Choosing people over algorithms

It seems to me there might be an emerging trend back towards human curation. LinkedIn’s new Pulse App features human curation, the same is true for Apple’s forthcoming news app, and Apple’s new music service similarly makes a big deal about it’s human editors, including former BBC Radio 1 favourite Zane Lowe.

Two of these examples are from Apple, making it early to generalise, but there’s definitely a strong meme that humans are better than algorithms when it comes to recommending content.

I wonder three things though:

  • whether large companies prefer human curation because it leverages one of their strengths vis a vis startups – capital
  • if this is more about positioning than substance – human curation sounds better to most consumers than recommendations from a machine
  • whether the real answer is a combination – human curation augmented via AI – we’ve made one investment on that basis and are considering another one now
  • Dan Field

    It’s interesting – as you say shows the strength of of the big companies… I wonder if there is a placebo effect in play too. If one were to use a fully AI curated system but fronted it/branded with a big name such as Zane would the every day user know the difference? And would they feel they are now getting a better service? Would we know?

  • http://www.internetbillboards.net/ Tom George

    Hi Nic,

    Curation is the future of the Internet, but AI will continue to get better. The best combination in my opinion will be the perfect marriage of the two. Already when it comes to publishing content on the web, many businesses are just now beginning to understand how beneficial curation will be. Not only for targeting better content, but also because at the end of the day, it is about relationships. It is kind of hard to have a relationship with a machine, right. Disclaimer, I am the Co-Founder of Internet Billboards. We are a leading curation website. Thanks for the excellent post. Do you have any contacts for startups located in the USA?

  • James Penman

    Hi Nic,

    I’m not so sure it’s expensive as long as you know what it is that you want to achieve from the human curated content. Getting to that point of knowledge can be expensive but that applies to a more techy approach (AI, data feed aggregation etc). In addition, ironically, search algorithms love expert human curated content so it lowers your ongoing traffic acquisition costs. Genuine expertise is critical.

    James

  • joeyizzy

    Hi Nic

    I’ve spend a lot of time over the last 4 years on this question with lots of different retailers. When I first started in recommendations/personalisation everyone believed “The Algorithm Knows Best”. Most companies would happily defer to the model when deciding, for example, which other products to show alongside a key piece on a fashion site.

    Over the last couple of years I’ve seen this shift to allow the algorithm to select the product unless the brand experts know of something “better”.

    What then becomes important is how you define “better”.

    In large scale A:B tests, on sites with thousands of products and across tens of thousands of transactions the algorithm will (nearly) always win.

    Why is this? Well, when humans merchandise a site they will tend to focus primarily on creating “looks” and matching items together with the expectation that the customer will like what they see and add all of the items together and drive high AOV.
    Most of the best performing recommendation algorithms tend to focus on conversion, they will understand what the customer is looking at, what the customer has done up to that point and then show them the very best things it thinks will make them convert.

    In practical terms, this means that if a customer has looked at three red dresses on your site and you are deciding what to show them next, rather that show them a jacket, a handbag and a pair of shoes, show them the hottest red dress that you have in stock. They’ve spent the last 3 clicks what it is they were looking for and they might have even typed “red dress” into the search box.

    But as you know real life doesn’t quite work this way. Leaving it solely to the algorithm runs the risk of sacrificing long term brand authority for short term conversion. As we all know, customers like to be sold to, and if you ask customers what they’d like to see and which stores they’d like to shop on they will tell you the one that gives them outfit suggestions and ideas.

    So if I were you I’d feel confident that a blended model of AI and Curation, one that finds the right balance between brand authority and utilisation of data, is good strategy for maximising engagement and lifetime value.

    Joey