“Too much data, not enough insights!” It’s a common complaint we hear when working with content owners and video service providers. In their 2024 Predictions, the DPP stated: “Everyone will need a Data Strategy”.
Most leaders in the media industry would agree that data is an organisational asset and that, coupled with people and creative ideas, it helps their business create value. This episode discusses the key components of a successful data strategy, considerations for media companies, and practical advice on how to get started.
Inside the episode
Simon McGrath and Alex Borland are joined by Matt Eaton of Nepholve and Merapar’s Klaus Seiler. Matt starts from the business, not the data: there is no such thing as a data strategy, as the saying goes, only data for strategy. Matt walks through how to begin one; Klaus then sets out the technical challenges, and why responsibility for data belongs with the teams that produce it.
The second half turns to AI: where it already saves time in a media organisation, why it makes an organisation’s own data more valuable, and the feedback loops between what is published and how audiences respond. It ends on architecture and on each speaker’s takeaway.
Highlights
- 06:24 “It’s data for strategy. There’s only one strategy and that’s the business strategy.” Matt Eaton
- 13:52 “This means it cannot be a central data team which handles all that, because that’s a bottleneck.” Klaus Seiler
- 14:36 “If you’ve got a very siloed organisation, you’re going to have very siloed data as well.” Matt Eaton
- 24:26 “An asset is only an asset if you can find it”: a line from Diageo at the Henry Stewart conference, and a case for getting the basics right.
- 36:25 “Historically we’ve put the cart before the horse and that we’ve thought about AI before thinking about the data on which it has to feed.”
- 38:35 “Start small, but start.” Matt Eaton
Key questions answered
Where should a media company’s data strategy start?
With the questions the business needs to answer, says Matt Eaton, working backwards from there. With one broadcaster the starting point was a workshop for the senior leadership team, with content, programming, operations, HR, finance and marketing all represented: what information do they need day to day, where does it come from, and how much do they trust it? Matt also audits existing reports and how they are shared: a report emailed to everyone at the end of each day, piling up in inboxes, is not accessible.
What principles hold it together?
Matt works from five or six: data as a strategic asset, understood by senior leaders and not just lip service; reusing data once captured; governance, covering security, regulation and ethics; collaboration; and accepting that the work is never finished. Defining data owners is fundamental, and a step many organisations skip. Culture matters as much as technology: Matt cites an AWS report in which, from memory, something like 92% of organisations named culture as the biggest blocker to becoming data-driven.
What are the technical challenges?
Klaus Seiler names data integration first: many systems, formats and schemas, and sometimes legacy systems that cannot easily be integrated. Then data quality, so the data can be trusted; scalability, as volumes keep growing, in live sport perhaps with sensor data from stadiums and cameras; real time; security and privacy; and serving AI and operational needs, not only BI. Since cloud storage is cheap and decoupled from compute, Klaus’s advice is not to throw anything away.
How does AI change the picture?
Matt separates machine learning, applied to narrow tasks such as transcription or spotting a logo, from large language models and generative AI. An LLM can turn a 1,000-word news article into a 200-word first draft for a teenage audience, and image generators work well in ideation sessions, but a human stays in the loop. Matt asks for care with ethics: automating repetitive work is one thing, taking the creativity out of an organisation another. Klaus adds that a general model trained on the internet will not help with every task: you bring the model to your data and fine-tune it, as open-weight models such as Meta’s Llama 3.1, just released at the time, allow.
What does the architecture need to look like?
Klaus sees data lakehouses, with storage and compute decoupled in the cloud, as a real change, but they serve BI and analytics; search and discovery needs answers in milliseconds, and more data stores. Klaus wants domains coupled by event-driven architectures, a choreography of systems that listen to each other, with each team responsible for its own data. And it is never a finished project: the platform has to evolve as systems and the market change.
Speakers
- Matt Eaton, co-founder, Nepholve
- Klaus Seiler, co-founder and CTO, Merapar
- Simon McGrath, CMO, Merapar (host)
- Alex Borland, VP Strategy & Business Development, Merapar (host)
