Most marketers I know are paying for six dashboards and reading two of them. Triple Whale for blended ROAS. Northbeam if Triple Whale is too expensive. Looker for the executive summary that gets exported to a slide nobody opens. Klaviyo's reporting tab. GA4. A spreadsheet somewhere with the actual numbers because none of the dashboards agree.
The notifications come in every morning. Spend alerts. Anomaly warnings. Weekly digests. After a few weeks, nobody reads any of them.
The strange thing is what those dashboards are actually doing. They read a few APIs and format the answer. That's the whole product. Pretty charts on top of API calls. Claude Code reads APIs and formats answers natively — it's literally what it does. So a year ago I stopped paying for dashboards and built a three-layer reporting stack out of skills. One layer per question I actually ask. This is what it looks like.
The Three Layers
Reporting isn't one job. It's three jobs at three different cadences, and conflating them is why dashboard subscriptions feel like noise. There's the daily check-in. There's the weekly client report. And there's the period comparison — the analysis we run when something changes and we want to know if it worked.
I built one skill per layer. Each is a few hundred lines of orchestration on top of the same handful of APIs.
Layer 1 — The Daily Recap
The first layer answers one question: what happened across all channels yesterday.
It's a Claude Code skill called yesterday-recap. I run it from the terminal with a client slug. It auto-detects which channels are wired for that client — Meta, Google, Shopify, Amazon — and pulls each one in parallel. It dedupes the Amazon FBM orders that sync into Shopify so revenue isn't double-counted. It computes blended ad spend, revenue by channel, and blended ROAS. Then it prints a clean recap to the terminal.
End-to-end it takes about thirty seconds. The output looks like a Slack message, not a dashboard.
Cobi — Sunday May 11
Spend $487 (Meta $312, Google $175)
Revenue $2,341 (Shopify DTC $1,820, Amazon $521)
Orders 34 (DTC 27, Amazon 7)
ROAS 4.81x blended
AOV $68.85
Notes: Meta CPM down 18% vs trailing 7-day avg.
This replaces the morning ritual of opening four dashboards and copy-pasting numbers into a notes app. It also replaces the Slack channel where everyone shares screenshots of yesterday's spend. One command, every client, takes less time than making coffee.
The skill knows which channels each client actually has, so it skips what it should skip. A client with no Amazon doesn't get an Amazon section. It's not a one-size-fits-all dashboard pretending every account looks the same.
Layer 2 — The Period Comparison
The second layer answers a different question: what changed when we did X.
This is the analysis nobody wants to do manually but everybody needs. Did the relaunch work. Did the new creative refresh move the number. Is this month better than last month and by how much. The skill is called compare-periods and it takes two date windows for the same client, pulls every channel for both windows, and outputs a markdown report comparing them.
The pulls are the same APIs the daily recap uses. Same dedupe logic. Same blended ROAS math. The only difference is two windows instead of one, and a written-up TL;DR at the top that frames the deltas in plain English.
The report writes to ~/Projects/clients/<client>/reports/<name>.md, which means it's automatically filed under the right client folder, version-controllable, and shareable as-is. A client asks "did the sun hoodie relaunch work?" — I run the skill with the launch date as the seam, paste the headline numbers into Slack, and send the full report if they want the details.
This is what replaced the manual Excel ritual. The one where you export Shopify orders, pivot by date, export Meta spend, paste it into a second sheet, and spend an hour matching things up before you can answer a question that should take five minutes.
Layer 3 — The Weekly Client Report
The third layer answers the question every client asks on Monday morning: what happened last week and what are we doing this week.
The skill is called monday-report. It pulls a DTC and Amazon revenue breakout, the daily breakdown table, Meta ads performance, product mix from Shopify, and renders it all into a branded HTML email. Then it sends via Resend, which is a transactional email API that costs basically nothing at our volume.
The output isn't a dashboard link. It's an actual email the client opens. Subject line, copy, tables formatted for mobile, my actual notes on what we're doing next. It looks like a thoughtful human wrote it because a thoughtful human did write the framework — the skill just fills in the numbers and the tables that don't change week to week.
This is the layer that replaced the freelance weekly reporting platform — roughly $200-300 per client per month for a dashboard most clients would never log into anyway. The version we send is better because it's an email, not a login.
Why Three Layers Instead of One Mega-Report
The obvious objection is why not build one big dashboard that does all three. Each layer answers a different question at a different cadence, and combining them makes every layer worse.
A daily recap that has to render product mix and Meta creative breakdowns is slow and noisy. A weekly report that updates in real time is anxious-making and harder to debug. One layer that reads four APIs and formats one answer is fast and easy to debug. One mega-report that reads four APIs three times a day is overkill and breaks in ways that take an hour to find.
The other reason is the broader AI-native agency philosophy — AI builds the tools, and the tools should be small and composable. Three skills that each do one thing well chain into bigger workflows. A monolithic dashboard doesn't compose with anything.
What This Stack Replaced
Here's the actual cost comparison, with the SaaS numbers being what we used to pay before the switch.
| Tool or service | Cost per month (before) | Replaced by |
|---|---|---|
| Triple Whale or Northbeam | $300-500 | Layer 1 (daily) + Layer 3 (weekly) |
| Freelance weekly reporting platform | $200-300 per account | Layer 3 |
| Manual Excel exports for WoW analysis | ~2 hrs/wk per account (time cost) | Layer 2 |
| Across a book of accounts | scales with each client added | ~$30-60/mo in API + Resend fees, flat |
The $30-60 is API fees on the Anthropic side (the actual model calls), plus Resend for sending the emails, plus the marginal cost of the channel APIs we already paid for to run campaigns. Everything else was rented dashboard infrastructure on top of those same APIs.
I wrote a longer breakdown of the full SaaS replacement if you want the full picture across creative, landing pages, SEO, and the rest of the stack.
What This Stack Misses
Being honest about the tradeoffs matters more than pretending the custom version wins everywhere.
- No real-time pixel attribution. Triple Whale still wins on multi-touch attribution modeling. Our stack uses Shopify source attribution as the source of truth — good enough for ROAS, not for attribution modeling.
- No fancy dashboards. Outputs are text, markdown, and email. If a client expects to log into a portal and see charts, this stack will disappoint them.
- Setup time per new client is around 30 minutes. Wiring API credentials and adding the slug to config files. In Triple Whale you connect a pixel and you're done. The custom stack pays this cost upfront and saves on monthly fees.
- Doesn't replace a real BI tool for cohort analysis. If you need to slice retention by acquisition channel across a portfolio, you want something else.
Who This Is For
Agencies and operators where the per-account dashboard cost is starting to eat margin. Solo operators with multiple accounts. Brands with a dedicated marketing lead who hates dashboards and just wants the answer. Anyone who has ever opened Looker, found the dashboard was stale because someone changed the underlying field, and closed the tab.
Not for: enterprise BI, finance teams who need audit-grade reporting, anyone who needs an executive dashboard with role-based access. Different tool for a different job.
The Takeaway
The reporting stack is the most replaceable part of the marketing SaaS bill. It's also the one most people don't replace because it feels harder than it is. The actual work is small — four API integrations, three orchestration skills, one email sender. The rest is paying for charts.
The three skills above sit in the Reporting Command Center pack on operatorstack.app — $99 for the full pack, or $497 for the Operator's Stack bundle with all six packs and the API crash course that walks through wiring Claude Code to your channels.
Or hire Clare Digital and we'll run them across your accounts.
Q: Can I run this stack without paying for an API?
Mostly yes. Meta's Marketing API and Google Ads API are free with rate limits. Shopify's GraphQL Admin API is included with your store plan. Amazon Ads and Selling Partner APIs are free but require approval. The only real cost is the Anthropic API for the model calls, which runs $30-60/mo at our usage, and Resend for the email send (about $20/mo for the volume we run).
Q: How long does it take to set up per new client?
About 30 minutes if the client already has API access set up for their channels. Most of that is adding the client slug to our config files and verifying the pulls return the right numbers. If they don't have API access and we need to request developer access, it can stretch to a few days waiting on Meta or Amazon to approve. Triple Whale handles a lot of this for you — that's the tradeoff for the monthly fee.
Q: What happens when an API changes?
The skill breaks, you fix it, you move on. Meta versions the Graph API every few months and Shopify deprecates GraphQL fields occasionally. The fixes are usually a few lines of code. SaaS dashboards eat that maintenance on your behalf — at 5+ clients the math still favors the custom stack by a lot.