Identify a Part from a Photo with AI on WhatsApp

To identify a part from a photo with AI, the customer sends an image over WhatsApp, a vision model describes what it sees and compares it with the photos and specs in your catalog, and the system returns the most likely part numbers for your team or the customer to confirm. It isn't magic, and accuracy depends heavily on your catalog.

Why spare-parts sellers care

Anyone who sells replacement parts knows the scene: the customer doesn't know the part number, describes it as “that threaded pin thing”, and a salesperson burns ten minutes trading messages. A photo settles much of that conversation, and AI cuts the time your staff spends hunting through the catalog.

How the flow works, step by step

1. The customer sends a photo (plus text or a voice note, if they like) to the company WhatsApp.
2. The system checks that the image is usable: sharpness, lighting and a visible part.
3. The vision model extracts features: shape, thread type, number of holes, apparent material, relative size.
4. Those features are compared with the catalog, using the photos, descriptions and dimensions already stored.
5. The system replies with the two or three likeliest options, each with a reference photo and part number.
6. The customer confirms, or the case goes to a salesperson when confidence is low.

What the AI sees well and what it doesn't

Works well

• Parts with very distinctive shapes: fittings, gears, bearings, caps and flanges.
• Photos with the part on a plain background, good light and a coin or ruler for scale.
• Catalogs with several photos per item and standardized dimensions.

Works poorly

• Near-identical parts that differ by millimeters or coating (same-head bolts in different lengths).
• Blurry, dark photos, or parts that are dirty and worn.
• Items whose identification depends on the source machine, not just the shape.

Accuracy limits: be honest with the customer

The system should treat its result as a suggestion, never a certainty. For low-value items, confirmation can be automatic; for critical, expensive or safety-related parts, a person validates before the order.

Ask for extra details when needed: equipment brand, machine model, the main measurement. Photo plus text usually gets far better results than the photo alone.

What your catalog needs to have ready

• A unique code per item, with no duplicates.
• A standardized description: type, material, dimensions, application.
• Two to five photos per item, from different angles.
• Cross-references and old or competitor part numbers, where they exist.
• Stock and price available through a system or spreadsheet.

A well-structured digital catalog is the foundation: without organized data, the AI just guesses. It is common for the biggest job in the project to be cleaning up the records, not configuring the model.

How to test before going live

Gather 50 photos from past real orders where your team already knows the correct part number. Run them through the system and count how often the right item came first and how often it landed in the top three. Those two numbers tell you whether the pilot is ready and which part families need more photos.

Connecting identification to customer service and the catalog is part of automation and AI work. Start with one product line that is well documented and expand gradually.

Frequently asked questions

Can AI identify any part from a photo?

No. It does well with distinctive shapes and a well-kept catalog, and errs more with items that look alike. Treat the result as a suggestion.

Do I need my own app for this?

Not necessarily. Photos can arrive directly on your company WhatsApp, which customers already use, with the flow connected to your catalog.

How many photos per item does the catalog need?

Usually two to five, from different angles on a clean background. Look-alike items need more photos and measurements in the description.