ITV: AI Video Analysis for Automated Segmentation Workflows

About ITV

ITV is a leading media and entertainment company, with the largest commercial television network in the UK and a global production and distribution business spanning more than 55 labels. It reaches over 40 million viewers weekly through ITV's family of channels and the ITV Hub, available on 28 platforms and more than 90% of connected televisions sold in the UK.

The Challenge

ITV needed to modernize a manual segmentation and timecode capture process used to identify 12 physical and creative content segments, including color bars, blacks, slates, recaps, PRTS, credits, program part segments, and bumpers. The process was managed through a MAM solution and a team of ITV tagging operators. Tagging was necessary, but it consumed resources inefficiently without adding value to the content.

Challenges in content segmentation:

• Manual and labor-intensive, with exposure to human error
• Time-consuming, creating inefficiencies and high costs
• Difficult to scale to high volumes, since it required linear manpower, multiple passes, and high accuracy for playout

The Solution: AI Video Analysis for Automated Segmentation

BRAHMA AI brought BRAHMA AI Core Segmentation to ITV, automatically identifying frame-accurate timecodes and segments through AI video analysis. Its cloud editor lets operators review and QC AI output and adjust timecode frame boundaries where needed. The toolkit is built for near-100% frame accuracy and full automation of the workflow used to extract content segments. For ITV, the solution was customized for high performance, scalability, and low manual review time.

BRAHMA AI and ITV jointly created the scope and solution design, covering integration with ITV systems and algorithms suited to ITV's TCC capture guidelines. The project moved through four stages, with agreed milestones and a smooth rollout plan.

Implementation Journey

Stage 1: POC and Trial

ITV and BRAHMA AI agreed a 12-week timeline to process a defined content volume and address the 12 TCC use cases at a preset accuracy level. BRAHMA AI demonstrated that its existing, off-the-shelf AI platform could identify at least 8 of the 12 use cases without changes to the existing algorithms:

• Essence
• Bars & Tone
• Clock/Slate
• Recap
• Opening Title
• Program Part Segments
• PRTS
• Textless Elements

These segments were identified at over 90% accuracy initially, cutting QC time to under 3 minutes, down from 15 to 20 minutes manually. BRAHMA AI processed about 50 hours of content during the POC, then moved to customizing the software for the remaining four use cases.

A key early challenge was aligning ITV and BRAHMA AI on how to measure and report AI accuracy. BRAHMA AI set up a weekly governance process, introduced a standard metrics and scorecard template, and worked with ITV's Content Operations team to revisit the content validation step. This removed ambiguity and shifted reporting to fact-based measurement. Within weeks, AI accuracy rose from 60% to more than 90% for most of the use cases that hadn't been addressed before the POC.

Stage 2: XML Integration

The ITV TCC process went live using an XML integration approach based on AWS notifications. ITV uploaded a new proxy asset to an AWS S3 bucket acting as a hot folder; BRAHMA AI picked it up for processing based on an SNS notification. The TCC Discovery engine populated results in the BRAHMA AI UI. Once processing finished, ITV's operator was notified, the QC team reviewed or edited the result and finalized it, and the XML was created and uploaded to ITV's S3 account.

Stage 3: API-Based Integration

The next integration used RESTful API calls to trigger jobs directly between ITV systems and BRAHMA AI. BRAHMA AI worked with ITV's Content Delivery, architecture, and engineering teams to understand the process and tune its AI engines, enabling fully API-driven workflow automation.

ITV submitted files through the BRAHMA AI API with the storage path and additional details such as work order type, series name, and episode title. BRAHMA AI processed the asset from the supplied data and returned job status and result URLs. After successful processing, ITV's QC team verified the result in the BRAHMA AI UI, then used the result URL to create XML for its existing MAM. A more enriched work-order function was underway as this stage progressed.

Stage 4: Generative AI-Led Enhancements

BRAHMA AI further enhanced the solution with generative AI. These improvements rolled out in four phases between February and May 2026, focused on improving detection, tagging, and localisation across ITV's content. The enhanced capability applies to new content and removes the earlier constraint of training Opening Title detection for a specific series. The improvements delivered:

• An increase in overall detection accuracy from 95% to 98% across all segments
• Localisation accuracy improved from roughly 75% to more than 88%
• End Credits, Opening Titles, and Recaps saw the biggest gains in detection and tagging, with accuracy rising from around 50% to more than 80% for each segment

AWS Deployment & Scalability

BRAHMA AI Core for ITV runs on AWS Ireland. The solution uses a well-architected deployment on AWS Batch with Amazon Elastic Kubernetes Service, supporting scalability and cost efficiency. A high-availability, multi-availability-zone architecture helps ensure uninterrupted service if an availability zone goes down. The deployment scales up or down as needed, optimizing costs during burst workloads.

The project faced two key challenges: education and integration. BRAHMA AI addressed both through a consultative, transparent approach, working closely with ITV stakeholders through workshops, demos, and POCs. On accuracy, the teams used a hybrid model: the software did the heavy lifting while humans carried out QC of the AI output, enabling continuous improvement.