 Kia ora koutou, na mihi nui to the conference Kavinas and attendees and na mihi nui to my workmates within the digital directorate here at Te Papa and workmates across Te Papa and na mihi nui to all the companies and individuals who have been involved in the work that I'm presenting today. A number of whom are here today so it's really great to see you welcome and I'm Amos Mann a digital producer here at Te Papa and this is a story about data and it begins in a bug lab. The bug lab exhibition was created by Te Papa working closely with Weta Workshop with Richard Taylor as creative director. So far in its 10 plus year touring life cycle it has completed two very successful runs. It opened at Te Papa in December and it recently finished its run at Melbourne Museum. This was the first exhibition in which Te Papa used digital labels alone to provide information about the specimens and objects on display and it was also the first time Te Papa built visitor activity data collection into almost all of the digital products within the exhibition. I should also mention that I am by no means an expert in data analysis however I do have a large amount of experience of the gallery context in which this data collection took place and as a digital producer on the exhibition I find myself in the best place to present the work which has been undertaken by many many people over the past year. So why collect data and we can look to one of the winning aspirations of Te Papa's digital directorate for an answer to put audiences and their needs at the heart of digital innovation and use insights to continuously improve experiences. More broadly I've come up with these five goal categories. Personally I'm most interested in the last one. How might we use data collection and analysis in creative ways. Data is being used across all these categories and increasingly sophisticated ways within the context of websites and software applications installed on single user personal devices. However there are challenges we need to overcome if we want to apply these web and app data practices to the gallery context and I believe that where there are challenges there are also opportunities opportunities to take new a new and creative approach. Here are some of the main reasons why there are challenges but put simply data collection in gallery is different to the web because we usually aren't talking about single user sessions on personal devices but instead we're usually talking about multi user sessions on public devices and these are these are physicalised experiences they are fixed within a spatial design and we are therefore positioning these experiences within a physical user journey which can be seen as quite different to a browser based user journey. The BugLab exhibition has 15 digital labels integrated into lab tables throughout the exhibition and the concept that I was working with for these was that the labels would act as control panels for the apparatus in which the specimens and objects were under investigation and there are three different types of labels each with a different user interface structure and different information architecture. The specimen label UI involves a photographic representation of the specimen display the label relates to the user taps a bug to reveal its label content the object label type has a simple three-step content structure sort of like a storybook and it begins with a provocative prompt and this is the details label where visitors tap on a point of interest and here's an example of a data log generated through visitor use I should point out that we needed to build a custom event logger for these interactives rather than using something like Google Analytics and this is because partly we partly because we were restricted in the hardware we could use for the touring show we built these as as HTML interactives running on bright sign media players and this was also a first for to Papa using HTML within the bright sign machines we worked closely with to loose to do the build on these projects and we worked with click suite to on the design and UX research and user experience so the activity that you see in that log took place over four minutes and 15 seconds at a glance you might notice that there's an inactivity home return event at the end of the series that I've selected but I suspect that we are seeing more than one session represented here because of the time length which is over three times the mean observed session length for the labels and because about halfway through there's a 15 second gap and then there's a change in behaviour see if you can spot it suddenly there's a lot of image swiping and I reckon we're seeing two different visitors using the interactive consecutively here as I've alluded to analysis of gallery log files has been less than straightforward and or less straightforward than anticipated for example we don't have a good way to detect those as two distinct sessions through automated analysis however some results have emerged through a number of different approaches we've taken to analysis here are some results from analysis that we undertook with the company D exhibit showing average daily sessions for one of the specimen labels and the atlas moth is the winner however it's also the biggest insect on the screen the least popular is the housefly which is also the smallest specimen on the screen the popularity slash size pattern holds true for the first three or four of the largest bugs but then the scarab beetle breaks the trend gaining higher than average sessions per day then the larger puriri moth is this a sign that the scarab beetle is punching above its weight should we display more scarab beetles would we ever replace the puriri moth with more scarab beetles maybe we would or instead we might look to break with the strict adherence to representing bugs at their relative size instead we might present them in a way that's more relevant to the science stories we're trying to tell so here's another example of a potentially useful insight across all the object labels this this data is not is applied it's a mean or averages from average daily sessions across all the the object type labels and we can see that more users ended their session on page 3 then any other page and that page 3 was the last page or I should tell you page 3 was the last page of content in a step-through content structure so we believe that this is an indicator that we have about the right amount of content for the object label information architecture although care must be taken results like this look all too familiar they look like website user data and we mustn't forget that web context is really quite different to gallery context we need to ask how many people were interacting together as a group to generate this data how many people were using the interactive in quick succession beginning their session on a page other than the home page and how many people just kept lovingly tapping the scarab beetle 20 times in a row we can't answer these questions through simple analysis of activity data alone to answer these questions we need to combine we need to combine activity data with other data sets here are some of the standard measures for websites of those when dealing with in gallery context these are the ones that we can measure using activity data alone and not very accurately however we can ask is there a parallel version of these measures can we come up with some new definitions that work in parallel for the in gallery context yes but we need to combine user activity data with other data sets to make these measures for example we ran observational sessions within bug lab and gained session length data and number of visitors per session data and this is data that we can't get from the event logs so we can use these figures gain from observation to extrapolate insights from the event log data in order to gain a measure of how many people are using the labels which is probably a good question that we should be asking so come and see me after the presentation if you want to know about the method I use or I developed to do that extrapolation it's a little bit it's just a little bit boring so here's a good example of combining data sets this is time and use data from another type of interactive in the bug lab not a label it's called wings in motion so the orange line shows weekly exhibition visitor numbers and the bars are the amount of time the interactive was being used as a percentage of opening hours and we can see something kind of interesting here we can see a drop in visitor numbers across weeks for five and six that that orange line starts to drop from its peak and yet time and use remains up around 60% and doesn't drop until the seventh week this might be an indicator that across our busiest period 26 December to 22 Jan visitor numbers were above a threshold where higher visitor numbers did not result in higher time and use for that interactive in other words this points to a measure of the maximum capacity time and use for this interactive and I think it's safe to say that it's about 60% of opening hours per week we could go into a finer grain analysis over that and get a little bit more of a bell curve over each day or over a week is busier on weekends but it averaged over a week 60% of opening hours it is in use somebody is using it when numbers are above wherever that drop sits to in the seventh week and when you think about context and how many and this bulking of visitor numbers that I've been talking about in the middle of the day or at the weekend well 60% of opening hours per week seems to be a pretty good capacity benchmark and I wouldn't be surprised if similar time and use capacity benchmarks were discovered for interactives of a similar nature and a similar location in the gallery and a similar style of exhibition in Wellington in New Zealand a big caveat there when we see an indicated like this what can we do about it well we could increase the number of copies of the interactive thereby doubling the capacity we could increase the visitor throughput by removing the least popular content and thereby streamlining the experience while maintaining satisfaction and we for this one actually we did spot a particular moth that really was not getting much attention at all so get rid of the moth maybe and then but here and I think most importantly I suspect that moving the interactive into a more prominent location might also bump up capacity at to Papa it was positioned a little bit out of the way if you there wasn't other things next to it if you could imagine people taking a bit noticing it's free and then taking a bit of time to walk over there and then start using it and that and so if it was more placed in a in a more prominent location I think we would get higher capacity in fact I would argue that the single greatest influence on the reach or conversion rate of a fixed location digital product within an exhibition is it's spatial positioning within the gallery or to put it another way yes it is a bit of a truism however I do like this statement because it speaks to the difference between exhibition context and online context really clearly the success of a website is not considered to have any connection to the spatial location of the user in relation to their personal device no the reason your website is doing badly isn't because users can't quite reach their phone but as you can see from this diagram for an accurate measure of the reach or conversion rate for an interactive we really need to know how many visitors can close enough to the opportunity to engage with the product before we can start to measure conversion in a meaningful way it would also be good to know how many of them glanced over at the kiosk and maybe even went up to it and decided not to engage which is a behaviour that we can define as bounce by combining these data sets we can start to come up with an in gallery definition of bounce rate and a definition of reach or conversion rate and we would need this kind of data to measure the impact of spatial positioning of a product within the gallery so what if we did move it well how would we know that this or how would we know that's a good position compared to another it's a dynamic scenario as you will know I'm sure to get this data you could sit and I'm talking about catchment zone data there to get that data you could sit and watch and see how many people come within the catchment zone for your interactive which we've done we have done that or you could use another form of automated data collection during the last week of the exhibition at Te Papa the mahuki team breadcrumb installed their centimetre accurate positioning technology throughout the through out the gallery and at the ticket desk they signed up 167 visitors and gave them a trackable tag to wear during their visit visitors were actually very keen to take part in the tracking and very happy to answer survey questions at the end of their visit also so it's a really valuable data set here is the high accuracy walking path for those 167 visitors these technologies can be used to analyse the success of the layout of experiences in relation to wider user journey journeys and also to calculate things like bounce rate and conversion rate which are pretty simple things for the web but yeah this is the dwell time heat map view of that same data there have been some really fascinating hints at areas for further research that have emerged from the study but on a more fundamental fundamental on a more fundamental level if we want that same level of accuracy and granularity that we get from websites and apps within a specialised user journey we need specialised data so far in this presentation we've been looking at ways we might use visitor activity data to reach the first four types of goal categories that I introduced at the beginning however I believe the most interesting future potential is to use visitor data and creative ways and ways that create transformative experiences and ways that empower our visitors and communities there are many kinds of data feedback loops commonly seen on platforms such as Facebook, YouTube, Netflix do they translate into the gallery we've seen a number of museums provide visitors with their own personal visit data such as Cooper Hewett and Mona and this might be this is acting like a souvenir or memento is the data generated from our collective experience in a gallery also of interest museum workers find this kind of data fascinating I take that as a sign that our visitors might find it to be of great value also they might even find that it is empowering and inspiring to consider this collective visitor data in relationship with specimens objects stories and experiences in gallery discussion on this has arisen over the last few months with another team are continue X continue X have been looking at what could be discovered by combining some of the data sources we have they are working on a product called more that as I quote and I'm quoting here that uses artificial intelligence tools to generate real-time visitor insights and predictions more customers can use these insights and predictions to deliver tailored exhibition tailored exhibitions targeted offers and personalised visitor experiences and we've been talking with continue X to look at how we might combine data sets such as location data and interactive event data to create new and dynamic experiences for visitors application of machine learning and artificial intelligence has the potential to take this creative thinking into a future where visitors build a dynamic and responsive lifelong relationship with the museum and its collections and stories through this technology in preparation for this talk I discuss some of these AI and machine learning potentials with my colleagues are Kate Wainless and Richard Hulse Kate as a UX researcher and Richard is a product owner for the digital experience delivery system and we felt that we could more easily accept a recommendation provided by AI then we could a decision made by AI on our behalf a decision kind of started to feel a bit creepy but hang on is it a recommendation actually a kind of decision made on our behalf just as we encounter these issues online we can easily imagine all kinds of dystopian decisions made for us in the gallery and we might be reminded of in an extreme case how from 2001 I wonder what how might decide is the best thing for me during my in gallery experience I might ask why did the museum recommend that painting for me is it because I'm Jewish is it because I'm male is it because I'm 45 am I being privileged am I being excluded we might never know why museum AI made a recommendation AI and machine learning tools often operate as a black box too complex for a human to unpack it's probably too late to just say no to AI but I believe that it's not too late for the glamour sector to ask what is our version of the infamously abandoned motto don't be evil in relation to how we use these kinds of data sets in creative ways in gallery and by meeting these ethical challenges I believe we will see even greater creative opportunities AI personas such as Alexa or Siri are becoming more and more prevalent but I believe that in the transfer of this technology from the home and the personal into the public gallery we will likely face very similar challenges to those that we've encountered over the past year when taking the first steps towards transferring data practice from the web or data analytics practice from the web into gallery context when working with Richard Taylor on developing bug lab one of the key questions he would ask as creative director was what is the conceit of the exhibition the creative result of this line of questioning was an exhibition with a sense of make believe that it was built by bugs what role could AI take in realising this creative approach more fully in this data-driven AI version of an exhibition built by bugs maybe the exhibition personas can listen to you and respond to what you want to tell them and maybe they can work and play with you towards a much much deeper transformative understanding and connection with the world of bugs and I'd like to say great thanks na mihi nui to everyone involved with input to this data discussion thank you very much