Retailers have a long-standing interest in modeling and in simulating the customer journey. For retailers, modeling may involve the use of a concept model, a data model, or a process model to frame the customer journey relative to their store operations and to parameterize relevant components of the customer journey, for example relative to key performance indicators (KPIs). Simulation is perhaps less widely used by retailers in their day-to-day operations. Simulation facilitates the exploration of what-if questions relative to a model (e.g., questions such as what if new products are introduced, what if a store alters its entrances and exits, what if an advertising campaign is launched to target a new customer demographic?), as well as to fill-in gaps in the model. The latter is very relevant to our discussion here, as a common gap in retailers’ control is the customer journey: a store may realign its layout, but until an actual customer begins to interact with that new design, the retailer cannot reasonably assess what implications result. This is where simulation scenarios based on agent-based models can come into particular usefulness, in allowing the retailer to experiment with synthetic versions of the customers that they seek to influence (whose custom habits are not guaranteed), and approaches that they might have available within their store operations (e.g., staffing, opening hours, location decisions, etc.) for which they do have tangible control. In particular, examination of customer journeys through agent-based modeling allows potentialities of those journeys to be conveniently subjected to experimental pipelines that move from theory through concept to data and modeling to simulation and what-if scenario building. Each part of this pipeline is useful in contextualizing the customer journey against (costly) retail operations, and in connecting the customer journey maps to service blueprints.
Ali and Moulin (2005), in their paper on modeling shopper behavior in indoor malls, mentioned a concern that permeates much of the agent-based modeling literature and that extends to modeling of customer journeys. That concern is worth repeating here: that “there are few MAS-based research studies attempting to simulate human ‘knowledge-based’ behaviors in micro-scale geographic environments (e.g., malls, shops, hotels, airports, etc.).” (p. 445). Ali and Moulin (2005) go on to argue that personal characteristics and spatial behaviors of individual shoppers work hand-in-glove with features of mall environments (in their example) to produce shopping behavior (p. 446). Batty et al. (2003) raised a similar point, arguing that at fine-scale resolutions of streets, people begin to identify “more closely with elements nearer to our everyday experiences” (p. 674). Batty et al. (2003) also discussed how movement may shift in its character at different scales (p. 675).
It is apparent, then, that the customer journey actually comprises quite distinct paths and purposes when considered at different scales. In our review, therefore, we focus on what might be achieved to develop models of the customer journey from the standpoint of individual agency.
(Purely) conceptual models
As we discussed in Sect. 3, the service blueprint is often the primary concept model that retailers utilize when considering how the customer journey may map to the offerings embedded in a service-scape. There may be as many service blueprints in use as there are retailers, but chain stores at least may have a common service blueprint to standardize the customer experience across their many locations. In many cases, a retailer may even tie changes in the service-scape to KPIs via the customer journey. In this way, the customer journey becomes the vehicle that animates the service blueprint and may be considered, purely conceptually, as “agent-adjacent.” (It is also therefore problematic that there are relatively few retail KPIs for the high street outdoors and that large portions of the customer experience along streets in opaque to retailers’ insight.)
Work by Titus and Everett (1995) introduced an incredibly detailed conceptual model of the Consumer Retail Search Process (CRSP), which constitutes portions of the customer journey. A critical contribution of their work was to classify search behavior on the basis of environmental support structures and drivers from consumer behavior. Environmental factors included environmental differentiation, environmental visibility, orientation aids, and spatial configuration (p. 108). Consumer behavior included the characteristics of individual customers or customer types, based on environmental search knowledge and sensitivity, environmental perception, navigational strategy, movement, contact, time pressure, task complexity, and considerations of search satisfaction based on effectiveness and efficiency (p. 108). Titus and Everett (1995) introduced these concepts for indoor retailing, but tailoring them for the outdoors would not involve much retooling of the concept (although feeding data to those concepts for high street settings would be challenging.) Indeed, we can consider that agent-based models could be useful in animating dynamics (and generating synthetic data) atop the types of conceptual models represented in service blueprints or the CRSP introduced by Titus and Everett (1995). In particular, agent-based simulation scenarios could provide (1) what-if experimentation with service-scape design, as well as (2) synthetic scenarios to examine unknown parts of the customer journey in ways that allow retailers to plan for the unseen and unforeseen components of high street dynamics. Aspects of this work have been demonstrated by others through agent-based approaches, albeit over a scattered literature and for applications that are initially removed from retailing. In what follows, we review how one might go about mapping key components and concepts of the customer journey to agents.
Non-stationary retailing
A critical consideration in examining customer journeys through retail high streets is that many vendors and shopping opportunities are not stationary on the high street. Consider, for example, that food trucks and carts selling consumer packaged goods are apt to shift location from day-to-day (or even moment-to-moment) along a high street. In some situations, for example when selling services or subscriptions, sales associates may mingle directly with pedestrian traffic. For applications to these instances, the non-stationary retailer could be modeled as an agent in and of itself, with a dynamic and synergetic (service) journey path of its own that might interplay with (e.g., intercept and interrupt) customer journeys. To our knowledge, agent-based simulations of non-stationary retailers have not been published in the literature, although the problem of positioning non-stationary retail vehicles relative to urban points of interest (POIs) has been covered by location-allocation models (Murray 2018). It would seem, at least anecdotally, that agent-based models of urban parking behavior (Waraich and Axhausen 2012) might be useful in analyzing how non-stationary retailers such as foot trucks might consider locations, although in most cases these retailers require permits and the problem of locating space at the streetscape is relatively unique.
Customer typology
There are natural affinities between typology-type approaches to customer journeys and agent-based modeling, where significant work already presents on classes and ontologies of agent characteristics and agent behaviors relative to urban processes and phenomena (Benenson and Torrens 2004), often at street-level (Torrens 2016a). Batty et al. (2003) discussed how the categorization of pedestrians must necessarily shift as the scale of observation of their behavior shifts. In other words, as you consider finer resolutions of space and time in human behavior, the commensurate typology of that behavior may need to be adjusted because people’s behavior takes on different foci at different scales of the city: “The law of large numbers also breaks down when the phenomena cannot be classified into categories from which general relationships can be inferred” (p. 674). Batty et al. (2003) therefore advocated for agent-based models that focus on mobility rather than location (p. 674). For our considerations of retailing, of course, both mobility and location are important, as shops are indeed tied to location, but they are also dependent on the mobility of high street customers at that location. Considered another way, Batty et al. (2003) essentially argue that at fine resolution one might bump consideration of behavior from the fixed structural characteristics of streets to the dynamic, mobile, and social tapestry of the crowds that populate them. This also implies that at fine-enough resolution, individual customers become relatively unique on a retail high street.
Significant progress has been made in customer journey modeling along parallel threads of inquiry into customer types. In some situations, the customer journey is considered as a high-level concept, used to differentiate the key agencies in customer journey systems, e.g., a service provider, the customer, and an intermediary that facilitates encounters between the two (Berendes et al. 2018) (p. 220). Underneath these classification hierarchies, customer journeys are usually considered for “typical customer behavior” (Berendes et al. 2018) (p. 219) and typologies can be (necessarily) rather simple. For example, Chen et al. (2019) developed a two-class typology of shoppers—those engaged in business or leisure shopping—in their eye-tracking study of customer journeys through airport retail concourses. One might conjure a much more “delicate” (Lee et al. 2013) (p. 902) typology of agents engaged in the customer journey; i.e., delicacy in the detail attributed to customer typology is particularly necessary when marketing efforts are being considered, for which fine-granularity targeting is often cost effective. These detailed classifications of customer journey types and of customer behavior are well suited to agent-based modeling; indeed, agent classes can be devoted to separating-out key differences in specific types of contextual customer agency from a backdrop of typical behavior. In this case, a base class of typical agency could be developed, with more detailed state and rule descriptors to handle specific agency.
For example, Lee et al. (2013) examined “malling behavior” in (indoor) shopping malls, i.e., categories of customer behavior in mall areas between stores. In mall contexts, as in high streets, it is important to note that not all behavior is focused on shopping, although much behavior has potential to transition to shopping. For example, in their survey, Lee et al. (2013) resolved distinct customer journey profiles for eating, “playing” (e.g., gaming), reading, resting, “seeing” (browsing), and shopping (p. 902). They were also able to delineate what they termed to be “store category selection” (i.e., visiting stores of a particular retail category) as well as phases of intra-store movement. Critically, Lee et al. (2013) were able to build a hierarchy of behaviors from these typologies. Although their work was not applied to agent-based modeling, it is perhaps straightforward to see how the hierarchical classification of customer behavior could be used to form agent classes as well as to build transition probabilities between agent behaviors within the context of economic events and conditions. The caveat, in mentioning these examples, is that most malls are indoors and well-controlled environments.
Similarly, conceptually-typological research has been undertaken to study outdoor retailing. Berendes (2019) introduced two value-based classes of customer in their work to examine high street customer trajectories. Berendes (2019) described utilitarian shopping (with an emphasis on efficiency, i.e., applying minimal time on the customer journey path for maximum price savings), and hedonic shopping (prioritizing enjoyment). Both types of shopping were considered as building value along the journey through experiential gains, e.g., through “idea shopping” (building knowledge about trends), “adventure shopping” (shopping-related search as recreation and socialization), and “exploring shopping” (shopping to gather information by browsing) (p. 315). Feng et al. (2020) discussed typologically-adjacent classification schemes for outdoor pedestrian behavior (of which customer journeys could be considered an important component) and they introduced a high-level classification based on behavior. This included strategic behavior, which takes place ahead of a trip and may have long-lasting consequences on the resulting journey, e.g., by determining the destination and activity schedule (p.2). A corollary for the customer journey would be shopping purpose and selection of a high street to shop on. “Tactical behavior” equates to route choice (p. 4) and involves distinguishing between types of spaces to be traversed, objects that may attract and repulse movement, information regarding the route as provided by signs, and the movement patterns of ambient pedestrians. Again, this typology is easily matched to the customer journey. For example, customers may choose routes through a high street based on distinguishing factors that include types of space (public sidewalks with road crossing, pedestrianized areas, public–private spaces such as outdoor malls); objects such as outdoor map displays and street furniture; and rhythms and motifs of ambient shoppers who may be walking around with large numbers of branded bags from a particular store or beverage containers with store logos. The relationship between tactical behavior and street signs is of obvious significance for retail high street customer journeys, which invariably take pedestrians past a large number of advertising and promotional signs as well as displays of retail goods. Lastly, Feng et al. (2020) described “operational-level behavior,” which they considered as pedestrian behavior that comes into play over small bundles of space and time in which pedestrians may need to dynamically react to and interact with conditions that they encounter. Feng et al. (2020) clarified that this can involve interaction with objects that attract, distract, obstruct, and repulse (p. 4); interaction with pedestrians; group behavior with pedestrians that share a “salient social identity and act according to the social norms of that group” (p. 4); and collision avoidance with proactive behavior to avoid future collisions. Many aspects of Feng et al.’s (2020) classification of operational-level behavior have matches in our consideration of touchpoints along the customer journey, including attraction to shops and the goods that they may display at the high street interface, interactions with greeters and sales staff at the entrance of a store, and social effects relative to peer customers in the crowd or egressing into and exiting from stores.
While the aforementioned typologies are largely concept-driven, there have been major inroads in developing empirical classifications of shoppers and journey classes. Millonig and Gartner (2011) used observational tracking, supported by GPS (outdoor) and Bluetooth (indoor) movement tracing, to develop data-driven typologies based on pedestrian movement characteristics. Their results have some synergy with existing typologies from consumer research [particularly the hedonistic/utilitarian typology introduced by Babin et al. (1994)]. Millonig and Gartner (2011) presented a very well-sourced typology of urban shoppers: they identified “passionate shoppers” (who stop often and for comparatively long times, mostly at fashion shops) (p. 13); “convenient shoppers” (who illustrate a higher average speed and display no significant preference for shops) (p. 14); “discerning shoppers” (whom they characterized like convenient shoppers, but with the added note that they like to frequent specialty shops) (p. 14); and “swift shoppers” (who do not stop often and usually visit food shops and supermarkets) (p. 15). Millonig and Gartner (2011) also used quantitative methods to data-match shopper types to utilitarian and hedonistic movement (p. 16–18).
Shop choice and selection
Considered hierarchically, from wide area to hyper-local geography, the selection of high streets and shopping districts is at the top of the customer journey pyramid. By far the most steady stream of agent research into retail location selection has been developed by Timmermans and colleagues and that work has traditionally been based on agent-based implementations of trip and path-planning models that are “micro-simulated” to agents from methods traditionally used in regional science (Arentze and Timmermans 2007), activity-based models (Arentze and Timmermans 2002; Zhu and Timmermans 2011), motion planning (Dijkstra et al. 2011), network models (Han et al. 2011; Ronald et al. 2012), and time geography (Arentze et al. 2010). Such micro-simulation has the advantage of establishing a natural affinity between coarse-level models of urban activity dynamics, and fine-level considerations of shopper journeys within those systems (Dijkstra et al. 2013). But, in essence, micro-simulation is a scaled-down derivative of the original high-level model and in that sense does not usually treat realistic behavioral agency at the micro-scale of shoppers.
Other scholars have investigated shop choice and selection at (or within) this “micro-scale.” For example, Yoshida (2020) introduced an agent-based model for what they termed to be “shop-around behavior.” This might perhaps be best considered as a micro-simulation counterpart to the stream of location-allocation research for route selection developed by Timmermans and colleagues. Yoshida (2020) clarified that their approach attempts to move beyond two common threads of existing route selection models: aggregate (coarse) treatment of space, and Markov-type selection heuristics (inertia-based probabilistic trees) to account for pedestrians that are engaging in choice among routes. These deviations from the usual stream of research have important implications for agent-based models. Attention to fine-scale detail would imply that customers could be exposed to a huge number of choice-points while on the customer journey, and a deviation from inertia would suggest that the transition probabilities for agent states relative to those choice-points would need to be assessed at each time-step in the model (from a discrete span of time t → t + 1), for each discrete state, for each discrete agent. This would yield an absolutely massive state-space for the model to resolve, even for a limited parameterization of agents on a customer journey. Yoshida (2020) argued that this is necessary in shopping models because consumer behavior is very diverse (p. 123). Recalling the law of requisite variety: for a given model to produce faithful dynamics relative to actual high streets, there should be a match between model detail and the equivalent real-world behavior.
Perception
As shoppers and would-be shoppers engage in a customer journey, their path through retail high streets connects them perceptually to their surroundings. Of course, many shops may have selected a high street location because people’s perception of that setting predisposes them to engage in consumption, e.g., themed food districts such as Brick Lane, or fashion districts such as Savile Row (both in London). For example, Lee et al. (2013) discussed how, in their study of indoor mall behavior, “customers spend much more time for visiting than moving” (p. 908). Agent-based models should be well able to accommodate these perceptual factors in simulation, because of their abilities to support sense–reason–act type exchanges between their state input and transition rule functions (Torrens 2018a). Roozmand et al. (2011) introduced a dedicated “perception module” (p. 1082 and p. 1084) in their agent-based model of customer decision-making. It was designed to accept spatial information and to assign meaning to that information, although the mechanisms by which this was achieved were not provided in the paper. In their Format-Store model, Mathieu et al. (2011) allowed agents to perceive signs in stores (through an interaction distance function), which they then used as targets of their movement. In this way, agent-customers “discover” inputs to their journeys through the store, with the result that “customers will therefore always take different paths, even when their shopping lists are similar” (Mathieu et al. 2011) (p. 125). Turner and Penn (2002) focused solely on the relationship between vision and the spatial configuration of built environment in their agent-based model based on space syntax (where “syntax” is interpreted as the logical structural progression of built space). Their approach was based on James Gibson’s idea for “natural vision,” i.e., that perception can be explained through the relationship between the environment and affordances, which are characteristics of the physical environment that facilitate physical behavior (Cutting 1993; Gibson 1950, 1966, 1979). Turner and Penn (2002) appealed for the research community to consider perception as a direct mechanism by which one might “regard the environment as the provider of possibilities rather than as a place to be rationalized” (p. 473). This plea actually speaks directly to issues that we consider in this paper regarding the customer journey. In other words, the retail high street manifests as a service-scape for customers that journey through it, with myriad retail-centered products and opportunities that are embedded within the happenstance of the adjacent social, historical, environmental, and technological substrate of urban streetscapes. Collectively, these factors combine to produce a perceptually deep setting for movement and custom. However, Turner and Penn (2002) were careful to conclude that their “results, though good, also show that a direct perceptual system does not suffice on its own” (p. 487).
Cognition
While the features of retail high streets might help to (and in some cases be designed to) evoke perceptual contacts between pedestrians and service-scapes, the content and meaning of those touchpoints is often critical in helping retailers to align would-be customers’ cognition of their service offerings with the branding of those offerings. Once again, agent-based models have long been used to build cognitive models for pedestrian dynamics in urban context (Frank et al. 2001; Funge et al. 1999; Mohsenin and Sevtsuk 2013; Paris and Donikian 2009; Penn 2003; Raubal 2001b; Raubal and Worboys 1999; Stern and Portugali 1999; Torrens 2016b, 2018b), and it stands to reason that they could be useful for retailing.
Brown (1994) discussed how shopping has been related to work on cognitive maps: “Despite the inevitable variations from study to study, most analyses agree that retailing facilities figure prominently in mental representations of the city centre” (p. 553). For example, Brown (1994) (p. 553) referenced the famous work of Lynch (1960) theorizing that how people conjure their images of city environments might explain their movement, and research by Sieverts (1967) to discern first-order and second-order “cognitive shopping streets” in Berlin. These mental constructs could form the basis for driving the perception of agents in simulation (see our work (Torrens 2015, 2016b, 2018b) on driving agent motion by mental maps, for example). Brown (1994) also pointed out that large retail stores often feature prominently on people’s cognitive maps of cities (what Couclelis et al. (1987) referred to as “anchor points”). Brown (1994) was also careful to point out that customers’ cognition is intricately bound to their “attitudes towards and emotional involvements” (p. 553). In other words, while the customer journey may bring people into contact with retail touchpoints on the high street, it is customers’ own internal agency that determines the cognition that forms around that context. In his review of work by Van Der Hagen et al. (1991), Brown (1994) (p. 556) made a salient comment: that “the decision heuristics employed by shoppers are strongly influenced by the complexity of the extant retailing environment and the idiosyncrasies of the individual location.”
A pedestrian’s shopping motivation and goal can be construed as inducing and driving their cognition along the customer journey (Berendes 2019) (p. 315). Berendes (2019), for example, discussed that utilitarian customers may have pre-settled products that they are seeking out during the customer journey, while hedonically motivated shoppers may be comparatively more open to persuasion through marketing and other recommendation methods (p. 322). Aspects of motivation thus factor into how customers’ cognition of the retail high street shapes their customer journey. Consider, for example, that a utilitarian-motivated customer may frame the geography of the high street through a lens that prioritizes efficiency in movement and eschews distraction. A hedonically motivated journey, on the other hand, may factor-in the atmospherics of a streetscape, media from storefront displays, and the hustle and bustle of a weekend crowd as part of the experience that they are trying to garner from a high street journey. In this way, then, shoppers’ cognition of the streetscape is intricately bound to their shopping behavior, with marked expression in the way that they pursue the customer journey. For example, Yoshida (2020) introduced a useful distinction in shopping behavior that touches on issues of cognition. In Yoshida’s (2020) scheme, “planned action” was used to categorize shoppers’ behavior before they approach the high street: “itemizing of stores to be visited, working out of time budget, ordering of visit so that the visit is ‘somewhat efficient,’ which then informs the route to be taken.” (p. 125). By contrast, “unplanned action” is improvised: “a visit that occurs when something motivates a pedestrian to visit a facility without having chosen an errand in advance” (p. 125). Both may influence customer journeys: planned action may have a strong influence on the brand journey that is pursued, while unplanned action may determine how much of the retail high street shapes that journey once enacted.
While it seems straightforward, at least conceptually, that connections between cognition and the customer journey could be accommodated within agent-based modeling frameworks, we did not find much existing work on the topic. Roozmand et al. (2011) introduced an agent-based model that treated customer decision-making as a function of culture, personality, and “power distance.” In doing so, they expanded on Costa Jr. and McCrae’s (1992) factorial model of personality, social status, and social responsibility (which actually was designed to examine personality disorder) (Roozmand et al. 2011) (p. 1075). They also built on Wilber’s four-class model of consciousness: interior-individual characteristics (desire, drive, emotion), exterior-individual conditions (body and objects), interior-collective states (common knowledge and norms), and exterior-collective factors (the environment and social structures) (p. 1080). Of interest to our paper is that the approach used by Roozmand et al. (2011) mapped these socio-behavioral factors to state update schemes in their agent-based model as part of the activation component for agent decision-making (p. 1075).
It is also salient to mention that retailers themselves also use (their own) cognition of the high street in determining their service-scapes. In a study of retailers’ mental maps of shopping areas, Brown (1987) detailed how the managers of stores view the local retail environment. Brown (1987) discovered that retailers were most aware of magnet stores, then the complementary or non-complimentary associations with adjacent stores, followed by proximity to traditional sources of customer traffic such as parking facilities and offices. Brown (1987) also settled on a 200 m “maximum perceived distance” within which retailers considered generating compatible footfall custom.
Path-planning
Path-planning involves the selection of a course of movement, usually between an origin and a destination or satisfying a chain of destinations. As we discussed earlier in the paper, the physical paths that customers and would-be customers traverse through high streets is one of the central concepts in considering retail customer journeys. Knowledge regarding how customers settle upon choosing particular high street paths is therefore incredibly useful for a variety of retail operations, from consideration of where to site a store to deciding where to place visual advertising. Path-planning is reasonably well developed in agent-based modeling, following decades of work on the problem of robot motion planning (Latombe 1991). It is often framed as a problem of modeling accessible paths through urban settings, and the likelihood of individuals to select among an assortment of paths can be decomposed to very efficient heuristics from computer science, which often take minimal parameterization to evaluate (Dijkstra 1959; Hart et al. 1968). Key in most agent-based path-planning models is that one may assume that a pedestrian is motivated to select a shortest path. However, this is not necessarily the case on retail high streets, where other factors beyond minimizing walking distance understandably come to the fore (Bitgood and Dukes 2006; Garbrecht 1971). Millonig and Gartner (2011) articulated this point well in their examination of wayfinding behavior among shoppers: “For pedestrians, the shortest path does not always represent the optimal route for an individual’s purposes, as studies have revealed that people often forgo to take the shortest path and prefer the ‘most beautiful’, ‘most convenient’ or ‘safest’ path.” (p. 3). Moreover, pedestrians have many degrees of freedom in their movement through urban streetscapes and their paths may often become highly dynamic, reactive, and adaptive, bucking the simple drivers that computational heuristics may suggest as proxies for their planning behavior (Torrens 2016a).
Nevertheless, path-planning agents can be useful in estimating coarse flows of pedestrians between fixed points on a retail high street (for example, between transit stations and anchor stores). Data to parameterize path-planning are often readily available: the number of pedestrian trips that originate at a parking garage, for example, may be available from ticketing kiosk data and the number of people that enter the front door of a store is usually known to retailers. Estimates of the likely flow of pedestrians between these locations can be built from input–output models or from spatial interaction models, leaving potential paths that they may have traversed open to estimation by path-planning heuristics, e.g., greedy search on graphs that represent streets as edges and stores and high street points of interest as vertices.
Turner and Penn (2002) presented a compelling criticism of traditional heuristics for path-planning in agent-based models (which usually rely on graph-based traversal cost functions to drive shortest-path-planning), which is relevant to consideration of specifically customer-centered paths. They posed the question as to whether it is “really plausible that the human brain continually reassesses an internal cost function, or is it that the human in led by less tangible factors—her curiosity or his desire?” (p. 474). They also went on to argue that in overlapping cost environments (such as high streets, which have temporal costs to move through, costs in effort, costs in distraction, costs in exposure to noise, costs in nuisance of crowds, etc.), it is difficult to separate costs. [Recent research in transportation planning has actually begun to explore how the multilayer nature of path heuristics might be accommodated within a cost structure: see the regret-minimization approach by Chorus et al. (2008) and the level-of-effort approach built into agent-based pedestrian walking models by Guy et al. (2010).] Turner and Penn (2002) argued that human movement is “natural movement” (p. 474), driven by affordances rather than the object-based approaches popularized in agent-based modeling that mimicked heuristics from robot motion planning (Latombe 1991, 1999).
Foot traffic
The volume of foot traffic along a retail high street or along sections of a street is usually estimated using some form of spatial interaction model (Daamen and Hoogendoorn 2004), which explains aggregate flow in terms of mass parameters that describe a source generating a flow (e.g., a transit stop or a parking garage) and a sink that attracts that flow (e.g., a particular retail store location), the physical or network distance between the source and sink, and some expression of the friction that can act on that distance to dissuade customers from making journeys of an excessive length. There are numerous theoretical and observational supports for estimating foot traffic flow in these terms. Brown (1994) described the work of Morris and Zisman (1962) to empirically measure pedestrian movement in Washington D.C., which revealed the strong relative influence of traffic to retail from nearby offices (p. 555). Brown (1994) also highlighted a similar conclusion from Ness et al.’s (1969) examination of lunch-time pedestrian traffic in Toronto (p. 555). Nelson (1958) introduced the “rule of retail compatibility” and “theory of cumulative attraction” (see Brown (1994) (p. 554–555) for a discussion). Respectively, the two notions conceptualize the common observation that compatible retail stores can often generate cross-traffic of customers between them (compatibility) and that stores can take advantage of shoppers’ habits of comparing similar goods [termed as “matching” by Brown (1994) (p. 562)] by siting stores of similar retail trade categories in proximity (cumulative attraction). In these two cases, stores essentially generate their own foot traffic dynamics (in terms of aggregate flows) within the micro-scale of the retail high street, and the specificities of the actual shopping experiences and service-scape (above and beyond their trade classification relationship) will explain the fine details of what form any flow among them may assume.
Wayfinding
Waypoints, considered generally, are points in space and time that people consider when planning and executing their navigation. Wayfinding (moving between waypoints, usually by navigating) is perhaps essential to understanding the customer journey (Dogu and Erkip 2000; Gärling and Gärling 1988), as visits to individual stores along the high street shopping trip constitute the significant touchpoints along the entire journey chain. Other significant waypoints along the customer journey could be identified between visits, including street signs, advertising, information kiosks, and features of the high street urban design such as pedestrianized areas, plazas, malls, and so on (Hess et al. 1999; Moudon et al. 1997). Wayfinding has been well treated in geographical applications of agent-based modeling (Frank et al. 2001; Hajibabai et al. 2007; Raubal 2001a, b, 2008; Torrens 2016b, 2018b). There would seem to be some straightforward connections between the work of geographers in this area and interests in customer journeys and touchpoints. For example, consider that waypoints constitute both touchpoints in people’s behavioral geography and in the organization of their movement (wherein the waypoint essentially allows them to check-in with the territory they are moving through) and with the service-scape (e.g., when high street’s offer “you are here” maps (Meilinger and Knauff 2008) of retail sites, often presented by trade type). Although not directly focused on modeling and simulation, Millonig and Gartner (2011) presented a very detailed review of the sorts of location-based data that are available for the support of wayfinding tools for shopping decision-support systems.
Steering
Steering is a large focus of many commercial customer journey mapping systems that work indoors. Generally, such systems use closed circuit television (CCTV) cameras to identify and track individual customers as they move around within a store. The end result is either a “heatmap” or a map of movement traces. These traces essentially illustrate how customers traverse the store by steering as a response to path-planning, locomotion, and interactions with staff and products. The topic of steering is very well covered by agent-based modeling. With the exception of Alasdair Turner’s (Turner et al. 2001; Turner and Penn 2002) work on examining steering in indoor art galleries, our review did not return academic work on steering-based agency for retail environments. It is worth mentioning that many pedestrian-based agent models rely on Helbing’s social force model (Helbing and Molnár 1995) to produce steering in crowded corridor-type settings such as sidewalks. Turner and Penn (2002) are critical of “social force” type approaches to collision avoidance, as is Torrens (2012; Torrens et al. 2012). Turner and Penn (2002) put their argument elegantly when they ask, “does the corporeal human bump through a crowd of corporeal humans, or does the human guide him or herself through gaps in the crowd?” (p. 474).
Proximity
The relevance of touchpoints in docking the customer journey to retail service-scapes brings to the fore the notion that proximity effects may be critical factors in how touchpoints interplay with pedestrians on retail high streets. For example, pedestrians are generally understood to maintain distance from physical objects to avoid collisions (Basili et al. 2013; Collett and Marsh 1974; Cutting et al. 1995; Huber et al. 2014; Kitazawa and Fujiyama 2010), and a personal distance from other people that variously relates to personal and social factors (Adams 1995; Aiello and Thompson 1980; Altman 1975; Hayduk 1983). When encountering touchpoints along a customer journey, pedestrians on a retail high street may depart from their adherence to these buffering effects: they may seek-out contact with things in the retail service-scape. Indeed, the locations along a high street in which pedestrians deviate from norms of collision avoidance—to patterns of contact-seeking along the high street—may be incredibly valuable information for retailers, indicating, for example, where visual advertising might resonate with pedestrians as they pass by [e.g., the spatial reach and temporal reach of outdoor advertising media (Bhargava and Donthu 1999)]. However, separating-out the contextual situation for proximity-based buffering in crowded and often hectic high street scenarios may be difficult. Our review of the literature did not return examples of agent-based modeling being used to examine links between proximity effects and customer journeys. Nevertheless, proximity effects between moving pedestrians in crowds and on street scenes have been well covered in agent-based modeling for applications in animation and special effects and these types of applications could feasibly be adapted for examining journeys on retail high streets. Proximity-driven movement is often used in motion control of animated characters in computer graphics and especially in computer games, where geometry can be employed to produce realistic-looking crowd patterns for large numbers of agents atop very efficient data structures (Stüvel et al. 2017). Adaptive roadmaps (a modification of probabilistic roadmaps used in robot motion planning) are computationally efficient structures for building personal space buffers between agents while they pursue navigation graphs in heavily animated graphics scenes (Gayle et al. 2009; Sud et al. 2007). Proximity effects are usually achieved using some variation of spatial tessellation between agents, based on their relative positions in a scene, while also taking into account the likely movement path that they will take through the same scene (Kavraki et al. 1996). The tessellation can straightforwardly be associated to agency, to account for example for an agent’s attraction to an object or its adherence to personal buffering space in a given street context (Torrens 2016b).
Locomotion
Locomotion is a critical component of the customer journey along retail high streets. That people are in motion along a high street customer journey can be useful information for retailers, as can determination of where, when, and with whom that motion is taking place. Similarly, certain types of locomotion (how people articulate their motion through stride, body language, how they walk while holding shopping bags, etc.) may suggest shopping behavior and actions, as when customers are walking past a store front and slow and begin to gaze at the displays (Burgoon et al. 1986; Neider et al. 2010; Patla 2004; Shimojo et al. 2003). Customers may also display different locomotion patterns that could indicate their demographics (consider, for example, how locomotion differs between children and the very elderly). Our review could not account for much published work that would explore the relationship between high street customer journeys and locomotion. Nevertheless, the topic presents an interesting opportunity for agent development. Locomotion is very well covered in agent-based modeling in animation and computer graphics (see Pelechano et al. (2008) for a detailed review of the field). It is less frequently used in computational social science or even in geography, where attention is usually placed on movement (we may distinguish movement such as path-following, navigation, wayfinding, steering, collision detection and avoidance from motion such as stopping, idling, leaning, reaching, and so on) (see Torrens (2016a) for a review).
We might consider the absence of locomotion and its relevance to the customer journey. Many aspects of the retail journey involve non-movement. Indeed, the parts of the service-scape that allow retailers to take advantage of customer’s behaviors to stop, pause, rest, queue, idle, etc., often align with touchpoint opportunities. Consider, for example, when a customer is walking purposefully along a retail high street but suddenly comes to a stop outside a store display. This example is a canonical opportunity for a retailer to evaluate the effectiveness of that display touchpoint: if a customer subsequently enters the store, the retailer may assume a connection between the display and a subsequent purchase registered at the point of sale. A simple query, of the like, “what brought you into the store today?” may provide enough information to formalize that connection. Lee et al. (2013), in studying (indoor) mall behavior, presented a method for detecting periods of “stay” in customer journeys by analyzing sessions of relative inaction via Wi-Fi localization (so-called Wi-Fi fingerprinting). In essence, their approach used movement traces from Wi-Fi to isolate rough areas of non-movement for customers in indoor shopping malls. Their test cases demonstrated that customers spend more time in browsing and contemplation type activities than they did in moving.
Group dynamics
High street retailers are often interested in groupings of people as they traverse the customer journey. Some of these groupings may be straightforward, as in the example of tour groups that might be led on high street walking tours by a tour guide. In other instances, retailers may be interested in isolating the customer journeys of particular groups of people from the general motif of foot traffic along the high street, e.g., groups of young people walking together past fashion stores. Batty et al. (2003) raised the issue of treating group-level phenomena on streets; they noted that “There is also the somewhat mystical property of large crowds being formed with their own momentum which binds them together and drives their movement. … such herding instincts due to identity of purpose—‘crowd fever’ so-to-speak” (p. 675). They went on to comment, however, that, “There is little descriptive material on which good models of these dynamics might be built, and the interpretations that do exist are not found within mainstream geographical, urban, or architectural analysis.” (p. 676). Regarding the high street customer journey, we might build understanding of group dynamics in two ways that associate directly to retail service-scapes. First, we may consider how customer journeys group at or in the vicinity of touchpoints. Batty et al. (2003) actually mentioned a related phenomenon when discussing crowding events at high street carnivals in London: “Canetti (1962) describes such events as being highly focused on single points of attraction which are spatially associated with agglomerations of individuals.” (p. 676) (Canetti 1962). This notion is already covered in agent-based modeling of pedestrian movement in the outdoor environment, usually based around physics-inspired models that can focus movement patterns around well-understood models of attraction and related continuum mechanics (Treuille et al. 2006) of particle-based crowd motion (Helbing et al. 2000; Liu et al. 2014; Schweitzer 1997). Second, one might perform a grouping of customer journeys based on particular attributes (of the people making those journeys, or of the journey geography that they express). This thread of research has also been picked-up in agent-based modeling of pedestrian movement (Nara and Torrens 2007; Torrens et al. 2012), using in particular trajectory-based classifiers to build typologies of movement paths for synthetic agents, with some experimentation to match those classifiers to real-world urban scenes (Nara and Torrens 2011).
Sociality
The social phenomena that people generate are an essential component of retail high streets. As Bartelheimer et al (2018) remarked: “customer experience accrues over time and is also co-created with other actors” (p. 3). Bartelheimer et al. (2018) made a passing (but very relevant) comment in his paper on community platforms for retail: that “Service ecosystems emerge and continuously adjust themselves based on shared institutional logics, such as actors sharing a common belief system regarding high streets” (p. 3). If we also consider that at least an ordering of retail shopping streets is a recurring feature of people’s cognitive maps of city environments, we begin to see conceptually that we might be able to establish a set of shared cognitive attributes for high streets generally, or even trends regarding specific high streets that are exchanged through the social actions of shoppers on their customer journeys. This is useful for agent-based modeling, in particular, which can use those attributes to form rules. Indeed, agent-based modeling of belief systems is well covered in much of the computational social science literature (Epstein 2007) (see work on belief–desire–intent (BDI) models to drive synthetic pedestrians in simulation (Ronald et al. 2007), for example). As a caveat, we mention the observation by Siebers et al. (2014) that, “in ABM, although most models have been inspired by observations of real social systems they have not been tested rigorously using empirical data and most efforts do not go beyond ‘proof of concept’” (p. 101). As Bartelheimer et al. (2018) pointed out, “there are little (if any) attempts made in research to establish or study social shopping communities in local high street retail” (p. 5). Our review for this paper returned very little research on these social aspects of retailing outdoors; there was also a dearth of existing research on how one might model social aspects of customer journeys indoors.
Fraud and customer journeys
Loss prevention is a major concern for retail operations. Ustun et al. (2006) provided some figures for the cost of fraud to retailers: in 2004, US retailers lost an estimated US $9.1 billion to shoplifting (p. 365), and in the UK, retailers invest US $750 million in security systems annually. As Ustun et al. (2006) described, security is costly for retailers, and understanding the journey that shoplifters might take through a store can assist in estimating likely returns for different security measures along that journey path. This is of relevance to our discussion of high street retailing straightforwardly: if somebody walks out of a store without paying for a product in their possession, it may be grounds for accusing them of theft. Some work on agent-based modeling of shoplifting has been published, and aspects of the research that it represents are relevant to customer journeys. Ustun et al. (2006) introduced a conceptual model, based on an agent methodology, for simulating physical security in shops (supermarkets). Although their model is based on indoor security, it maps neatly to customer journeys, taking into account the physical layout of the store, as well as the (visual) reach of fraud-detecting sensors through the same environment, as well as the paths that security staff routinely take through the store. Lopez-Rojas et al. (2015) introduced an agent-based model, RetSim, that was sourced in sales data to identify fraud in interactions between sales staff and customers of a Swedish shoe retailer (fraudulent use of coupons, returns, and voided sales). An interesting proposition would involve examining customer journeys on outdoor high streets for loss prevention purposes, e.g., customer journeys of shoplifters that pass through multiple stores. We have not seen work of that nature in the published record, although at least anecdotally we are aware that security staff on high streets do coordinate to identify and share information about shoplifters.
Time
The ability to model time (and related dynamics) is a central component benefit of agent-based modeling (Torrens 2009). Time is obviously crucial in considering the customer journey and in mapping journeys to shops’ service-scapes. For example, last-meter logistics of loading and unloading on the high street must often be managed relative to the timing of opening hours and customer rhythms so as not to interrupt the customer experience by physically disturbing customers’ journeys (Glaser 2016). Also, high street retailers are often sensitive to collective timing of customer journeys, particularly surges in visits. Retailers therefore have an interest in tracking the rush hour dynamics of commuting along the high street, trips to and from school, as well as lunch hour dynamics for nearby offices (Hess et al. 1999; Moudon et al. 1997; Ness et al. 1969). In many instances, proximity to other retail sites and their hour-by-hour or weekly dynamics are important for determining likely customer journeys to other stores: examples include traffic to banks and ATMs on Fridays (pay-day) and theater and movie opening times for restaurants. Within the customer journey, timing can also be useful, for example in singling-out the spaces and durations of dwell time, as well as aspects of efficiency in how quickly customers move in and out of a store from one entrance/exit to a high street or another (Spearpoint and Hopkin 2020). For retailers that work with relatively narrow profit margins (such as high street convenience stores selling consumer packaged goods), frequency of transaction is critical to their relationship to customer journeys and so even small savings in transaction time can be actionable (Antczak and Weron 2019). Similarly, advertisers may be interested in determining which parts of the high street customer journey are taking up lots of time; slow-moving crowds of customers may yield a lot of views past physical displays, for example (Garaus and Wagner 2019). While time is well treated in almost all agent-based models, our review did not uncover much work that specifically focused on time as a simulation scenario for consideration of outdoor retailing. Antczak et al. (2020) recently introduced a NetLogo-based (Wilensky and Rand 2015) agent model of queuing within supermarkets, tied to point-of-sale data.
Atmospherics
Atmospherics are often a significant part of the customer experience affecting indoor journeys. These most commonly relate to lighting (Custers et al. 2010), aroma (Morrin and Tepper 2021), product placement (Tan et al. 2018), and design features (Stevens et al. 2019), among other factors. Atmospherics are also important in outdoor high streets; indeed, the atmospheric character of a high street setting may contribute to retailers’ decisions to site their facilities there (Goffmann 1963, 1971). Aspects of high street atmospherics could be directly adapted from indoor store considerations. For example, Choi et al. (2006) considered connections between street lighting and space syntax, and Omer et al. (2015) examined connections between different types of street layout and agent-based movement in simulation. Satoh (2021) has recently introduced an innovative scheme for pairing customer journey to smart signs in a way that allows for the effecting of atmospherics within a store. Satoh (2021) described a digital sign that used radio frequency identification (RFID) sensors to build a contextual and spatially-bound record of a customer’s interaction with it (as a touchpoint) (p. 71). The system then uses an agent running on the local system to activate programs to drive signs in the vicinity. It is straightforward to see how Satoh's work could be extended to the high street and adapted for street-facing signs, for example. Interestingly, the agent programs “migrate” (p. 72) between the touchpoint artifacts through communications, without having to negotiate centralized servers within the broader the retail information system (p. 73). The system is implemented on reasonably light technology: Satoh (2021) demonstrates the concept on Raspberry Pi devices (p. 78). Akhter et al. (2019) introduced a scheme for counting pedestrians on streets using infrared sensors embedded in smart city type systems, which are designed to detect pedestrian humans from a backcloth of dynamic street activity.
Artifacts
In many high street retail settings, the service-scape is landmarked with artifacts. These include outdoor seating areas for dining, street-facing counters, vending machines, outdoor displays and racks for goods, and information kiosks such as “you are here” maps. In these cases, artifacts extend the retail service-scape into the public space of the sidewalk, drawing the retailers’ operations into direct contact with customer journeys. In this way, then, we might consider some artifacts as key touchpoints in the customer journey. For example, through questionnaire surveys, Lee et al. (2009) examined customer use of self-service kiosks and information kiosks (indoors). They concluded that general trends in customer patronage at retail kiosks were related to the experience that was enjoyed at those artifacts. How customers interact with high street artifacts, and then go on (or not) to visit stores would seem to be an area that would be straightforwardly explored by agent-based modeling. We were not aware of any such work in our review. However, Batty et al. (2003) discussed the significance of objects on the streetscape in building their agent-based model of festival goers. They found that at a small-scale (a high resolution of the high street), “elements or objects vary in such a way that temporal dynamics are intrinsic to their representation and explanation” (p. 673).
Location-based services
Many retailers are now experimenting with service-scapes that manifest in the omnichannel that forms at the hybrid of material and virtual retailing. It therefore makes sense that retailers would consider the customer journey as being co-determined between the two. Sometimes this connection straightforwardly separates out pre-purchase components of the customer journey and post-purchase stages. For example, many high street stores facilitate the purchase of goods online with in-store pickup. In the same way, Online stores may accommodate the purchase of goods through their virtual platforms, with options to physically collect them at another retailer’s tangible presence on the high street. More sophisticated connections are also possible across the omnichannel. For example, location-based services that function on top of location-aware technologies that commonly feature in the devices that we carry while moving on the high street facilitate a range of customer touchpoints that may flit between the physical components of customer journeys and their digital counterparts, as well as novel techniques to use digital technologies to augment physical journeys (Niantic Labs 2016).
Millonig and Gartner (2011) have explored the connection between the positional data that location-aware technologies yield and the movement and inferred interest locations (which we could consider as touchpoints) of outdoor shoppers. Their fieldwork to examine the location data of shoppers using spatial clustering analyses showed that inferred movement could be used to classify shopper types (“passionate,” “convenient,” “discerning,” and “swift”) (p. 13–15). In some ways, these types also confer customer journey attributes: customers engaged in discerning shopping may be more apt to browse window displays along a high street, for example. Further, coupled with interview data, Millonig and Gartner (2011) used spatial analysis to build estimates of shopping trip types (“utilitarian” and “hedonistic”) (p. 16–18).
Another stream of agent-based modeling is focused on the development of location-based services as shopping agents. This represents an approach that focuses on the design of broker-type agents [“agent intermediaries” (Tewari et al. 2003)] that serve as shopping assistants on location-based commerce platforms. They are of high relevance to our discussion of customer journeys, because the broker agents are usually tasked to poll high street stores’ inventory systems [what is often termed as “location filtering” (Fano 1998)] to present product availability and pricing to customers via their smart devices, but while they are physically engaged in a customer journey. Various location-based services have also been considered for delivering advertising and coupons to smart devices and customers as they move. Research-oriented systems include Easishop (Keegan et al. 2008), Impulse (Tewari et al. 2003), and Shopper’s Eye (Fano 1998).
Staff-side agency
Staff are a critical component of the service-scape in most indoor customer journeys. In many cases, staff will physically introduce themselves into customer’s paths when they recognize key aspects of a customer’s journey behavior. For example, staff may identify that a customer is searching for goods and offer to assist in identifying the product location. Staff may notice that a customer has a hurried path and offer to assist them in checking out quickly. Shop staff may recognize that a customer journey has taken that customer back to a display multiple times and determine that they have an opportunity to influence a sale. Further, security staff may identify unusual customer journey paths of shoplifters. Agent-based models have been used to explore staff interactions with customer journeys. Of particular note are a series of immersive game environments that allow staff to train with virtual customers (with each customer programmed as a non-player character (NPC) agent). Mathieu et al. (2011), for example, introduced the “Format-Store” environment to assist training shop staff for a variety of interactions with customer journeys. Again, we should mention that this model is designed for in-store training, rather than high streets, but we note it here because of the innovation that it introduces, and because it raises several open research questions about how staff-side interactions might interplay with customer journeys outside the store to extend the service-scape from the inside to the outside. Format-Store was developed as a three-dimensional (first-person gaming type) environment, focused on providing virtual customer scenarios (mostly effected through chat-bot type textual interaction) for customer relationship management (solving customer problems, providing necessary information to customers when requested), managing the store’s physical environment (spill hazards and store lights flickering), and restocking products when shop floor inventory runs low (p. 120). Mathieu et al. (2011) discussed how the gaming environment can assist in experiential learning, particularly in providing what they term as “in situ” learning. Key, in providing realistic situational scenarios, is the development of what we would recognize as customer journeys for the NPC agent-customers in Format-Store: “The customers wandering in the store at any time are merely going to their business—shopping for goods—trying to fulfill internal goals—purchasing items on a shopping list or querying for information—instead of following a scripted behavior.” (p. 122). A particularly innovative aspect to their treatment of simulated customer journeys involved the introduction of “disturbances in the environment” (p. 122), e.g., parameterizing virtual customers with missing information, blocking parts of the store with spills, removing informational cues such as signs, etc. In essence, the model agents were designed with the capability of interrupting the (simulated) customer journey, thereby allowing for training on the staff-side of the service-scape to address how staff can respond. The virtual customers in the model were designed to exhibit traits of being upset and complaining, for example.