Nobody phrases it this way, but I think that artificial intelligence is almost a humanities discipline. It’s really an attempt to understand human intelligence and human cognition.
– Sebastian Thrun, “Reputation in the Age of Artificial Intelligence,” contentconnection.PRSA.org, March 16, 2018
Imagine the next generation of Product Innovation:
Rapid advances in AI/ML and Deep Learning will lead product innovation processes to innovate themselves. The tools available right now for that work are bound to evolve in Moore’s law timeframes – and faster.
The impact of AI technologies on business is projected to increase labor productivity by up to 40% and enable people to make more efficient use of their time.
– “Artificial Intelligence Poised to Double Annual Economic Growth Rate in 12 Developed Economies and Boost Labor Productivity by up to 40 Percent by 2035, According to New Research by Accenture,” https://newsroom.accenture.com/, September 28, 2016
“The only constant is change,” the saying goes. The saying is right. Obviously, marketing professionals can benefit from being able to predict these changes in advance, in order to prepare for them and take advantage of them. A trend is an indicator of change. All time-series data sets contain some kind of trend.
Data really is of a few kinds. Non-conscious data refers to data about the non-conscious mind of the consumer. This usually refers to the kinds of music they listen to, the movies they watch, the binge-watched TV shows they consume “mindlessly,” the YouTube channels they tune in to late at night, and the Pinterest and Instagram pictures they devour ravenously. Popular culture codes are shaped and fine-tuned through this mega consumption in the non-conscious mind. This popular zeitgeist forms a fundamental and important data source for algorithmic investigation.
Some of this data changes over the course of time, and some of it remains incredibly popular – the Beatles have fascinated generation after generation, and Charlie Chaplin still wows those who discover him.
Western zeitgeist, via pop culture, finds its way to infiltrate and influence minds around the world. New hybrids and variants are created in this confluence of non-conscious cultures.
Then there is information contained in conscious data. This is the data pertaining to conscious human actions: what people speak, write, tweet, blog, search for, post, consider, purchase, advocate for, rant and opine about, and return. These conscious acts contain in them the source and treasure of the next big innovation or desire. In addition, all call center data, focus group data, and consumer survey data contain consumer articulations.
Online search data coupled with real time fluctuations of CPC (cost per click) provide important inputs to innovation.
Demographic data pertaining to emerging, growing, and disappearing segments of the market provide important data for algorithmic exploration.
Global issues, be they environmental, sustainability, or social causes, provide additional information on the cultural zeitgeist that shapes consumer desires.
Desires shaped through aspirations in the non-conscious usually seek validation and support through conscious actions.
Machine Learning and AI analysis for product innovation and R&D requires assembling this necessary data set. The fully assembled data set has a number of elements that cut across brands, geographies, and categories of products.
In a survey of over 1,600 marketing professionals, 61%, regardless of company size, pointed to machine learning and AI as their company’s most significant data initiative for next year.
– “Survey Finds Machine Learning and Artificial Intelligence Are Top Business Priorities,” https://www.memsql.com/, February 7, 2018
Historically, product innovation required a team of researchers to continually conduct “trend clinics” and perform site visits to “cool places” to determine what consumers desire. Such activities were referred to as “trend hunting” and “cool hunting,” and consumed quite a lot of time, energy, and resources. Innovators would then pore over the data that was collected, and brainstorm using human ingenuity to determine what might be interesting innovations. Identified concepts were then presented to focus groups of consumers in standardized concept templates. Consumers were then required to vote on what they thought might be interesting ideas, which would then be taken forward.
Naturally, a great amount of “noise” is injected into this process:
It remains a mystery as to how many wonderful ideas died on the vine as they were swept through this well-intentioned yet archaic process.
A modern Machine Learning and Artificial Intelligence approach to product innovation comprises a nine-step process.
Whether it is an R&D department trying to create new products and services, or an internal service provider like HR trying to see what new services could be offered to employees, these steps for algorithmic exploration prove valuable.
Metaphors form the basis of non-conscious thought. The first step in the process is to extract metaphors that pertain to the topic or the focus of innovation. If the category is Soup, then the first step in the process is to algorithmically understand what Soup is a metaphor for.
Algorithmic extraction of metaphors uses semantic frames that structure the bridge between conceptually different domains. So for instance, Soup as Warm Comfort might be a bridge between the domain of Food, and the domain of Feelings or the domain of Home and Parenting.
Such an extraction of semantic frames is possible, and is done by the algorithms of artificial intelligence. These algorithms embody thousands of years of human knowledge, and cultural expressions across the world.
Extracting the global primal metaphors that pertain to the topic at hand becomes the first step of product innovation. Note that in traditional research, this is inferred from thousands of conversations with consumers.
Extractions from consumer conversations are problematic as inferences are drawn from conscious outputs of consumers about the state of their non-conscious. A better approach is to mine the inputs that shape the non-conscious, and draw inferences from them.
Metaphors extracted from the non-conscious mind are now validated through data pertaining to the conscious behaviors and actions of consumers. Algorithms look for support in the vast corpus of the conscious for that which was extracted from the non-conscious.
These are nicely grouped based on the level of resonance in both the non-conscious and the conscious.
Dominant metaphors and trends are trends that have already infiltrated our culture. They are hugely resonant in the non-conscious, and find great resonance in the conscious as well.
Emerging metaphors and trends are the trends just around the corner, and they usually have great resonance in the non-conscious, but are not present as much in the conscious corpus. This becomes an easy and clear algorithmic way to evaluate and extract emergent trends. Be they the rise of “naturals” in food ingredients, the rise of Ayurveda in health and beauty, the growth of ETF in mutual funds, these are emergent trends around the corner. They become a vital group to capture.
Fading trends are those that resonate low in the non-conscious, and yet are highly talked about. These are guaranteed to fade, and have short life spans, so building products in line with these metaphors is quite risky. These metaphors allow for rapid and short-term exploitation, however.
Finally there are metaphors and trends that are past. These should be appropriately called the Graveyard of Dead Metaphors. It is important to note that products and services that are built around these metaphors are activities that need a second and third look and must be put to an end.
Inputs to the non-conscious mind shape its desires, fuel its fears, and solidify its decisions. Algorithmically identify the particular instances in which the category of product or service appears. Whether it is a fragment of lyric in a song, or a jingle that refers to the category and is lodged firmly in the mind, identify each and every one of these instances where the product or service manifests itself.
Next determine if the context is relevant to the category in question. That is to say that money, if spoken about in the context of economic conditions, is relevant to financial services whereas the context if it is being spoken about as a precondition of romance may be more suitable for dating services. Algorithmic parsing of contexts classifies them into contexts that are category relevant and contexts that are not.
Now further filter these contexts based on their connection to the brand. A brand typically represents itself as a set of functions, a set of feelings, a set of values, and a set of imagery and semiotics.
Once metaphors are chosen, metaphor-based filtering can now be accomplished to extract contexts only where the chosen metaphor is activated as well.
Once these steps are systematically carried out, there exist a corpus of contexts that are ready for algorithmic exploration.
Evaluation of hundreds of thousands of product ideas reveals that product ideas are usually a response to occasions, locations, life pressures, cultural tensions, daily activities, aspirations, disappointments, work pressures, and the like.
There exists a core set of foundational building blocks for innovation that can be algorithmically explored. These building blocks are neither finite, fully fleshed out, nor deterministically defined, but rather are born out of decision and graph analysis of numerous ideas that were successful and an equal number of them that failed.
Algorithmic parsing of contexts feeding the non-conscious mind brings out core creative ideas that can be combined into concepts in an n-gram sense.
Very similar to how neuronal connections are formed in the brain, concepts are chained to form concepts and product ideas.
Algorithmic combinations of n-gram concepts result in literally a combinatorial explosion of millions of product ideas and concepts that satisfy a multitude of consumer needs and expectations.
Consumers increasingly expect the products and services they buy to perform more than one task well. One glance at the mobile phone nearby you proves that point; it serves multiple functions, and serves them well. Today, we expect our food to be both good-tasting and good for us, and our cars to be powerful and long-lasting. It is no longer exclusively an “either/or” world we live in.
Engineers can deliver products that feature many capabilities – some to the point almost of overkill, it could be argued. Product designers, and their marketing colleagues, have to determine which of these features to single out for the consumer’s attention. Some of the time that can be obvious. But other times, it can be a vexing dilemma; how to know best which individual or suite of features are going to be the ones that appeal the most to the buyer?
Because AI and ML systems are designed to gather, analyze, prioritize, and synthesize findings on a scale that is Herculean for any human, they can deliver specific findings that product designers and marketers can rely on for guidance when it comes to product feature prioritization. From endless oceans of data in countless databases, AI and ML tools extract the most relevant and meaningful indications of consumer preferences, both conscious and non-conscious in nature.
These systems’ capability to look backward and forward simultaneously, to deduce potential trends before they even fully materialize in the marketplace, to gauge product features’ competitive pros and cons, and even to suggest “counter-programming” measures to help offset a competitor’s product features perceived to be superior by consumers . . . these are all real-world reasons for employing AI and ML in the selection process for product features.
The millions of n-gram product ideas need validation and prioritization. Algorithmic validation and prioritization occurs for these product concepts through conscious consumer data.
A product concept that is truly innovative will be only lightly referred to in the conscious. If however, no reference is made to it, it might either be entirely undesirable, or just “too far out” for the consumer to even take notice.
Rule-based heuristics, historical innovation success and failure data, and experiential knowledge are coded into algorithms that can then extract, evaluate, and predict the consumer acceptance of a product idea or n-gram. Algorithms may be fine-tuned for a category, for a brand, and for a particular geography based on consumer attitudes.
That is to say, algorithms that look for the next big recipe, the next big ingredient, and the next big health and wellness thing can all be different based on brand and geography.
The next generation of intellectual property are the rule bases and algorithms that perform such evaluations.
Much as with product feature prioritization, AI and ML-based systems can help improve the effectiveness of product bundling. Their ability to assimilate vast amounts of data, from vast arrays of sources, and process it all at blinding speed means that marketers can gain a broader understanding and a larger vision of what particular product bundles are likely to appeal to consumers the most.
These systems also enable marketers to build different virtual models of product bundles and test out assumptions quickly and with more accurate and reliable results.
A product bundle, also known as a “package deal,” describes a marketing innovation where several products are sold as one complete package for a set price.
There are numerous examples of product bundling. For instance, an Office Suite software package typically offers a word processing program, a spreadsheet, a presentation builder, and other features. Priceline offers a discount if you purchase your flight, hotel, and rental car online all at the same time. Restaurants often combine a number of food items into one complete meal (such as the McDonald’s Happy Meal).
Algorithmic bundling is merely a classification of n-grams of product concepts into natural bundles – a product that satisfies A, does B, and addresses C while residing in D. What is typically a laborious process for humans to evaluate and perform is done in milliseconds by a well-trained algorithm.
While innovation can identify product extensions and service extensions, a very interesting by-product also results – category adjacencies and expansion potential. If for instance the contexts containing the product category in the non-conscious also contained another product category, it clearly indicates a neural binding and association between the two products. As it is in the non-conscious, consumers may be entirely unaware of that binding.
This binding provides a business a wonderful opportunity to diversify and move into an adjacent category. If it might not be strategically prudent to offer that expanded service, it might be useful to partner with another entity to offer the service.
Partnership opportunities, acquisition opportunities, and category expansion ideas naturally arise from algorithmic exploration of contexts in the non-conscious.
Services in conjunction with the product arise when the n-gram of product idea is only partly covered by a product the company offers, and the remaining becomes a service accompanying the product. The Direct-to-Consumer (DTC) approach can always be traced as a service that accompanies the n-gram of product innovation. Indeed, it becomes a powerful way to validate just exactly when a consumer might require that DTC or home delivery as a service.
Premiumization essentially poses a simple question: What needs to be added to the product or service that will enable it to be viewed as an upgrade and therefore command a better price in the mind of the consumer?
Algorithmic handling of the question is quite simple – figure the n-grams of product innovation, and extract the contexts where the innovation originated. Simply mine the contexts for cues of luxury and premiumization. Feelings, ingredients, and occasions usually provide cues to premiumize.
Product descriptions, or features that connect to the cues of premium, automatically trigger those bindings in the brain, and the consumer subscribes to the notion of premium.
Studies of thousands of products, and core neuroscience, reveal common bindings with premium. Typical cues of premium include, but are not limited to things that have colors of gold, black, red, and pure white. Natural textures and materials evoke premium. Associations with celebrities evoke premium. Time – for instance, whether a cheese is aged or a sauce is freshly made – evokes premium. Bigger, heavier, and stronger evoke premium.