In buildEst. Q1 2027monitoringagentops

ops-watchdog

Watches a business signal. Ships a morning brief. Answers 'why did this move?' on demand.

ops-watchdog preview

§ 01 · What it is

Configure it to watch a signal — stock levels, review-site rating, competitor pricing, ad spend, warehouse throughput. It records daily snapshots, detects anomalies against your baseline, and delivers a one-page morning brief. Attached is an agent that can drill into any anomaly on request: 'why did five-star reviews drop this week?' → correlated timeline + likely causes + linked source data.

§ 02 · Money it saves

£8,000-£35,000/yr per watched signal
for operations managers, e-commerce founders, retail buyers, and marketing leads without a dedicated analyst
The math: One out-of-stock event on a top-10 SKU during peak season costs £2-8k in lost margin plus reputation. One competitor price move you miss for a week can cost 5-15% of category revenue. A single BI analyst headcount is £45-70k; this covers the 'daily anomaly-watching' 15% of that job at ~£100/mo of infra + LLM cost.

§ 03 · How it works

  1. 01

    Signal adapters ingest the raw data — Shopify inventory, Google Reviews, competitor price pages, ad platform APIs, a warehouse CSV export. Snapshotted to Postgres nightly with a small feature-engineering layer (rolling averages, deltas, seasonality baselines).

  2. 02

    Anomaly detection runs in the morning: statistical (Z-score against rolling baseline) plus LLM-narrated ('reviews dropped from 4.7 to 4.4 over 5 days, this is a real trend not noise').

  3. 03

    Delivers a morning brief by email or Slack: headline number, deltas, anomalies with severity, and a one-sentence recommended action per anomaly. Optimised to read in 60 seconds over coffee.

  4. 04

    Drill-in agent: reply to the brief with a question and the agent has full access to the snapshotted data + web-search for external context ('what launched this week in our category?'). Every answer cites its data source.

Stack

PythonPostgres + TimescaleDBPrefect (schedulers)Claude (anomaly narration + drill-in agent)Slack + Resend (delivery)Next.js dashboard (config only)

§ 04 · Why it beats what's out there

General-purpose BI tools (Metabase, Looker) are excellent for people who know what question to ask. Ops managers usually don't — they need something to *tell* them what to look at today. That's the gap.

What's out thereThe gap it leaves
General-purpose BI dashboards (Metabase, Looker, Power BI)Wait for you to open them. Assume you know what to look for. Great for scheduled reporting, bad for anomaly-first workflows. This inverts the model — it comes to you.
Custom analyst headcount£45-70k/yr salary plus overhead. Vastly overkill for a 5-person e-commerce team that just needs 'tell me when something weird happens.' This covers the boring 15% of an analyst's day for the cost of a modest SaaS.
Alert emails from source platforms (Shopify low-stock, Google review alerts)One channel at a time, no cross-signal correlation, no narrative. A 'low stock on SKU A' alert isn't useful without knowing 'and a competitor just went out-of-stock on the same category, so demand is about to spike'. This connects those dots.
Want this in your business?

I'm building this now. If it fits your problem, let's talk before it ships.

Follow along

Build notes and progress on this system will be posted in the Notes. Or check GitHub when the repo goes public.