I was recently speaking at a Google event in Sydney. The topic I was asked to speak on was 'Digital storytelling', so let me tell you a tale.

Humans love stories. We use them to make sense of the world. When we cannot understand something, we use stories to explain it to ourselves, and each other. Stories are everywhere because everything is literally complex, the product of many things interoperating and creating endlessly emergent effects.

The first stories were explanatory fictions. All cultures have them, different in form but not in function: where did life, the universe, and everything come from? We have a cognitive bias towards narrative because we live and act in real time in a world that is inherently unpredictable - and we desperately want it to be predictable. Stories are devices for learning, for encoding cause and effect among human affairs, for explaining how societies reward or punish certain behaviours. In this way, every story has a moral, not just the explicitly didactic ones. Stories are simplified models of human behaviour.

The memory palace is a well-known mnemonic trick where you store things you want to remember in rooms of an ever growing house. Lesser known is the memory story, where you can remember long chains of information if you connect them through character, cause and consequence. This is because, as Kahneman has pointed out, our memories are stories we tell ourselves with us as the hero or victim. We don't remember most moments of our lives; we remember the stories, as we decide what we have learned from experiences, or what we want to believe. Politics are stories about society. "Every election is a competition between two stories about America", as former Obama speechwriter Jon Favreau recently said.

Kahneman's research suggests that what defines stories are significant events and endings - the 'peak end rule'. Kurt Vonnegut's (rejected) master's thesis makes the same point in diagrammatic form. He showed how to graph stories using the axes of positive and negative events over time. You can tell what kind of story it is by the shape. Shakespearean tragedies end in death whereas comedies end in marriage.

Stories, even ones about real events, must be in some sense fiction, picking some elements to highlight and suppressing others. The beam of attention pulls some things into sharp relief, rendering others invisible.

When we try to understand human behaviour, we create simplified models to make sense of complexity: we tell stories. One of the creation myths we tell in advertising is built on a model of communication effect known as AIDA: attention, interest, desire, action. This is mapped onto a quantitative model which is usually drawn as a funnel because it's narrower at the bottom. At the top of the model we have the total number of people to whose attention an advertiser has bought access. Then we use surveys to understand how mid-funnel intermediate measurements like 'unaided brand awareness' or 'favourability' have changed among target audiences. Thanks to digital media, we can observe countless intermediate behavioural measurements like shares or likes or views. At the bottom of the funnel are sales.

We are measuring stuff to (ultimately) understand if advertising is working, based on our simplified model of the world. All well and good. Kind of. We know it's more complicated, and there are myriad influences on purchase decisions - 'all models are wrong but some are useful'. The problems arise when we start to hear the story about digital, and confuse the map with the territory.

We assign targets to media agencies based on impressions or CTRs, to creative agencies based on brand trackers, to digital agencies based on traffic, to social agencies based on shares. Then Goodhart's law kicks in, which states: 'When a measure becomes a target, it ceases to be a good measure.'

Goodhart was emeritus professor at the London School of Economics and pointed out that when we set measures as targets, we distort the incentives around them. People start trying to hit their targets, however they can, which renders the measure ineffective as an indicator of advertising efficacy.

This is what leads to all kinds of distortions in the market, such as fraud, bots and bad surveys. The client that gets the bonus based on call volume or video views has the same issue. (That's why we always ask our clients what metrics they get bonuses on. It saves so much time.)

The story of digital was that, with enough data from the fictional funnel, we could unpick the drivers of human behaviour. But it was only ever a story.