In product innovation, a company needs information about consumers, existing products and the competition to build-on when developing its ideas, concepts and designs for candidate products. It is usually a multi-stage process that involves planning, research (i.e., market, consumer & product), development and testing wherein information from different sources is likely to flow-in all the time. A particular issue of importance is identifying promising target segments and understanding how distinct potential segments may differ in their preferences and expectations from the product concept (e.g., attributes, uses). However, segmentation studies can take time and may stall progress in the process of new product development (NPD). Eric Tayce (VP, Burke Inc., Corporate Innovation) proposes a segmentation methodological approach that can be especially beneficial in an early stage of an innovation process. His approach turns the focus from using primary consumer data to employing existing data pools within a company to set direction in the beginning of the process.
Various techniques for segmentation are utilisied in marketing research, from basic schemes with a few variables and simple analysis to complex multivariate models and statistical analyses (e.g., clustering). In the context of NPD, a model may entail consumer needs & wants, preferences, beliefs and expectations, and additional background characteristics (e.g., demographic, interests, lifestyles). The more complex or sophisticated multivariate models based on primary data can provide richer, well-founded and more instructive consumer knowledge with respect to the prospect product, but they are likely to take much longer to devise (e.g., measured in weeks). Tayce points to a conflict in the objectives of executives between seeking “the speed and flexibility of an entrepreneurial start-up” and expecting “right alongside a level of analytical rigor” that is possible with the more advanced segmentation models described above, which are usually based on large samples. This conflict can get executives into a ‘bottleneck’ situation he calls “precision paralysis”.
A caveat of more elaborate modelling, while achieving greater analytical rigour, is the delay in obtaining decision-support findings, that could arrive after a crucial decision had to be made; this event may be particularly critical in the early steps of an innovative product development process. A more proper question seems to be: at what stage building a more detailed and rigourous segmentation model would prove more conducive? Tayce suggests that in some cases obtaining “good enough” answers would be better than perfection. He calls his approach to push ahead the innovation process “Upcycled Segmentation” — it is the title of his article in Quirk’s Marketing Research Review (May/June 2026, read in webpage or magazine format).
The intention in ‘upcycling’ is to cut or hurry-up long cycle times but still give innovation teams adequate and useful guidelines and insights to make necessary decisions in order to proceed. The approach is aimed at “creating purpose-built segmentation hypotheses” — the apparent emphasis is on generating hypotheses (rather than complete solutions) on the basis of “high-signal data” held by the company, such as CRM logs, digital activity (e.g., online, apps), and user experience (UX) transcripts. Concrete segments would be derived through advanced and rigourous segmentation analyses at later stages (i.e., test, validate or substantiate the earlier hypotheses). The key objective of the ‘upcycled’ approach is to provide guardrails for the early-stage decisions.
Tayce outlines in the article “a 5-step framework for faster innovation decisions” that describes his approach. The five steps, brought here in summary, are:
- Step 1: The Decision Audit — screen consumer inputs and keep those that point out concrete ‘problems’ which may lead to actionable product or marketing decisions;
- Step 2: The Data Scavenger Hunt — here is the core of searching for relevant inputs in existing data pools that have not been synthesised before, wherein gathering and analysing those “disparate data points” should reveal “outlines of potential segment schemes” (with respect, for example, to sales & customer experience, product R&D, digital marketing);
- Step 3: Pattern Synthesis (the AI/human mix) — identify themes that tie the various inputs into potential segment concepts in a way that makes sense and creates purpose, recommended to perform these analyses with the help of AI models but under the supervision of a human subject-matter expert;
- Step 4: The Naming Game — while taking the consumer viewpoint, assign labels that describe, for instance, a barrier consumers face or a motivation driving their decision (e.g., “Friction-Averse Optimizers”);
- Step 5: The Micro-Validation Loop — before proceeding to implementation of the results, it is proposed to perform a sort of face validity test by exploring the outcomes of the “upcycled segmentation solution” with a small, targeted group of consumers (“pressure test the logic”), which may also help in illuminating the solution to teams within the company (“support internal communication”).
In Step 1 Tayce suggests an interesting idea: segmenting the (marketing, R&D) decisions and not just the people (consumers). The essence of this objective is to prioritise routes to decisions that would have meaningful implications for segments of consumers. In Step 3 Tayce points to leveraging AI methods and algorithms, particularly Large Language Models (LLMs), apparently applicable for interpreting the verbal content of all the inputs or ‘constructs’ revealed in the previous step and extracting themes. It is unclear whether those ‘themes’ would take the form of a segmentation structure or set a collection of descriptive statements (e.g., concerns, goals, decision drivers) that may constitute the basis for segmentation — more likely the latter. The advice of Tayce sounds reasonable, therefore, that it is critical to involve a human expert for seeing that the analyses with the aid of AI output delivers or leads to “the highest-fidelity recommendations”.
- Comment: One may think of several mixes of provisional segments from which a more concrete scheme or structure (of exclusive or fuzzy segments) may be derived in later analyses — yet it is not clear that the method goes that far. Constructing a concrete model structure of exclusive or fuzzy segments should require a type of an unsupervised analytic AI algorithm; the objective of utilising LLMs is seemingly to generate suitable propositions for segments.
Tayce remarks that Step 4 is not a simple task as it may look and one has to be aware of assigning catchy but trivial names of little informational value. However, this step implies that we are not naming yet segments but components of consumers’ decision-making processes on which segments may be founded. It is understood that by effectively classifying and elucidating the factors underlying consumer decisions, it would be possible to construct more meaningful segments; thereafter, it would enable more practical and impactful product and marketing decisions (i.e., that would have real significance for the targeted consumers).
As we come to the final Step 5 in the proposed framework, it seems that the process provides us with leads and guidelines to plausible segmentation schemes. However, it is hard to see how the term “segmentation solution” qualifies to describe the kind of outcome described (e.g., do we get provisional segments with multi-facet profiles?). As an “upcycled segmentation framework” it could suggest directions to constructing a segmentation solution or model, such as what to focus on and the types of information that should be considered in defining segments. On the bright side, the upcycled segmentation process appears to set foundations that are based on consumer decision making, particularly anticipating what would truly matter to consumers in different segments. That would be a commendable contribution of the upcycled segmentation approach, and in which light it may be presented.
Tayce clarifies how innovation teams and management should view employment of the upcycled segmentation framework: “While upcycling data allows you to move much quicker, I am not suggesting it is an excuse to abandon rigor” (p. 54). Developing a unified and coherent segmentation structure will still be needed (one that “survives real-world pressure”). He raises three precautions to be taken, and that indeed deserve special consideration: (1) the Recency Bias Trap — do not let inputs from a recent qualitative study, even when they seem relevant and insightful at the moment, to automatically undermine or derail multi-year observations established from the company’s larger behavioural datasets — weight them carefully; (2) the Inertia Risk — treat the provisional segments proposed as a bridge to a permanent well-founded resolution, hence do not be tempted to integrate the upcycled segments into the corporate strategy while avoiding the cost of a primary study that would allow a deeper established segmentation solution (taking this risk could result in an “innovation debt” according to Tayce); (3) the Logic Gap — the upcycled provisional segments of consumers require testing and verification of the logic of connections, particularly between the behavioural outcome (“what they did”) and attitudinal grounding (“why they did it”), that underpin the characterization of each segment (i.e., ensure that the segment resonates, capturing a real signal rather than noise).
Tayce dedicates the last part of his article to the transition ought-to-be from the provisional upcycled segmentation propositions to a rigourous segmentation model based on a “gold-standard primary study”. He suggests that the appropriate timing for the shift to a more elaborate primary study usually relies on three factors: capital entangled with the innovation enterprise, its scope, and risk. Tayce further explains the benefits and advantages that can be leveraged from the knowledge captured in upscaled segmentation when developing the stronger-founded final segmentation model. It should give researchers a much-improved starting point that would bring them to the destination model faster and more sensibly. However, he emphasises the importance of keeping an eye on the “behavioural anchors” that would yield a segmentation model better grounded in reality. Tayce also recommends five criteria of professional competencies for choosing a research partner agency to support and consult in performing the upcycled segmentation: data synthesis, augmented intelligence, segment validation, strategic consulting, and iterative design.
Certain ambiguity is still hanging over the article on the nature of the ‘provisional segments’: are they similar to a familiar form of segments or are they more like ‘pre-segments’? That is, numerous propositions may be generated, each suggesting a motive or reason for defining a segment (e.g., a goal or motivation, a concern, an attribute preference, use intended, a task or ‘job’ to accomplish). A meaningful segment would usually embody a combination of such motives or characteristics (i.e., how do the themes come through?). Furthermore, and perhaps more crucially, it remains unclear if and how the upcycled ‘provisional segments’ are connected or related to each other (e.g., on what critical aspects consumers differ between segments, and what are the commonalities within segments, how segments are complementary). The nature of the “segmentation solution” produced (with the aid of AI/LLM) is not sufficiently explained.
One should also keep in mind that the upcycled segmentation is likely to be based on inputs (‘data-points’) about consumers, actually customers, that are in some level of interaction or deeper relationship with the company, its brands and products. That is admissible indeed for creating hypotheses or propositions about segments, in the upcycled preliminary stage, but this will be inadequate for a final model that has to reflect the mindsets, considerations and behaviours of a wider span of consumers who could potentially be interested in the innovative products. Not covering wider target consumer populations for the final segmentation model may also lead to a sort of “innovation debt”. That only stresses more urgently the need to proceed to a final segmentation model built with analytic rigour in a primary study.
Tayce proposes in his article a productive framework for ‘upcycled segmentation’ that can help innovation teams to speed-up and push ahead their NPD process in its critical early stages. Moreover, it would provide an instructive roadmap with clues and propositions to direct the development of a final rigourous segmentation model in a later stage. It is also notable that the foundation is built around aspects of consumer decision-making. Nonetheless, the ‘provisional segments’ possibly have a less tight form and could be less connected. Hence, one has to account for any gaps that have to be mediated between the specifications of segments in the upcycled stage based on existing behavioural datasets and the final stage in a primary study. The responsibility is in the hands of the managerial leadership not to suffice with the early resolution, but leverage the insights gained as they proceed in due time to build a complete and concrete segmentation model for supporting advanced design and marketing decisions.
Consumer / Shopper / Customer Behaviour 