1. Introduction
The opportunity for using mobiles to deliver collection content and related information to museum visitors is fast being embraced. Apps for smartphones are proliferating from MOMA to LACMA in the US, from the Centre Pompidou in Paris, to the Reina Sofia in Madrid. But what if these phones could also be used to understand visitor patterns?
The recent Horizon Report by the Marcus Institute for Digital Education in the Arts 2010 Museum Edition examined emerging technologies for their potential impact on and use in education and interpretation within the Museum environment and highlighted four significant trends, in order:
- Rich media, i.e. images, video, audio, augmented reality being increasingly valuable museum assets
- Digitisation of museum collections continuing to require a significant share of museum resources
- Museum visitors expecting to be able to connect to social networks in all places and at all times
- Open content repositories and social networks challenging the roles of museum education.
Alongside these trends, the Report identified some significant challenges arising from them. Amongst them and relevant to this paper is the documentation of the impact of programs delivered via digital technologies; i.e. is the visitor really getting greater value and a more enjoyable visit through this form of content delivery? The Report also identified adoption horizons for new technologies, predicting that the next two or three years will see the adoption of a technology that is fast growing in popularity in the consumer sector; namely, location based services.
This paper brings together a number of these issues to examine how location identifying technologies in mobile phones can provide both quantitative and qualitative information to museum staff. It looks at the evaluative value of this technology. It does not look at the way in which location identifying technologies can be used to provide museum visitors with content linked explicitly to their current location, though this is an obvious benefit.
2. Visitor evaluation
Visitor evaluation is a central tool in the modern museum. Numbers through the turnstiles are still the primary measurement of a museum's success and determine ongoing funding more than any other factor. The level to which visitors will either return for a repeat visit or recommend the museum to others is generally driven by how much the visitor enjoyed the exhibits. This in turn is driven by how much the visitor engaged with the exhibits, often determined by how accessible the exhibits were and the information about them.
How can this be assessed? Various active methods exist, such as conducting exit surveys, or asking visitors to wear lanyards with embedded RFIDs so they can be tracked. This yields useful data but is compromised by the visitor's responding or tracking with the knowledge that their responses/actions are being evaluated - likely to subtly (or not so subtly) alter both. There are also passive methods, a crude one widely used being for a staff member to latch onto an unaware visitor or visitors and track their movements with a felt tip pen on a plan of the museum, the pen's ink causing a bigger mark where it is held still, reflecting more popular exhibits or at least places where the visitor dwells.
This data is then used to modify existing exhibitions or inform the design of new ones. It helps inform design at both the broad level, e.g. it is known that more visitors favour turning right upon entering a gallery and they tend to pay less attention later in a visit; and also at the micro level, e.g. that panels with 150 words are far more likely to be read than those of 250 words.
With new exhibitions costing sometimes millions of dollars, it is not surprising that many museums dedicate 10% of a project's budget to evaluation.
The case for accurate visitor tracking in museums
Meanwhile, with the rise of the museum website as a major resource platform, there is an increasing mismatch in the technological ability to evaluate use of the physical as against the virtual visitor. Where the physical visitor is evaluated in the manual, somewhat haphazard and heavily staff resource dependent methodologies as detailed above, the virtual visitor can be assessed in considerable detail purely by accessing such tools as Google Analytics.
In simple terms, this is achieved by Google observing the IP address of the visitor's computer, following the viewing path, and then aggregating the data. The result is a compendium of data, including visitor numbers to the website as a whole, specific page visits, dwell times per page and for the site, and geographic location of the visitor's computer.
Compare this with a typical physical visit to a museum, where beyond the likelihood of the visitor being counted upon entry, anecdotal information from guides and security staff about visitor paths and exhibition popularity, and exit surveys, there is likely to be very limited knowledge of where visitors went, and the length of their visit.
So the opportunity not only to track the paths of visitors unobserved, but also to record anything from their overall dwell time in the museum to dwell times in front of specific exhibits, their previous visits to the museum and the time between visits, and even the country where their phone is registered, has the potential to dramatically assist museums in areas from exhibit design to marketing. For instance, an exhibit may have cost tens of thousands of dollars to develop but is being ignored by visitors, or a much more significant proportion of visitors than thought may be international, requiring a different focus for marketing plans.
The technology for museums to receive this data currently exists, using mobile phone signals, along with Bluetooth and Wi-Fi. And the reason for the opportunity within museums is the dramatic change in recent years to acceptance of mobiles in museums. No longer are visitors greeted with signs requesting mobiles to be turned off before they set foot inside. Mobiles are now welcomed and seen as a means by which more in-depth and tailored information can be provided to visitors. For example, the contemporary private art museum MONA in Hobart, Tasmania, opened in January 2011, entirely relies on the iPod Touch (1340 of them in total) for visitor content, with no labels on any of the exhibits (http://www.mona.org.au).
Learning from the retail sector
The museum sector is small and consequently does not have the resources to make use of this opportunity. However, significant components of museum operations have synergies with the retail sector, a part of the economy with deeper pockets for exploring new technologies. The museum sector has a history of piggy backing on the technological developments of its retail cousins, whether in the overt area of streamlining their own retail operations (both in the museum shop and on-line) or more subtly in using retail counting systems to accurately count museum visitors.
Shopping malls in particular share many physical characteristics with museums. They are both likely to be large masonry structures with a limited number of entrances, to contain a series of retail or exhibition spaces along with catering areas, joined by large open spaces. Both shopping mall and museum operators want to know where the more and less popular areas are located, what the dominant paths followed by visitors are, how long they spend in catering outlets and retail stores, and how long they spend in the mall/museum as a whole.
For the retail mall operators (known in the business as RAMs or Retail Asset Managers), critical to their thinking is adjacencies and synergies of the retail and catering mix so as to maximise rental income through more intelligent leasing. This in turn allows them to charge top dollar for shops in prime positions, and also control who gets a lease in the first place.
Exploring the retail sector a little further, department stores are in many ways small shopping malls, the departments being the equivalent of tenants. The types of information which the department store managers are seeking include:
Expanding management metrics
- How long do shoppers stay?
- How frequently do they return?
- How many shoppers walk through the store to adjacent spaces?
Quantifying departmental performance
- What are the conversion rates for each section of the store?
Optimising store layouts and operations
- How can floor sets be optimised to improve shopper flow?
- Measuring marketing effectiveness
- How does the marketing increase traffic to the specific targeted department?
- Do marketing campaigns change shopper behaviour in-store?
Optimising labour scheduling
- Helping retailers to understand traffic patterns to each part of the store so as to staff the patterns more effectively
It is not a big leap to see the application and value of this information to the museum sector when the department store is seen as the exhibition/museum.
Potential technologies for tracking
The synergies between the retail and museum sectors are therefore in this instance similar. What then are the technologies being developed by the retail sector for tracking?
GPS is an obvious starting point. It is a technology that is widely used for tracking outdoors, but it has significant limitations indoors, due to the relatively weak signal involved and signal blockage caused by buildings. With extra receivers added to a building it can be made effective, but GPS was never intended for indoor environments and thus will always perform poorly.
RFIDs (radio frequency identifiers) are being widely used in the retail supply chain management processes, both in passive and active (i.e. with builtin battery) form. Whilst the unit cost of a passive RFID is unlikely ever to drop to that of a bar code, and thus the concept of staff-free check out counters is not going to be a reality in the near future, RFID technology is one with which customers/visitors are increasingly familiar and comfortable. Many corporate staff wear an RFID on a lanyard round their necks as part of their firm's security procedures.
This familiarity would therefore make it relatively easy to ask customers/visitors to wear one during a visit. Each RFID is uniquely numbered, so it can be individually tracked. RFID readers would then be placed in multiple locations around the facility. As the RFID came into range the reader would identify it, and transmit back to its base station this data for as long as it is within range. Multiple RFIDs can be read at any one stage.
This data can be stitched together to identify the locations visited by that RFID carrier, the time spent at each, and the order in which they were visited, thus building an assumed track. The data can then be aggregated to build a pattern of customer/visitor tracking.
All this is achievable, but it has a number of complicating issues. It requires contact with the customer/visitor, and therefore staff resources to instruct the carrier, and also can change the normal behaviour of the carrier. It requires assumptions to be made about paths followed as the cost of providing total reader coverage for a shopping facility is impractical due to the large number of readers required to cover for their relatively short read range (c max 5m for passive RFIDs).
The alternative is to use active RFIDs which have a far greater read range (up to 50m), the advantage being that triangulation can be achieved between a number of readers, and an accurate tracking process undertaken. This is a technology already in use in the art security market (http://www.isisasset.com/products/for_the_art_world.htm), but involves significant costs as each active RFID has a unit cost of c US$ 30.
RFID technology is therefore a feasible way of tracking but has substantial downsides around the significant implementation costs, both in staff time and hardware, along with the major issue of requiring interaction with the customer/visitor.
Mobile phones have the immediate advantage that most visitors carry them, and if so, the phones are likely to be on and therefore emitting a signal, whether or not in use. The result is, as is generally known, that the network providers are able to estimate at any time through cell triangulation the approximate location to within about 100m of the mobile unit. This is however too inaccurate for the level of tracking data required to be useful for the retail sector.
3. Using mobile phones for tracking
Whilst tracking information from network providers is not accurate, and anyway would have complicated privacy issues around it even if it were accessible, the signals that are being emitted by a mobile can be harnessed in other ways. Three forms of signals are particularly relevant to this: TMSI, Blue-tooth and Wi-Fi.
Important to all three for tracking purposes is that the subscriber remains anonymous, alleviating privacy issues, and that the subscribers are unaware of their signals being tracked, thus ensuring that behaviour is not modified.
TMSI (Temporary Mobile Subscriber Identity) is the random number assigned to each mobile by the local area. It is used as the recognition method between the mobile and the local base station. TMSIs are frequently changed to avoid the subscribers being identified, and to ensure privacy because the mobile numbers of the subscribers cannot be accessed through them except by the network provider. The only exception to this, it should be noted, is at the moment of initial assignment of the TMSI when the IMSI (International Mobile Subscriber Identity) must briefly pass between the mobile and the base station.
Once assigned, the TMSI is used as the identity for the periodic communication that all mobiles have to undertake with the local base station to regularly report its location. The same TMSI is maintained within a location area, an area that may be covered by tens or even hundreds of base stations.
However, once the mobile passes into a different location area, a new TMSI is assigned; likewise if the mobile returns to the original location area.
The period between required regular communication of the TMSI varies between countries from a few minutes to up to an hour. By picking up these moments of communication using commercially installed antenna and receivers, sufficient in number to triangulate the position, and registering at the same time the signal strength, a location can be identified. The accuracy of that location will vary according to the number of readers installed, but hardware is currently available to cover a shopping mall of 50,000 sq m with 6 receivers to allow for location identification within 1-2m.
By recording each TMSI communication (which comes in the form of a date and time stamp), and stitching together the data, a movement pattern can be estimated, the accuracy of that pattern obviously depending on the frequency of communication. This frequency is critical to the process as if it is too infrequent, visitors who are within range for less than the period may be missed entirely. However, it is possible to trigger more frequent polling of the TMSI signal by arrangement with network providers.
Bluetooth is an open wireless technology for transferring data over short distances. These distances depend on the frequency being used by the transmitter but cannot extend further than 100m. Bluetooth technology is embedded in almost all new mobiles, each of which has a unique identifying number. Bluetooth devices can be queried for their MAC (Media Access Control ) address, a unique identifier that does not change over time. By transmitting a Bluetooth query all Bluetooth enabled phones within range can be identified. A number of pieces of data can be gathered from this, including the country in which the mobile is registered. By frequent contact, which is a characteristic of Bluetooth interrogations (occurring typically each minute), the length of time that the Bluetooth signal is within range can also be used to identify how long the dwell time of the mobile user is within that location.
Bluetooth on its own is not a technology that is suitable to be used to track in this context, but it provides useful supporting data for other signal tracking technologies such as TMSI and Wi-Fi. It does require the Bluetooth functionality of the mobile to be on for tracking, a relevant issue as Bluetooth drains battery strength and is therefore actively turned off by many subscribers. Anecdotal evidence in the US suggests that only about 6% of phones have Bluetooth in discoverable mode.
Wi-Fi is a term that covers a wide range of wireless protocols. In this instance, it refers to the 2.4 GHz frequencies that mobile phones use to access the Internet. This signal network is separate from the network provider's signal, thus enabling the phone to be used whilst accessing the Internet. Most Wi-Fi enabled phones are programmed to undertake a probe request every minute or so, ascertaining what networks are available. It is this probe request that can be identified using mobile Wi-Fi 'sniffers'. In addition, mobile sniffers can send out their own probe requests and all phones within range will respond. Once identified, the phones' Wi-Fi unique identity can be tracked using time and date stamps to build a picture of the path followed and the dwell time at each stage of the path. By keeping track of which wireless access point (WAP) the mobile has roamed to and by analysing signal strength of all nearby WAPs, it is possible to achieve slightly better than room accuracy; namely, 5 to 20 metres.
This technology has its challenges, most substantially around the way in which Wi-Fi radio waves tend to reflect off building elements, thus complicating interpretation of the data. In addition, there is a limit to the distance that mobile sniffers will penetrate though masonry walls, currently about 50m, and also a limit to the number of Wi-Fi signals that can be picked up, currently about 100 mobiles per sniffer unit. The solution to this is to increase the density of Wi-Fi access points, but this substantially increases the cost of installation.
4. Case study
A pilot program to understand the efficacy of mobile phone tracking technology was undertaken using a combination of TMSI and Bluetooth technologies in a US department store. Eleven access points were hard wired into the multi- floored building of approximately 10,000 sq m to allow for a sufficiently granular level of location specific data recovery. The results were gathered over 3 months during 2010.
The technology worked efficiently, but was somewhat limited by being able to pick up only GSM 2G mobile phones, which comprise only 37% of the US market, where the dominant technology is CDMA.
However, the data that resulted were granular enough to be able to be analysed in a way which provides highly useful parallels to the museum sector, as the following questions and answers illustrate.
1) Is a specific marketing campaign effective ? Yes, Doorbusters ( i.e. those specifically responding to marketing campaigns) visited the store more frequently than regular shoppers. And on average, Doorbusters visited the store 2.8 times in three months compared to 1.4 times for regular customers.
However, on average their dwell time was less than normal shoppers - was this due to the store being more crowded, a lower staff-to-shopper ratio, or just a different type of shopper?
Fig 1: Comparison of visitors responding to marketing campaignLikewise, a Back-to-School promotion showed increased shopper traffic,
Fig 2: Weekly review of comparative response to promotional campaign
but decreased dwell time.
Fig 3: Daily comparison of dwell times during promotional campaign
2) Were improved sales achieved by improving the store lay out? Comparative heat maps showed there was marginal difference ( example of heat map generated).
Fig 4: Example of heat map with red representing highest areas of traffic and cream lowest areas of traffic
3) Where did shoppers go during marketing promotions as compared to normal shopping? The Back to School promotion showed that the highest increase in patronage was in the Men's and Women's Apparel departments.
Fig 5: Comparison of visitor traffic by department
4) How long do shoppers stay in the store?
Fig 6: Average dwell time by day
5) What is the path which shoppers predominantly take through the store using 5-minute gradations? This showed, for example, that 50% of shoppers enter through the main doors, and after 5 minutes 53% are still shopping in the Women's section.
Fig 7: Visitor tracking by department in five-minute gradations
6) How often are shoppers visiting the store?
Fig 8: Frequency of visit for return visitors
5. Conclusion
Each of the technologies that have been reviewed, TMSI, Bluetooth and Wi-Fi, provides opportunities for achieving the level of tracking of mobile phones that is going to be valuable for museum evaluation purposes. It is likely that the optimum solution will combine all three technologies.
One function of these technologies that needs to be clearly identified is that the data they collect cannot be used for accurate visitor counting. Despite the universal use of mobiles, there is never going to be 100% tracking achieved, whether due to mobiles not being with visitors, or not turned on. Mobile phone tracking therefore relies on extrapolating the results to build a picture of trends.
But the benefits are that the data is likely to be more accurate than that collected by existing manual tracking methodologies, and considerably more granular. In addition, it is far less staff dependent, as it happens automatically as long as the hardware is functioning, and it provides a wealth of data beyond that collected by current methods.
In many ways the real challenge is not the technology itself, but the analysis of the data received. As the Horizon Report spells out in commenting on the parallel issue of delivery of content through new technologies:
Museums are good at traditional program evaluation, and expect it as a normal component of museum activities. Too often, however, it is the technology that is the presumed focus of assessment in digitally delivered programs rather than the changes in knowledge, attitudes or skills that may result from the activities of the program
As a sector, we must quickly embrace the existing technologies, harness them to the process of visitor evaluation, and work out what the data is telling us to continue to improve the quality of museum visits.
6. References
New Media Consortium.The 2010 Horizon Report: Museum Edition. The Marcus Institute for Digital Education in the Arts. http://midea.nmc.org
