AI Vision · Case Study Computer Vision in production workflows

Reading water meters with computer vision.

For a South African water utility, 10decoders built an AI Vision solution that reads meter values directly from a photo or video — turning a slow, manual field task into an automated, auditable digital workflow.

Client
South African water utility
Domain
Utilities · Metering
Capability
AI Vision · OCR
Stage
Demo / Proof of concept
Overview

Manual meter reading, reimagined.

Water utilities depend on accurate meter readings to bill customers and manage supply — yet most readings are still captured by field staff who travel to each meter, read the dial by eye, and transcribe the value by hand. Across a large, distributed meter fleet, that process is slow, costly, and prone to human error.

10decoders set out to show a better way: capture the meter with a standard phone camera, let computer vision read the value automatically, and return a clean, structured reading ready for billing — with low-confidence reads flagged for a human to confirm.

Challenge & Approach

From a manual field task to an automated read.

The challenge

  • Manual reading is slow and labour-intensive across a distributed meter fleet
  • Hand transcription introduces errors that flow straight into billing
  • Varied meter types, lighting, and angles make reads inconsistent
  • No easy audit trail to verify a disputed reading after the fact

Our approach

  • Capture the meter with a standard phone camera — photo or short video
  • Computer vision locates the meter and reads the digits automatically
  • Low-confidence reads are flagged for quick human confirmation
  • The captured image is retained as evidence for a clean audit trail
How It Works

The AI Vision reading pipeline.

A simple capture-to-reading flow that runs on everyday devices and returns structured, billing-ready output.

Step 01

Capture

Step 02

Detect meter

Step 03

Read digits

Step 04

Validate

Step 05

Structured output

What It Does

Built for real-world meter conditions.

Reads the meter automatically

Computer vision detects the meter and recognizes the dial digits from an ordinary photo — no special hardware required.

Handles varied conditions

Designed to cope with different meter types, lighting, and capture angles found in the field.

Flags uncertainty for review

Low-confidence reads are surfaced for a quick human check — keeping accuracy high where it matters.

Works from a phone

Field staff capture a photo or short video on a standard device — no specialist equipment to deploy or maintain.

Returns structured readings

Outputs a clean, structured value ready to flow into billing and operational systems.

Keeps an audit trail

The source image is retained alongside each reading, making disputed values easy to verify after the fact.

Why It Matters

The value of an automated read.

Qualitative benefits demonstrated by the proof of concept. Quantified field results can be added once measured at scale.

Less manual effort

Removes hand transcription from the meter-reading workflow.

Fewer reading errors

Consistent digit recognition reduces the mistakes that reach billing.

Scales across the fleet

A repeatable, device-based capture process scales to many meters.

Auditable & transparent

Image evidence behind every reading supports disputes and audits.

Start with a thirty-minute conversation.

No 50-page proposals. We'll tell you which level fits your situation, what a realistic engagement looks like, and what it would cost — in one direct meeting.