STRAT7 Bonamy Finch

How we used AI to analyse viewer conversations for ITV

Analysing viewers’ conversations to inform future marketing initiatives for the UK’s longest running commercial network.

Meet the client

ITV is the oldest commercial television network in the UK. Their free-to-air service is host to popular shows like I’m A Celebrity, Get Me Out Of Here!, Love Island and Coronation Street. 

Millions tune in every week to watch shows live or to watch the wide range of shows available for free on on ITVX, the company’s streaming service. 

The network’s popularity with the public reflects their efforts to appeal to the diverse tastes of the British public.

Dormant data: an opportunity in waiting

ITV has partnered with our sister agency, STRAT7 Researchbods, for over five years. Researchbods run ITV’s Insight Community, The ITV Village, where they conduct ongoing research with thousands of viewers. 

ITV have been testing their own advertising via this Insight Community for several years. This means they have an account of viewers’ reactions to commercials going back years.

These comments were valuable, but were being looked at on a weekly or semi-campaign by campaign basis, without consideration of trends over time. Which begged the question: was there an opportunity hidden within the vast piles of unstructured data? The answer was undoubtedly yes. 

Because each member of ITV’s community is deeply profiled, viewer information (such as demographic, viewing habits, etc) could be combined with insights from the unstructured data to help ITV understand how to appeal to different audience types.

ITV had one core question they wanted us to answer: how could we deliver more impactful commercials to get more people watching our shows?

Finding order in chaos

Working closely with Researchbods, we extracted 40,000 mentions and created a searchable data dictionary of shows to search through the unstructured text. 

The engine collects, analyses and visualises unstructured data quickly and efficiently. Careful pre-processing of the data, including sentence tokenisation and emotion detection, helped us focus on the most informative mentions. 

We applied topic modelling to group together the different areas of conversation, as well as analyse the precise language used by viewers. We used the topics and shows combined with the viewer information to build a rich picture of what appeals to different types of audience. 

Using this information, we created a data-driven evaluation of what shows generated more of a reaction from viewers, so ITV could focus their creative learnings on creating new campaigns using this rich seam of feedback. 

We were also keen to understand how responses to the topic areas vary by age and viewing habits. Younger viewers, for example, are attracted to humour and engaging soundtracks while older viewers prefer to build on previous viewing experience and acting talent.

The results: 5 key insights

The engine enabled us to collect and analyse vast amounts of data at a fraction of the cost and time of traditional research techniques. And five clear strategic insights emerged.

  1. We uncovered five high level topic areas in the 40,000 open ended responses, after filtering. These areas included music and soundtracks, the quality of acting and talent, and creating new impressions. These topics were then profiled against ITV’s shows and audience demographics to spot opportunities on how to position shows in future commercials. 
  2. We detected a clear shift in music and soundtracks since early 2021 (pictured below), which fit with ITV Creative’s strategic pivot in terms of the types of music in their commercials. They suspected the music was helping to increase cut through for their campaigns with viewers, which was expected, but now the data confirmed it.
  3. We identified the precise phrases that people used to describe trailers and the music being used, words like ‘suspense’, ‘well-suited’ and ‘upbeat’ which hadn’t been used by community members about the campaigns before 2021 . This helped the Creative Team focus on the specific choice of music which was generating a positive reaction.
  4. In an age where getting on to a viewer’s consideration list is essential, a strong first impression is key. This research identified a shift in attitudes around generating strong first impressions. The overall volume dropped over the lockdown period where ITV had a smaller pool of new content to draw, but saw signs of recovery as the world opened up again.
  5. A clear data picture helped us benchmark viewers’ reactions to the rollout of ITV’s new streaming service, ITV X. While the research was carried out in the early days of the service, the findings showed a considerable increase in comments signposting a sense of anticipation or trust.

Where next?

We’re a long way off letting ChatGPT write our ads, and the headline learnings from this project are being fed back into wider discussions internally about campaign assessment. However, this research was a useful proof of concept both for the application of AI, and for ensuring that ITV takes heed of the valuable lesson that sometimes the answer to a question can be found in existing datasets.

Turn your unstructured data into a competitive advantage

Your proprietary unstructured data is likely an untapped source of consumer insight. Deep exploration and analysis using could be a cost-effective route to a competitive advantage.

Discover how AI can uncover fresh insights from your existing data. Click here to find out more or speak to one of our consultants.


Using, we extracted 40,000 mentions and rapidly created a searchable data dictionary of shows to search through the unstructured text.

Craig Milligan

Global Head of Consumer Analytics & Insights | Decision Sciences, Diageo​