What can AI really do for publishing?

< Blog home

Highlights from the Society for Scholarly Publishing's 2018 UK Event

Earlier this month, the SSP organised their first UK summer event at the welcoming and cool offices of Work.Life Clerkenwell in London. The event's theme was 'Humans, AI, and Decision Making' - how can we make use of data to help us make better decisions? 

Isabel Thompson, Senior Strategy Analyst from Holtzbrinck, kicked off the evening with her insights into this area. Our co-founder and CEO Mads Holmen had the pleasure of speaking as well, and explored what artificial advice systems can do to help humans make better decisions.

Read on to get the highlights of these talks. You can grab their presentation slides too:

---

#1 AI has tremendous capabilities, but it's dramatically under-utilised

"AI is an intuition machine that learns through experience, but not like humans do. With these qualities, what can it do?" - Isabel Thompson

---

Isabel Thompson

Companies in pretty much every industry, including publishing, say that they're unsure what to do with AI. Isabel pointed out that, when talking to other strategists and business owners about using AI, the most common problems she comes across are all strategic - not technical:

  • Competing investment priorities
  • Unclear use cases
  • Unknown ROI
  • Not enough, poor or siloed data
  • Un-agile organizational set-ups
  • A lack of data scientists and tech talent

In order to get anywhere with AI, she said companies need to address these types of problem first. To do this, it helps to know what the immense potential of AI is. To look at it in more depth, Isabel used PWC's great visualisation, which shows how AI’s different fields, capabilities, and applications fit together:

SSP18

#2 Don't invest in 'AI'. Invest in solving a business problem. 

"AI can do a lot of stuff. But you need a business case, and AI is not always the answer." - Isabel Thompson

---

AI is a tool, and there’s a lot of hype around it. Isabel stressed in her talk it should be evaluated like all other tools. Is it actually the best thing for the job? 

If you look at where companies globally are getting meaningful benefits from AI, she mentioned, you tend to find that they are using it in the core part of their “value chain”. Isabel showed us the below image from McKinsey, highlighting four areas in the value chain where AI can really help:

Value chain

On the revenue-generating impact side of things, there is a lot you can do to improve discoverability, sales and marketing, e.g. better prediction of demand, advertising, and article recommendations. AI also has an impact on the cost-cutting and efficiency side of things; it can help with production costs, supply-chain management, and stocking.

Despite all these possibilities, integrating AI effectively into your business will only work for you if you consider these five things (Isabel thanked McKinsey for making these aspects available too):

  1. Do you have the right business case?
  2. Are you building a data ecosystem?
  3. Have you got the right tools and techniques?
  4. Are you integrating AI throughout your workflows?
  5. Are you adopting an open culture and reskilling your workforce?

#3 Recommendations are a key area for exploiting AI

"For a long time, publishing was about printing and distribution. It's now about filtering and curation." - Mads Holmen

---

Mads Holmen

As mentioned in highlight #2, article recommendations and promoting the right products to people are a key application of AI. With recommender engines, techniques like natural language processing, classification and ranking can be used to come up with the best suggestions for e.g. what to read next. 

Improving the discoverability of the content, Mads explained, boosts the loyalty of your consumers. Every visit to your page or platform is an opportunity to deepen user's bond with your brand using content. A consumer or reader who bounces because they hit a dead-end is a missed opportunity. 

So makes a good recommendation? All recommendations will be based on one of three approaches (or a blend of them), Mads explained. They'll be based on the content, on the audience behaviour and/or the individual preferences and goals:

The ingredients of a good recommendation

#4 To make AI work, think differently about competitive advantage, partnerships and vendor relationships 

"The way publishers are creating value is changing. For example, discoverability, TDM, analytics and author services are now key parts of publisher offerings, when they weren’t at all even a decade ago. AI will exacerbate these kinds of changes." - Isabel Thompson

---

AI means that developing and sustaining competitive advantage will increasingly be driven by data, Isabel explained. This means that access to unique or privileged data sets that your competitors don’t have will be critical: this data enables you to generate unique insights, products, and services.

Unfortunately, lots of data is siloed across - and within - vendors, publishers, research institutions and all the rest. This means no one organisation is going to have all the data it would like to. This means two things, according to Isabel:

  1. Publishers will need to be more creative and open with partnerships, also with 'frenemies'
  2. As most AI talent sits with vendors, publishers need to work with consultants and vendors to figure out what might be possible for their businesses

Working closely with other publishers (highlight #1) could bring the pleasures of browsing a site like Wikipedia to all consumers. Mads said: "Why can't we have consumers discover content across publishers' sites? Why is it that Google, Facebook and Amazon have run away with the whole buffet and we, the publishers, just fight over the fruit bowl?

"More publishers see the value of working together with 'competitors' over buying traffic from a platform like Facebook, because consumers coming in from other quality content sites are more engaged and add more to the bottom line too:

 

Syndication

Regarding highlight #2, Mads, as an 'AI vendor', talked about why you might choose to purchase an AI solution if you were a publisher looking to power recommendations:

"It’s possible to build a recommender system yourself, if you’ve got 18 to 24 months and the budget to hire machine learning PhDs. You wouldn’t build your own web-hosting architecture though - it just doesn’t pay to reinvent the wheel."

Thank you note - and the presentation slides

TC

It was a pleasure to attend the Society of Scholarly Publishing event to hear Isabel and Mads share their expertise and network over drinks. Tom Ciavarella from Clarivate Analytics was a brilliant organiser, and the Work.Life offices were a great venue and host. 

If you enjoyed the highlights in this blog post, why not grab the slides of Isabel's keynote here and learn more on defining an AI strategy? Interested in what recommendations can do for publishing? Mads's lightning talk slides are right here.

Recommended for you

Interested in this topic? Get in touch to talk more