Acelerar · Opportunity Assessment Lead: Thomas Kiran · KSA

Sourcing 15,000 product images for a Blink grocery store

How much of this job runs on an AI pipeline versus needs human hands — so you can price it and commit to the timeline with a real number, not a guess.

70–85%
of total effort is automatable end-to-end
2 / 2
live test: sample barcodes resolved to clean, open-license images
1 gate
a 100–200 barcode sample sets the price & the 5–10 day promise
The Ask

What the client actually needs

A Saudi grocery retailer has products already loaded in Blink but no images. They want us to find, resize, rename and hand over the correct image for each SKU. They do not have the images — we source them online, guided by universal barcodes.

Platform
Blink · blinkco.io
Market
Saudi Arabia (+966)
Catalogue
~15,000 images
Type
Grocery · food · personal care
Deadline
5–10 days
Note
Budget-conscious
The Pipeline

Step by step: machine vs. human

Grocery and personal care is the best-case category for automation — Open Food Facts is a free, barcode-keyed, open-license image database built for exactly these products. That's the engine we tested live.

Fully automated Human-in-the-loop Client / out of scope
01
Ingest catalogue mapping
Barcode ↔ SKU ↔ product name — parse the export from their Blink system.
Automated
02
Source image from barcode
Open Food Facts + UPCitemdb / Go-UPC fallback. The core engine — hit rate is the one unknown.
Automated
03
Recover the misses
Image search on name + barcode; AI proposes candidates, a person confirms the right one.
Human-assist
04
Verify correct variant
Vision AI matches image to "Amul Milk Powder 500g Vanilla." ~90% auto; low-confidence flags go to a human.
Automated*
05
Resize & format to Blink spec
Batch script once we know Blink's required dimensions, format and max file size.
Automated
06
Rename to product name
Driven entirely by the step-01 mapping.
Automated
07
QA & skip list
Script flags unfindable / watermarked / ambiguous items; a person decides skip vs. request-link.
Human-assist
08
Upload to Blink
Client's own team bulk-imports — unless they give us API access. See the fork below.
Depends
Step 08 · The Fork

What if they hand over Blink API access?

This single decision changes the whole shape of the deliverable — from "a folder of images" to "an AI pipeline that pushes straight onto their live products."

Scenario A · as described

No API — client uploads

We deliver a folder of correctly named, resized images. Their team runs Blink's bulk import. Clean handoff, but the loop ends on their side.

Upload is out of our scope
Pitch this
Scenario B · API access

We push via Blink's API

Map image → product by SKU, then POST to Blink's media endpoint. Source → verify → resize → rename → live, with no manual upload on anyone's side.

Near-100% automated, end to end

Confirm before promising B: (1) does Blink expose a product-image endpoint (likely partner-tier, their team asks Blink support) · (2) auth model + write rate limits for 15k items · (3) exact image spec, so step 05 outputs it right the first time.

The Honest Ceiling

What AI can't do here

No conjuring images that don't exist. These are the buckets that land in the human queue — and why the sample test matters.

Near-identical variants

500g vs 1kg, flavour differences where the packaging looks the same. Vision helps but isn't perfect — human eyes finish these.

Local SKUs with no online image

Regional Saudi brands with zero web presence. Nothing to source → skip or ask the client for a link (they already agreed).

Rights on web-search fallbacks

Open Food Facts images are clean and licensed; anything pulled from general web search needs a rights judgment call.

The One Gate Before Quoting

Measure it, don't guess it

The whole price and the 5–10 day feasibility hinge on one number: what share of their 15k SKUs are mainstream (auto) vs. local (human). We don't estimate it — we test it.

Step 1
Ask Thomas for 100–200 sample barcodes
Step 2
Run them through the pipeline in an afternoon
Step 3
Measured hit rate → fixed price + honest timeline