Birds Eye View

What media can learn from music’s art of improvisation

By John Clarvis, Data and Insight Director as originally featured in WARC.

As in making music, luck and improvisation aren’t enough when it comes to making sound media investment decisions. But there are transferable lessons between the two worlds, writes John Clarvis, Data & Insight Director at The Kite Factory.

When I was 18 and learning how to improvise music, I was disappointed to learn that just “jamming” is probably the worst possible way to create meaningful, artistic, emotional and musical improvisation. The greatest jazz musicians did, in fact, do the opposite. They learn the changes, the theory and make their own personal connection. To connect emotionally, musicians must understand what is going on in the first place. Lucky strikes and happy accidents can happen, but it’s not good enough to rely on.

In media, there are direct parallels. Data teams create ever more impressive technical feats while creatives leap into the emotional – rarely meeting in the middle. The data team know the theory that the creatives need to be able to improvise in a way that emotionally connects with their audience. While agencies over the last 10 years have craved innovation, what they needed to do was to improvise.

So how do musicians improvise and how can agencies learn to do the same?

1. Keep it simple and make a lead sheet

A lead sheet is a fast and easy way to provide musicians with a broad view of the changes in a piece of music. They do not contain the details of melody, or individual notes but provide the essence of the music. Lead sheets are how Miles Davis turned Disney musicals into jazz standards. While they contain the essence of the song, the provide latitude for improvisation.

Data teams are prone to over provide details, in a large part due to the deluge of data overwhelming agencies with irrelevance. Just because something can be measured, it doesn’t mean that doing so will be useful. The job of a data team should be to trim down the data into the essentials such as the essence of the audience and the brief.

The gold standard for providing the essence of an audience is segmentation. A technique that takes one large sample of an audience and then divides them into distinct groups. This provides planners and creatives with a lead sheet of the intricacies of their audience without getting bogged down in excessive detail. A recent example I’ve worked on identified seven distinct audiences with unique barriers and motivations to charity giving. Mapping these to media consumption provided the planners and creatives to improvise in harmony with the audience. Eliminating one promising idea but providing seven more in exchange.

A chronically underused technique is factor analysis, which in essence clusters variables. When eight questions on a survey or 30 variables in GA all act similarly there is little point in reporting them all. Factor analysis lets us combine these into understandable factors rather than reams of indistinguishable, uninspiring graphs.

2. Look in weird places

The most used drum loop in the world is the “Amen” break from a B side of a Winstons single. The second most popular drum sample comes from a bongo band’s cover of The Shadows song, “Apache”. In hip hop, crate diggers spend countless hours listening to obscure vinyl for clean and unique samples. Rarely does a sample come from a well-known song. Crate diggers have an ear for the weird, often resurfacing artists and elevating them to levels of fame unknown during their career.

Miles Davis’s most critically acclaimed work, Kind of Blue was inspired by a form of music not examined since 1547, Gregorian chants and church modes. An improvisation approach that builds new relative scales based on derivations of a key. Once forgotten it now forms the foundation of any training in music.

Many statistical techniques long reserved by statisticians and academics have been resurrected by big tech. The same algorithm that drives segmentation (cluster analysis) now serves as one of the backbones of machine learning. The same applies to regression analysis, once used exclusively by academics it now lives on in econometric modelling.  As well as techniques, data sources can also be unusual. In today’s digital world, almost everything is recorded as digital data. I’ve used audio profiles, image patterns, telemetry, weather stations, recordings, personal sensor recordings, CVs and traffic data to add something new to the planning process.

The best assumption to make is that if it exists then it is probably recorded, and if it has been recorded then there is a way of analysing it.

3. Embrace dissonance

The best musicians know when to create dissonance and when to resolve. Doing so is what drives emotional connection. Entirely consonant songs tend to be considered uninteresting and predictable. Of course, while entirely dissonant genres exist, they appeal to a very small niche audience. Dissonance resonates with listeners by shocking them out of their comfort in a song. Dissonance can be entirely musical. Amongst his many innovations, Bach introduced dissonance in the final bars of the St Matthew Passion to evoke strong emotional connections with the audience. More recently, Radiohead sprinkle moments of dissonance across their music to shock and intrigue their listeners. Charlie Parker referred to dissonant notes as the “pretty notes” and played them an octave above the harmony to soften their dissonance while retaining their essence.

Lyrics can also provide moments of dissonance. Steely Dan are famous for not only their musicality but for the juxtaposition of their cheery sounding music with incredibly dark lyrical themes.

My pet hate is when data is selectively fitted to match a narrative. This only serves to provide bland, uninspiring work. The role of insight should be to surface dissonance, as these are the interesting bits.

Statistical analysis in the main part is concerned with providing consistency and smoothing out outliers. This is with good reason; most outliers do not tell the whole story and disrupt an otherwise nice story and robust figures on which to base planning. In a pre-digital age this was essential as the sample numbers of research were typically too low to allow outliers to skew results. Digital data offers us the chance to examine outliers more closely. Clustering and factoring data allow us to provide discreet samples that include the majority but include the weird, interesting stuff.

The most important facet of dissonant data is that it can provoke discussion and through discussion lead to further digging into the data insight. In the luxury automotive industry, new car buyers express interest and preference for the performance of electric cars, but in blind tests prefer the look of petrol cars. This leads to a brilliant brief for product designers: how can you make an electric car as visually appealing as a petrol car?

The impact of Barbie vs. Oppenheimer is entirely due to dissonance. What could be more dissonant than the development of apocalyptic nuclear weapons to the bright pink world of Barbie? The dissonance intrigues and amuses people. Insight should do the same.

4. Is it a banger?

By far the most important facet of music is whether or not it is a banger. There exists a world of technically brilliant music, but, as any DJ will tell you, nobody cares about technicality if doesn’t bang.

James Brown, Dr Dre, Timbaland, Mark Ronson and even Simon Cowell understand what makes a hit. James Brown would dance in front of his band until what they played matched his movements, anything that wasn’t deemed funky was brutally dismissed. The X Factor and The Voice serve a pop audience with the same brutal cutting of anything that will not make a hit. DJs live by the reactions of their audience, feeding off and manipulating the emotions of their audience. This is not to say that expertise and technicality don’t make hits. At over five minutes long Bohemian Rhapsody, a baroque rock opera inspired by Beethoven and Bach, is perhaps the most unlikely hit of all time – but it was.

Insight work is typically the opposite of a banger and in making it too long, too technical and too detailed the ability to inspire creatives, clients and ultimately great, effective work is lost. This is where analysts need to become more like a DJ than a composer. Technical and methodology focused information should always be given, but as an appendix. The biggest, most interesting insight should lead and be presented to make the biggest insight. Tensions and dissonance in the data should be emphasised and resolved through the skill of the analyst into a plan of action.

Much like music, insight should be considered a platform for improvisation. There is no such thing as a completely correct answer to a brief, the winning insight is the one that inspires, entertains and excites its clients. If it doesn’t do this, then it shouldn’t make the grade.

My final test for an insight to determine if it is a banger my simple but effective pint test: “Would I tell this to someone over a pint?” If the answer is no, then it needs to be discarded or at least reworked until you would.