
Start with the-kpi-blueprint - decide what you're measuring and how before you start analyzing numbers. Add the-anomaly-alert once a dashboard is live, so you know when something in it actually needs a look. The-benchmark-check and the-cohort-tracker come next, once you have a baseline worth comparing against or a customer base old enough to track across cohorts. Save the-lever-finder for quarterly planning.
What to never hand over
Never let AI make the final call on what a benchmark means for a real business decision - the-benchmark-check tells you the gap and how solid the comparison is, a person still decides whether that gap is worth acting on given context the source can't see. Same with the-anomaly-alert: it flags which point is unusual, it doesn't know why, and treating a flag as a root cause is how you chase the wrong fix. And the-lever-finder prioritizes levers, it doesn't know your team's actual capacity or a commitment you made to the board last quarter - that call stays with a human every time.
When you outgrow the manual version
Intempt's Data Analyst agent runs this same job continuously: ask it a plain-language question and it detects the right report type - trend, funnel, cohort, path - renders the chart inline, and offers to build the segment, journey, or experiment that addresses what it found, without switching tools. The skill runs on what you paste in. The agent runs on your live pipeline.
Frequently asked questions. Answered.
No, not the role. It replaces specific tasks inside the role: drafting a dashboard spec, scanning a time series for outliers, building a cohort table, synthesizing 50 survey responses into themes. Those are real hours back. What it doesn't do is decide what a number means for your specific business, whether a benchmark is even the right comparison, or which growth lever is worth the team's time this quarter. That's still a human call, and it's the part of the job that was always the actual job.
