stostr

Case study · Homemade food · Katy, Texas

Radha's Kitchen: from social posts to local orders

A home kitchen making 15-minute curry sauces had great food and loyal regulars, but no way to tell whether posting online brought in a single order. We set up the posting, the ordering page and the tracking, so every week's effort can be measured.

Business
Radha's Homemade, Katy TX
Goal
More online orders, locally
Started
6 October 2026
Status
Live, tracked daily

See Radha's customer page →

The problem

  • Posting on Instagram and Facebook with no idea which posts, if any, led to orders.
  • An ordering page pointing to a channel the business didn't use, with a price list that went out of date.
  • Checking numbers meant clicking through Meta's tools by hand, so it rarely happened.

What we set up

  • A weekly post calendar: drafted for Radha, approved by her, scheduled ahead.
  • A clean customer page: the menu, one-tap ordering, and an assistant that answers common questions.
  • Automatic daily tracking of views, reach, followers and order-link taps.
  • A weekly review that proposes one change to try, for the owner to approve.

What we're learning

  • The farmers-market teaser was the best post of the first week, and Facebook carried it: local, event-based posts beat recipe posts.
  • Views are growing but taps on the order link haven't started yet, so the next tests target making ordering easier.
  • In-person tasting at the market is the first test of turning attention into sales.

Live results

The dashboard, updated daily

This is the same tracker the business uses, published read-only. Early numbers are small on purpose: it's a cold start, measured honestly.

Growth case study

From posts to orders

Radha's Kitchen sells to local families in Katy. The goal is more online orders. Each week an AI agent plans and drafts the social calendar, Radha approves it, it's scheduled on Instagram and Facebook, and these numbers check whether it turned into orders. Tracking started 6 October 2026.

PlanAI→DraftAI→ApproveYOU→ScheduleAI→MeasureAI→LearnAIYOU

AI: an agent plans the calendar, writes posts, schedules them and proposes the next experiment. YOU: a human checkpoint. Measuring is automated: every morning a script pulls Instagram and Facebook numbers from the Meta Graph API into the tracker. Orders are still logged by hand.

Day 1 of tracking

Distribution

–

posts published on Instagram & Facebook

Views

–

Meta views on those posts

Order-link clicks

–

taps on the order link

Followers

–

Instagram + Facebook

The funnel: do posts turn into orders?

Posts

–

–views per post

Views

–

–click rate (clicks ÷ views)

Order-link clicks

–

–clicks that ordered

Orders

–

Online orders so far: the finish line

–

Experiments

One change at a time. The hypothesis is written before acting. Before and After are daily averages from the tracker: the window before the change vs the same window after it.

#Hypothesis and metricBeforeAfterChangeStatus
Loading…

At this volume, results are directional, not statistically significant. The verdict (Won, Lost or Inconclusive) is a human call, made once the window closes.

Numbers update daily from the kitchen's tracker.

How it works For the technically curious

Measure: Meta Graph API → Google Sheets

  1. A read-only, development-mode Meta app holds a Page access token for the business's Facebook Page and linked Instagram account. The token lives in the script's private settings, never in code.
  2. A Google Apps Script runs daily at 6 am. It reads Instagram media insights (views, reach, total_interactions), bio-link taps (profile_links_taps) and Facebook post insights (post_media_view, post_total_media_view_unique, post_clicks_by_type).
  3. It sums by local date and re-writes the last 7 days each run (posts keep gaining views), never touching the hand-entered Orders column, and logs every run.
  4. A Dashboard tab of formulas is published read-only; this page fetches it in the visitor's browser and draws the charts.

Learn: a weekly agent, with a human approving

  • Every Monday a scheduled Claude task reads the tracker, writes the weekly read shown above, and proposes one experiment with "Approved" left blank. The owner approves, sets the start date and gives the verdict.
  • Before and after for each experiment are calculated from the daily numbers automatically.

Answer: a grounded question assistant

  • Facts live in one structured file (dishes, ordering, pickup, events). Rules answer common questions instantly; anything else can go to a serverless function that calls Claude with those facts only.
  • Refunds, complaints, allergies and large orders always go to the owner by phone, never to the AI.

Want this for your business?

Every project starts with a free 20-minute ops check. Email me a line about your business and I'll reply the same day.

Email me