JB Product
← Back
AutomationProduct ThinkingAIMarketing

Turning operational data into automated marketing

Published on September 16, 2025·2 min read·Lire en français
Share:X / TwitterLinkedIn

We produce a lot of content without realizing it.

In our delivery platform, drivers already upload a POD (Proof of Delivery) for every job: a photo of the site, plus the weight transported, travel time, and client information.

The question I asked myself: why create marketing content manually when all the raw material is already there?

The starting observation

Every delivery produces a real photo of real materials on a real construction site. That's exactly the type of content we need on Instagram for a building materials company: concrete, grounded in daily operations.

But manually turning hundreds of PODs into Instagram posts is repetitive work that nobody will do. Or that nobody will do well, for long.

What I built

The feature works in four automatic steps:

  1. POD retrieval — the system identifies newly validated deliveries and extracts the photo and metadata.
  2. Material recognition — an AI model (Claude) analyzes the photo and identifies what's being delivered (PVC, tiling, insulation, etc.).
  3. Caption generation — delivery data (route, weight) enriches the automatically generated caption.
  4. Scheduling and publication — the post is queued and published to the company's Instagram account.

Instagram post management interface automatically generated from PODs

What's visible in the interface: 456 scheduled posts, 30 published.

What this changes in practice

  • Social visibility with no operational overhead — field teams keep doing their jobs normally. Content generates in the background.
  • Authentic content at scale — no staging, no stock photos. Real deliveries, real job sites, real materials.
  • Repurposing an existing workflow — the POD already existed for operational reasons (proof of delivery, invoicing). It now does additional work without any change to the driver's process.

The real product thinking

The AI in this project is an implementation detail. It's not what makes the feature interesting.

What's interesting is the question that preceded it: what data are we already producing, and what else could we do with it?

Many product opportunities don't require new data sources.

A POD is a proof of delivery. It's also a job site photo. It's also a performance indicator. It's also potential marketing content.

It's not the same data depending on who's looking at it. So I leaned into recycling :)

Comments