You can (and should) read this over here in its original setting but they’ve said they’re OK with me having it here as well. No-one reads personal websites these days anyway - so it’s just an online archive really.

Every marketer thinks they know who their audience is. But what if the best audience strategy requires improvising with data?

If you’re a classically-trained marketer - which, these days, is like starting a music career playing acoustic guitar - you’ll understand ‘audience’ as a customer segment that drives growth for your brand. “Our target is Egregious Ethels who like bingo and knitting and believe in the power of prayer.” For you, Ethel becomes the cudgel you can use to beat media, creative, TikToks and, perhaps, colleagues in more far-flung offices into harmonious submission.

If you write briefs at an ad agency - like I used to as an account planner (aka strategist) at J Walter Thompson, in the days before my creative licence was revoked - audience is an intuition pump; a sleight-of-hand to help people come up with the right kind of idea.

For media planners however, both Ethel and the ‘creative target’ suck because they are audiences money just can’t buy. So you fake them with proxies to make everyone feel good, and frankly also to mollify your TV buyer, who spat in Ethel’s face and told you to focus more on ABC1 men “because Momma needs to hit deal next quarter”.

And on the other side of her deal targets, if you’re one of the TV sellers I hang out with nowadays, audience can be price discrimination. Sometimes you sell ABC1 men, not because they over-index (although they do), but to gate access to parts of the schedule other audiences cannot reach.

Meanwhile, if you ‘play electric guitar’ - ie you’re a thoroughly modern marketer whose audience clicked on the most recent Meta campaigns or bought denim from last year’s spring-summer collection - you get your kicks feeding first-party audience data into the fuzzbox marked ‘lookalikes’. And what comes out is merely a howl of statistical noise or, worse, the whine of self-fulfilling retargeting prophecy. “Radical dude! People searching for cars sometimes buy cars. Who could have guessed?” (Although also, frankly, who cares when “the CPA is so freakin’ low I’m crushin’ it here”.)

Jazz music, not sheet music

Despite these differences in craft, one thing all these marketers have in common is they’re rigidly playing sheet music written from their preconceptions about audience. They’ve decided what song they’re going to play - demographics, behaviours, media deals - and they’re banging it out, whether it fills the floor or clears it.

The opposite, and perhaps better approach, is data jazz, where you start playing, listen to how people respond, and adjust your tune accordingly. Sheet-music marketing assumes you know what the audience wants to hear. Data jazz lets the audience in.

It’s decades since Jeremy Bullmore wrote about people forming brand perceptions like birds making nests, “from scraps and straws we chance upon”, but it seems that in 2025 we’re still objectifying audiences. On JB’s telling, they are subjects not objects, who may or may not pick up the scraps we’ve strewn before them. They are the ones constructing their brand perceptions, not empty ears waiting to be filled with our marcomms music.

Yet, rather than remain humble before the riddle of human behaviour, we persist in our fantasies of control. Sure, we allow ourselves a bit of leeway in the research phase, but when it comes to running and optimising campaigns, our worst instincts of command-and-control take over.

We’re playing by rote when we should be riffing.

Don’t send a human to do a machine’s job

There’s a digital ad platform, LoopMe, which runs entirely on one-to-one brand-lift experiments. You start by telling the platform who you think the best audience is for your ad, and then, using your suggestion as a starting point, the platform experiments continuously, iterating to find the people whose hearts and minds are actually being won over by your chosen creative.

I once asked LoopMe’s chief data scientist, how do human-selected audiences compare to those discovered by the machine? He said there are two main differences. Machine audiences are bigger than those hand-picked by marketers. The humans go too narrow, over-targeting ‘common-sense’ audiences that are really anything but; left-handed golfers and the like.

And the other difference? That the audiences discovered by the machine through continuous testing are often far more interesting. The dimensions that distinguish people who are most likely to pick up the straw and add it to their nest of brand perceptions are nothing like the simple ones chosen by professionals, but are more tangled and complex.

Ironically, these are signals whose complexity does need a human to interpret, tune in and figure out how to jam. Like all good jazz, data jazz starts with listening to what’s going on around you, feeling the pulse of your fellow humans and finding a way to join in which is both true and new. And the only way to do that is to understand the inner logic of the music, not to bash out four-to-the-floor or stick slavishly to some metronomic baroque pulse, simply because you’ve decided in advance that that’s who your audience must be.

Every underdog has its data

And if that’s too anecdotal, how about this? Outside big agency land, in the world where smaller direct-to-consumer (DTC) businesses roam, it’s common practice to start outcome-powered Meta campaigns with an open target - which is to say, with no audience at all. Just listen to this Andrew Faris podcast with Brett Curry from OMG Commerce.

In those businesses, marketers (although here they’re more likely to describe themselves as ‘working in growth’) use the magic of machine learning to discover groups of people who are most likely to buy or sign up or visit, depending on the desired business outcome for their campaign.

They have learned to outsmart the power asymmetry at the heart of modern advertising. Meta’s targeting models are trained to use behavioural signals from billions of users to predict purchase data freely given to them by millions of businesses. They know who is likely to convert with far greater accuracy than any human could achieve with mere demographics or first-party hunches. If you don’t believe me, you’re in for a shock - because those models are about to get even better - just listen to this podcast where Meta reveals its new strategy for machine learning in its ad products.

So before they play another note, another campaign, these DTC nu-marketers listen to the composition of the people who responded, to understand what the signals are trying to tell them. To learn more about, well, the audience.

From playing to tuning in

The skill they are cultivating is the reverse of what classical guitarists and marketers alike are trained to do. Rather than proceed in an orderly fashion from ‘plan’ to ‘activate’ to ‘measure’, these growth musicians start with activation. They execute first and ask questions later - to understand what the performance data is telling them, to (in)validate their hypotheses.

This isn’t just a clever hack, using an ad platform as a research tool, it’s a fundamentally different approach. Instead of assuming they know their audience and then going out to hunt them down - putting a target on their back - they treat audience as a working hypothesis that gets updated based on who actually responds. Not as part of some fusty annual planning cycle, but continuously, in response to the business outcomes that are being delivered. Data jazz.

For them, audience isn’t a target, it’s a hypothesis about how to avoid the waste both of talking to people who are never going to buy and of talking to people who are going to buy anyway.

We’ve had a decade dominated by debate about avoiding the first kind of waste - usually with trite examples about not targeting high heels to men or dogfood to cat owners - but increasingly, through conversations about incrementality, we’re all learning to think more about the second sort. The perfect audience isn’t a demographic or a behaviour or even, bless her, Ethel - it’s the people your marketing needs to influence and actually has a chance of doing so.

More sax, less axe

Or, to put it another way, if you’re the marketing leader trying to get your musicians playing together, you need to be less conductor and more Miles Davis - right up there at the front, leading with your sax, not the axe. Stick with rigid segments and each department will optimise in isolation: digital teams chasing ever-smaller data pools, media teams locked into demographic proxies, creative planners tinging their lonely triangle.

But if you can get them to put aside their differences and treat audiences as shareable, testable hypotheses, then everyone’s listening to the same groove, discovering larger, weirder, more incremental groups that could transform your business. By embracing data jazz you unlock the ability to discover conversion patterns no humans would have played on their own.

So next time you find yourself writing or talking about an ‘audience’, try swapping in the word ‘hypothesis’ and see if you’re still making sense. If not, maybe think about whether you’re letting your desire for control and consistency get in the way of learning and discovery. Of data jazz. Ask yourself where in your business are you closed off to opportunities by thinking you already know the score, rather than giving yourself the chance to riff on a new one?

But if the swap works, if what you’re saying still makes sense in terms of testable hypotheses rather than fixed audiences, then congratulations. You are on your way to exploiting the same data advantages that let scrappy DTC brands outperform established competitors. Learning to improvise rather playing by rote can be the difference between declining returns and compounding growth.