Brand management a la Affinova

Bizarre piece in Business Week about a company called Affinova.

Affinova uses geeky software for industrial packaging and brand design. The software, “a “genetic algorithm” called IDDEA, essentially simulates evolution, creating generation after generation of possible products or packaging until the strongest possible design emerges.”

At each stage the designs are scored by a panel of 750 consumers, and “By looking at selections over multiple generations and across the whole panel, the software identified preference patterns—a tendency toward a certain color or font or wording—and ultimately identified the top concepts. ”

Affinovas site explains it in more detail:

“Consumers are recruited by email to participate in the study. Respondents that pass client screening standards take a brief, pre-exercise survey followed by instructions for the concept evolution exercise. The exercise consists of 20 to 25 choice set screens where each choice set includes 3 or 4 concept alternatives generated from the concept space. The consumer experience concludes with a post-exercise survey and a note of appreciation.”

Then, “Behind the scenes, the evolutionary algorithm intelligently determines the contents of each choice set based on the patterns of individual and aggregate selections. As many as 15,000 choice sets are created in real time throughout the exercise. In the beginning, the choice set concepts are pulled from a subset of the total concept space (the “starting population”), which is randomly populated to give all variants an equal chance of survival. Over time, the starting population regenerates with recombinations of the strongest performing variants. Random jumps systematically reinsert weaker performing variants to give all variants several opportunities to survive in new combinations. By the end of the exercise, discrete clusters of similar concepts emerge, featuring the most effective variant combinations.”

Finally: “The Output
1. Top Concept identification: Representative concepts are generated that best represent the element variant combinations within each emergent cluster. These “Top Concepts” are the fittest concepts in the considered space.
2. Preference segments analysis: Consumer segments are created by assigning respondents to emergent concept clusters based on their choice patterns. The pre- and post-exercise survey responses are then analyzed by preference segment to either (a) prioritize the Top Concepts or (b) uncover multi-concept segmentation schemes.
3. Element/variant analysis: The concept clusters and consumer choice data are analyzed to determine the relative importance of concept elements (choice drivers) and the strength of alternative variants for each Top Concept.”

I assume that the scoring takes place over a period of time, and all that their software does is to redesign the options presented to each new scorer, based on the survival of options from the previous sets, with some randomness.

Its difficult to know where to start in criticising this idea:

1. so much depends on who sees the options first, since their choices determine the next generation, and so on. Seems to me its not really 750 people, its 7.5 people 100 times over. The system actually guarantees a very small sample size, particularly if you want to talk about segmenting the results.

2. as any colour theorist will telll you, colours scarcely exist on their own: its only in combination with other colours that they take on life. So a tendency toward a certain colo(u)r is highly misleading. I suspect the same applies to other design elements: you need a holistic approach. (“they dont like red, so lets try blue – no? green – ah, yes, they like that”). But what if they only like red if its in a sans-serif typeface? and you tried it with a serif font?
3. I also think theres a fundamental flaw in this whole focus group approach: sometimes people respond best to something new that catches their imagination. (eg the Cadbury drumming gorilla advert. No focus group choosing between Times Roman and Helvetica would have got to that!)
4. also, the notion that typefaces or colours can evolve is a difficult one. People can evolve and adapt, yes, by developing new skills or abilities. But colours dont change: this is only about our preferences. The design does not evolve – it cant, it just changes. Evolution works by measuring fitness, not attitudes. (Prehistoric man did not become modern man by having people comment on his looks. He did it by becoming smarter. But one design is not inherently better than another. Some people may like it more than another, but thats a different thing.) And of course evolution is a continuous process, redefining its standards as it goes. (The best ape can climb to the tastiest bananas; the best man can define and popularise a share price option valuation equation that buys him the best caviar.) Whereas Affinovas final evolved packaging design would be judged by exactly the same standards as the first one they tried. (ie its possibly a better way for the ape to climb the tree).

Perhaps the geeky software should be compared to the systems by which Amazon rates books or Google rates websites: a collective intelligence rating. But these only really work when you have thousands or even millions of results, and can genuinely segment them. Even then, the evidence probably suggests that memes and fashions change quickly over time, which may be OK for a dynamic rating system such as Googles page ranking, which changes ranks regularly and often, but is less useful for a packaging/ brand designer who wants to put something on the shelves for the next few years.

The basic idea, though, claims to be another one: that an evolutionary algorithm can scientifically develop a design by randomly varying its component parts, and then gradually selecting the best (ie the most popular.)

Looks to me as if they read Segarans brilliant Collective Intelligence and liked all the algorithms so much they decided to stuff them all into one piece of software, without considering the underlying purpose and activity of each one.

What it does do, however, is to sound convincing and scientific – which sells the software. There is a marketing lesson here: (perhaps just not the one Business Week think): long words sell.

The real evolutionary test will come when the market buys or does not buy the product.

Leave a Reply

Your email address will not be published. Required fields are marked *