Krista Seiden, Analytics Advocate at Google, shared her tips for building a culture of optimisation in your business at the Out of the Box Confex 2016, supported by Green Hat.
Here are her top takeaways to help you build a smart framework to continuously test and optimise your marketing and sales efforts.
Control for your variables
Most marketers miss the bigger picture when they test. If you launch a new website and traffic doubles, you may reasonably infer that the new site is a smash hit.
Yet in reality, there are too many variables to validate that assumption. It may well be true, but you need to control your variables to pinpoint actionable insights. Run your tests as control as many factors as you can to ensure you get an accurate, repeatable read on your data.
Know your limits
Few of us have the time, budgets or operational capacity to test everything, all the time. Plan and prioritise strictly so your testing is focused on where you think you’ll get the most useful insights; and when and where you can control those variables usefully.
Tap the crowd
Testing doesn’t have to be the domain of your analytics specialist. Many in your organisation will have ideas about what might make a useful test – and those ideas can come from areas you least expect. Open up and gather testing ideas from all over; get a bigger brain!
Make your prioritisation collaboration too, and make sure you have sign off from the relevant people. If a test crushes it, there’ll be disappointment all round if it can’t be pushed permanently live because the right people weren’t involved in the process.
Design your test smartly
If you’re testing ideas side by side, they have to be sufficiently comparable… apples and apples.
Testing design should control those variables and focus in on a clear choice of similar enough elements that the data will tell you which was more successful.
Check – then check again
If you’ve uncovered a strong insight and a clear test winner, don’t rush to implement it. Take a breath and run the test at least two more times. There may have been elements you missed, variables you forgot to control for, or the results may be a legitimate fluke.
Triple check your tests and data before moving forward with any insights they reveal.