Unlike last year or the year before that, I’m not going to share blow-by-blow insights from speakers’ presentations. Let’s limit this post to what I found really interesting and actionable. Don’t expect to find seven types of panacea; instead think of it as a series of pointers that ought to motivate you to study the outlined concepts.

Yes, the wi-fi sucked and some presenters were not the best English speakers. And you know what? I don’t care. As long as they had useful content, it’s all fine in my book. Also, Craig Sullivan rocked as the host!

Lukas Vermeer

  • Don’t stop your A/B tests once they hit significance threshold. Let them run much longer (ideally two business cycles). Wait, that’s what Craig Sullivan said two years ago. It’s still true nonetheless…
  • Avoid these traps:
    • continuous testing – By randomly refreshing the tested page, chances are the test will hit its significance earlier, so filter out the peekers.
    • selective attrition – having users struggling with the task (purchase, filling out a lead form) leads to a drop-out. When the data aren’t included, it skews the results. Thus, look at the sample that isn’t in the test. This is, I think, the most difficult job of a tester.
    • Simpson’s paradox – ramp-ups mess with averages. Make sure you analyse ramped-up tests separately.
  • Understand the p values. Learn stats.
  • Good prediction trumps big data. That applies especially to low-traffic sites.
  • Focus on the money – your end goal is to help a site to convert like a motherfucker.
  • Go out and ask your users what makes their life difficult when interacting with your site.

Kevin Hillstrom

  • Have a process for revenue, get it sucked into your ecosystem.
  • Take a break from constantly executing. Create a development process for new merchandise.
  • E-commerce stores tend to cannibalise audience in stores in favour of the website.
  • Center analytics around merchandise instead of campaigns and ad spend. Well, that’s a bit self-serving point for merchandising analysts, but it’s valid nonetheless.
  • Measure how a customer behaves over time.
  • Stand out, be different, have a point of view!
  • People who buy once or twice can constitute a majority of your profit. Therefore, focus on customer acquisition. That works for products with long lifecycles.
  • Personalising merchandise assortment is crucial.
  • Keep your eyes on the prize – tie every marketing initiative to merchandise and buying.

Lucie Šperková

  • Last click sucks, because all heuristic models undervalue all channels between the first and the last click.
  • Take into account spillover effect (one channel drives other channels) and carryover effect (one channel drives the same channel).
  • Look at the data within a context (e.g. if you have only 10 million people on the market, don’t listen to a model that optimises for 30 million).

Colin Woon

  • Having an in-house SEO team reduces dependency on agencies.
  • SEO and natural search are different. Natural search focuses on clicks, CTR, position. SEO focuses on PR, linkbuilding.
  • Brand traffic matters a lot. SEO challenge: try to minimise leakage to other sites from your own brand terms.
  • Google seemingly gives priority to reviews sites during product pre-announcement phase. During the post-announcement, it’s the customer and specialist reviews that drive traffic.
  • You need to have control over the site and a good process management (especially, if you are a corporate fellow)

Michal Pastier

  • Content must have a personality (Haru Urara story or Tama the cat story).
  • Humans judge experience based on how we felt about it during its most intense point and at the end (peak-end effect).
  • To create a great ad, you’d better have these ingredients: engagement, happiness, character, emotion, peak-end effect. To have a great character, create motivation for it and stay consistent in further communication.

Jim Banks

  • Every business segment has unicorn keywords – seek them. For instance, by going through log files and search terms reports.
  • Don’t wait until your project is pretty to launch – you can always change it later.

Andrej Pančík

  • FB pixels report with a +/-15% error, and they have no lifetime value tracking.
  • Pixel checker checks only for text (says the pixel is there even if commented out) – use WASP
  • Sometimes even a small bid is good enough to reach your audience. No need to overbid your competitors – e.g. if you have a celebrity audience.

bid-volume-assumption bid-volume-reality

Frederick Vallaeys

  • Automation, on-demand economy and artificial intelligence will influence the world of PPC heavily.
  • If you let automation run without control, it gets unhinged.
  • Do not use averages for bid management.
  • Anchoring bids: when changing one bid, you have to change the modifier as well to have it grounded in reality.
  • Use Google Preditction API to gain insights on what AdWords might do to your bids.
  • Some automation is too conservative. If you add a human element, you will become more aggressive and more likely to win over fully automated accounts.
  • Break complex PPC management into small tasks that can be done by a freelancer.
  • Show a set of KWs to multiple humans, make changes they all agree on.
  • AdWords equivalent of artificial intelligence is Quality Score. QS is nearly identical to CTR.
  • It is extremely necessary to differentiate yourself in AdWords – headlines, calls to action.

Stephen Anderson

  • Game design is pivotal to good UX design.
  • Adapt this framework for your UX efforts:
    • Offering: what happens before the purchase starts
    • Behaviour people exhibit when purchasing
    • Experience – how people feel during purchasing
  • You can control offering, but the behaviour and experiences are its outcomes that get shaped by systems we design.
  • Adopt a game designer’s mentality: Until my players feel ___, we will not ship.
  • Stories and anecdotes win hearts and minds, but things told with numbers get formal support. That is because… If you can quantify something, you can rationalise it.
  • It’s dangerous to equate behaviours (external) with experiences (internal).
  • Freemium business model might become a problem, as people are getting sick of it. It drives up behavious, but drives down experience
  • Good example of how to create a good experience.
  • Until we have a reliable way of measuring experiences, we have to triangulate from behavioural data, surveys and outside statistics.


Michael Aagaard

  • Get the notes from his talk on his landing page
  • Understand the difference between system 1 and system 2 from Daniel Kahnemann’s Thinking Fast and Slow.
  • Do not overwhelm people with choices – it maxes out on cognitive strain. Create a conversion experience that facilitates cognitive ease instead.
  • Do a funnel walk-through without thinking like a marketer who built it. Identify where users get stuck the most.
  • Manage users’ expectations, as they don’t want to be surprised. Having said that, don’t lie on the landing page or CTA buttons.
  • Take advantage of a priming effect (exposure to one stimulus changes the way you interact with another stimulus).
  • Perform 5-seconds tests, record people’s behaviour, conduct usability tests.
  • Know and use the framing effect. There is a difference when you say that we are having a dead fish and seaweed for dinner vs. having sushi and wakame salad.
  • Use the word get instead of order.
  • Conduct interviews with sales and support – identify your customers’ pain points.
  • Use feedback polls to get answers.