WPS7064 Policy Research Working Paper 7064 Radio Frequency (Un)Identification Results from a Proof-of-Concept Trial of the Use of RFID Technology to Measure Microenterprise Turnover in Sri Lanka Suresh de Mel Dammika Herath David McKenzie Yuvraj Pathak Development Research Group Finance and Private Sector Development Team October 2014 Policy Research Working Paper 7064 Abstract Accurate measurement of stock levels, turnover, and profit- employ, than had been envisaged. Second, the technology ability in microenterprises in developing countries is difficult works reasonably well for paper products, but very poorly because the majority of these firms do not keep detailed for most products sold by microenterprises: on average only records. This paper tests the use of radio frequency identifi- about one-quarter of the products tagged could be read and cation tags as a means of objectively measuring stock levels there was considerable day-to-day variation in read-effi- and stock flow in small retail firms in Sri Lanka. In principle, ciency. Third, a comparison of survey responses and physical the tags offer the potential to track stock movements accu- stock-takes shows much higher accuracy for survey mea- rately. The paper compares the stock counts obtained from sures. As a result, the study concludes that this technology RFID reads to physical stock counts and to survey responses. is currently unsuitable for improving stock measurement There are three main findings. First, current RFID-technol- in microenterprises, except perhaps for a few products. ogy is more difficult to use, and more time-consuming to This paper is a product of the Finance and Private Sector Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at dmckenzie@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Radio Frequency (Un)Identification: Results from a Proof-of-Concept Trial of the Use of RFID Technology to Measure Microenterprise Turnover in Sri Lanka # Suresh de Mel, University of Peradeniya, Sri Lanka Dammika Herath, Kandy Consulting Group, Sri Lanka David McKenzie, World Bank Yuvraj Pathak, World Bank JEL Codes: O12, O16, C81, C93, M41 Keywords: Microenterprises; Survey Methods; Profit and Sales Measurement; RFID. # We gratefully acknowledge funding for this project from the Knowledge for Change (KCP) trust fund, and the assistance of Kandy Consulting Group and Innovations for Poverty Action in project implementation. Institutional Review Board (IRB) approval for this study was obtained from the IPA IRB (#:13November-003). Corresponding author: David McKenzie, dmckenzie@worldbank.org. 1. Introduction In a 2008 nationally representative survey of urban microenterprises in Sri Lanka, 81.3 percent of firms with no employees say they do not keep any accounts for their business. 1 This lack of formal recordkeeping is true of many microenterprises around the developing world, and makes it challenging for researchers to collect accurate data on inventory levels, sales, and profits in such firms (Vijverberg, 1991; Daniels, 2001; de Mel et al., 2009). The large genuine volatility of incomes in such businesses (Collins et al, 2009; Fafchamps et al, 2012) can make recall more difficult, and make it harder to distinguish measurement error from actual fluctuations. A further complication arises in evaluations of interventions, where the receipt of a program (such as access to credit) may affect an individual’s incentives to report accurately, or, in the case of business training, may even change the accuracy with which respondents can report on their business. As a result, many studies of microenterprises suffer from high levels of imprecision and of item-response on profits and sales, or otherwise struggle to measure these concepts at all (see McKenzie and Woodruff, 2014 for a review). Yet accurate measurement of inventory levels, turnover, and profits is crucial for answering many questions of economic interest, such as determining the returns to credit or training, to understanding choices between wage work and self-employment, and measuring levels of poverty and inequality. New technologies have begun to offer the potential to improve measurement in a number of domains (McKenzie and Rosenzweig, 2012), raising the question of whether technology can also provide an objective (not self-reported), accurate, and time- and cost-effective measure of business activity in microenterprises? This paper reports on a trial of the use of radio frequency identification (RFID) tags to measure inventory levels and turnover in Sri Lankan microenterprises. RFID tags are increasingly being used in large U.S. retailers like Kohls, Walmart, LLBean, and Best Buy 2 for inventory management. In principle one can apply the tags to new stock as it comes in, and then use a reader to measure stock levels at any point in time without having to physically scan items one 1 Data from the baseline of the Sri Lanka Longitudinal Survey of Enterprises (SLLSE). See de Mel et al. (2010) for survey details. 2 E.g. http://www.rfid24-7.com/article/kohl%e2%80%99s-deploys-rfid-chain-wide-launches-consumer-engagement- apps/ [accessed 28 July, 2012]. 2 by one as would be the case with bar codes. Measuring the flow of stock coupled with price data then can provide data on sales, which could then in turn be coupled with unit cost or mark-up data to provide a measure of profits. We implemented this process in 22 microenterprises in Kandy, Sri Lanka. We accompanied the tagging of inventories with physical stock-takes, and with survey elicitation of inventory levels from the firm owners. This enables comparison of the accuracy of RFID reads compared to survey responses. In addition, we tested the accuracy of RFID tagging on a larger range of products in our field office in order to provide evidence on which types of products this technology works for best. There are three main findings of this proof of concept trial. First, available off-the-shelf technology is more difficult to use and more cumbersome than we had envisaged, and than is suggested by media accounts of the spreading use of this technology. Setting up the system required a period of fine-tuning and overcoming technical obstacles, and then the time taken to scan inventory levels at a firm was approximately 30 minutes per firm. Second, in terms of proof of concept, our results show that i) it is possible to get firms to agree to use this technology; ii) the technology is able to work reasonably well for paper products and some clothing items; but iii) the read-efficiency of the technology is very poor for many products offered by microenterprises, and varies from day to day. As a result, RFID technology does not enable accurate measurement of stock levels or turnover in most microenterprises. Third, survey questions on stock levels are much more accurate in terms of matching the results of our physical stock counts, providing some reassurance that relying on survey self-reports can yield reasonably accurate measurement. The remainder of the paper is as follows. Section 2 provides an introduction to RFID technology, and discusses how it is currently used by large firms, and how it could be used in principle to provide measurement of turnover and profits in microenterprises. Section 3 provides details of our trial, including the technology used, how we selected firms, how the tagging process worked in practice, and our office trial. Section 4 provides the results, and Section 5 concludes. An 3 online appendix provides photograph and video illustrations of the products used and the tagging process. 2. RFID Technology and Its Use to Measure Inventory Levels, Turnover, and Profits 2.1 RFID technology Radio-frequency identification (RFID) technology uses radio frequency waves to transmit information. 3 The basic technology consists of an RFID tag and a reader. The RFID tag has an embedded microchip which allows it to store data, and an embedded antenna to transmit this information. Each chip contains an electronic product code (EPC) which allows for unique identification of the tags, along with customized information chosen by the user. The reader is a device that has one or more antennas that emit radio waves and receive signals back from the tag. This information can then be linked to a database on a computer. There are two types of RFID tags: active and passive. Active tags have their own battery attached to them, and use this power to constantly emit their own signals. As a result they can communicate over ranges of 100 meters or more. An example is the E-Z Pass used to automatically pay tolls on some roads in the U.S. Active tags can be relatively large in size and can cost $15 to $50 per tag, so are typically not used to track inventories apart from a few large, high-value, items. In contrast, passive tags do not have their own power supply, and instead rely on the radio-frequency energy transmitted by the reader to run the circuitry on its chip and reflect a signal back to the reader. This reflection is a weaker signal, and so the reader has to be much closer to the tag in order to be able to read – typically distances of between 5 centimeters and 3 meters depending on the strength of the antenna (Lee, 2003, Holloway, 2006). The range is larger the larger the antenna, which results in a larger tag. The passive tag is much smaller in size than the active tag (typically the size of a sticker or credit card), and considerably cheaper, averaging 20 to 30 cents per tag. As a result of its size and cost, it is the main type of tag used in inventory and supply chain management. Passive RFID tags have been trialed for inventory-management in several large retailers in the U.S. and U.K., including Walmart, Marks and Spencer, Sainsburys, Dillards, and 3 The description of RFID technology which follows is drawn from Violino (2005), Gaukler and Seifert (2007), and Holloway (2006). 4 Bloomingdales. 4 The main use appears to have been in stockrooms, with these organizations using passive RFID tags to track and inventory large boxes or pallets of inventory. Williams (2008) notes that despite much hype about how RFID would take over retailing, it has been slow to get embraced on the sales floor. However, there have also been several trials of their usage for tracking individual products, with a recent example being fashion store Zara implementing the use of RFID tags in 2014 to track items from factory to point of sale 5. One of the main barriers to more widespread usage at the individual product level has been cost, with the cost of a tag too high to justify use on high volume, low margin goods (Gillmore, 2011). RFID technology in principle offers several advantages over barcodes. In particular, they can be programmed to store more information, which can be unique for each item. Common examples given are the possibility of adding expiration dates to perishable products, and manufacturing batch numbers to pharmaceutical products (allowing easy identification of which items to remove from shelves in case of a recall). Moreover, they do not require line of sight reading, and can enable inventory counts without the need to physically scan each item’s barcode. However, since passive tags rely on transmission of radio waves between the reader and tag and back, there are several factors that can prevent accurate reads (Roberti, 2013). 6 The three main issues are liquids, metals, and tag shadowing. Materials containing a large amount of water absorb radio-frequency energy, so that the tag fails to receive enough energy to reflect back a strong signal. Metal can reflect energy away from a tag, or reflect the tag’s signal away from the reader. Finally, if items are stacked so that tags are lined up behind one another, the first tag can capture the reader’s energy, shadowing the tags on items behind it. The result can be that the first item is read but those behind it are not. Typical descriptions of RFID technology describe these as potential issues, but we have not found any numerical evidence of their importance in field settings of the sort seen in the typical microenterprise, and discussions of the lack of take-up have focused mostly on cost issues (Gillmore, 2011). 4 See for example the set of case studies at http://rfid.auburn.edu/research-papers.cfm. 5 http://www.reuters.com/article/2014/07/15/inditex-zara-idUSL6N0PQ3MY20140715 [accessed August 1, 2014]. 6 These are typically not issues with active tags, which produce their own signals. 5 It is therefore an open question as to whether the promise of RFID can be realized in enabling accurate measurement of microenterprise inventories, or whether these technological issues and operational issues limit its effectiveness. 2.2 How can this technology be used in theory to measure sales and profits? The goal is to enable the research team to measure stock levels and stock flows in the microenterprise without having to require the business owner to do anything at the time of each sale. 7 This can be accomplished in principle through the following steps. Step 1: Itemize the different products carried by the store, print tags, and apply them to the products. This could involve a physical stock-take to determine how many of each item the store has to begin with, or just a listing of all the different products the store carries. Then tags would be printed which would identify the date of tagging, product, and price of the product. For example, a tag could identify a particular product as a large bar of yellow “Sunlight” brand soap, priced at 45 SLR, and tagged on December 1, 2013. Step 2: Immediately after applying all these tags, scan them to obtain a read of the total stock level of the firm on this day t. This provides both information on physical stock numbers of each item (e.g. 8 large bars of yellow “Sunlight” brand soap), and, through using the prices of each item, the value of inventory at date t, denoted Stockt. Step 3: When new stock is purchased by the owner, apply tags to this before it is offered for sale. There are several ways this could be done. Tags could be left with the owner to apply him or herself to the items, so long as they clearly indicate which product they should go on; or research assistants could arrange to meet the owner when new stock is being delivered or purchased and tag this new stock. In the firms in our study, the two most common restocking frequencies were weekly (36.1 percent of items) and monthly (32.2 percent of items), with only 2 percent of items 7 An alternative approach would be to introduce bar code scanners and/or cash registers in an attempt to get business owners to record every transaction. This requires much greater behavior change on the part of firm owners, and we are unaware of studies that test such an approach. 6 being restocked daily. So depending on the type of business, it may be feasible to have research assistants do this new tagging. Denote the value of retagged inventories between t and t+s by Retagt,,t+s. Step 4: Return to the business and scan the tags on day t+s This should ideally be done at the same time of day as the initial read, and will provide a read of the number of items and value of stock levels on day t+s, Stockt+s. Step 5: Calculate sales over the period of s days Sales can then be calculated as: Salest,t+s = Stockt + Retagt,t+s - Stockt+s (1) Step 6: Calculate an estimate of profits based on mark-ups or unit costs Using the price and unit cost for each item, or the mark-up, one can determine the profit made from selling each item. Since sales will be available at the product level, profit on each product over the period of s days can then be calculated and added up. 2.3 Practical Issues to Consider The great advantage of this procedure is that it does not require the owner to have to do anything at the point of sale. That is, we are not reliant on the owner to record or remember every single transaction made, nor to adopt a new sales process such as scanning bar codes at the time of sale. However, there are several practical issues to consider. The first is that it may not be cost- effective or feasible to tag all of the different products sold by the firm. In this case one can then take a sample of the products sold and at least track movements in inventory levels and sales for this subsample, potentially then scaling this up by some elicited sales share to get an approximation of total sales. Second, for some products there may not be a fixed price, with the owner negotiating with each customer. Using an average price charged should still provide a reasonable approximation in most cases, with the opportunity to update this average price at the time of each retagging. Third, the procedure above would treat as sales items which are thrown away or given away or used for home use. The former is more of a concern for highly perishable products, and in principle regular RFID scans of the trash pile could help alleviate this. Items 7 given away would have to be recorded through survey questions, while in principle having the owner keep the tags of items taken for home use could allow recording of this component. How important these issues are will depend on the types of products sold. In most cases the procedure should in theory provide a reasonable approximation, and moreover, not be subject to differential reporting bias between treatment and control groups in experimental interventions. 3. Details of Our Proof-of-Concept Trial 3.1 Technology Used We invited quotes from leading manufacturers of RFID printers and chose to buy the Zebra RZ400 printer ($2950), from their reseller Barcoding based on a combination of responsiveness to our queries and price. This printer is described as an “easy to use, robust, industrial-strength printer” with favorable ratings on industry websites. 8 This was coupled with the NiceLabel software 9 ($450) which is designed for use with leading RFID printers, and which encodes the RFID tags. To read, store, and extract the RFID data from tagged products we used the Motorola MC 3190 Z hand-held RFID-enabled reader ($3900) supplemented with the recommended rAgent mobile software ($1300) for processing of the RFID read data. The total fixed costs of hardware and software were thus $8,600. We purchased 24,000 Alien 9629 2” by 1” RFID passive tags for $5280 (an average of 22 cents per tag). 3.2 Selection of Firms and Products We selected 24 microenterprises operating in markets around Kandy, Sri Lanka to participate in the study. We carried out an initial screening exercise based on revenue, profit, varieties of goods, number of items and percentage of varieties that were non-taggable in eight geographic regions. To be included in the sample, firms needed to have monthly revenue less than 500,000 LKR ($3846), monthly profit less than 100,000 LKR ($769), number of varieties of goods less than 100, number of items less than 2,000 and non-taggable varieties to be less than 50%. The justification for using this screening criteria was to ensure a focus on small scale enterprises where RFID technology would be feasible. In order to select one firm that met all the listed criteria, research assistants had to visit about 4-5 enterprises. Microenterprise owners were told that the purpose of the study was to test the feasibility of a new technology for helping monitor 8 See for example http://www.itpro.co.uk/609036/zebra-rz400---rfid-printer [accessed July 29, 2014]. 9 http://www.nicelabel.com/Solutions/Applications/Label-Design/RFID-label-design [accessed July 29, 2014]. 8 stock, and were offered 5,000 LKR ($38) to compensate them for their time and cooperation in the study. Two of the firms decided to drop out after an initial pilot tagging exercise leaving us with a sample of 22 firms. Of these 2 firms were closed on repeated occasions due to health related reasons and so were dropped during the course of the study, leaving us with complete data on 20 firms. These microenterprises are retail stores with no paid employees. The majority of them sell food and beauty items, with a couple of stores selling plastic goods or cloth. Table 1 provides some basic descriptive statistics of these firms. Although the median firm has been in business for 9 years, 95 percent keep no business records. The median value of stock on hand is estimated at 60,000 LKR ($461), with median monthly profits of 15,000 LKR ($115). In the baseline survey the owners were asked to list their 30 highest selling products, the aim being for us to cover the products that contributed most to their profits. The median firm said that 85 percent of total sales came from these top 30 products. Our field team then did a physical stock-take of these products. The mean number of items sold by a firm in the survey among their top 30 products was 579, with a median of 484 items. Given that existing literature had noted the possibility that liquids and metals can interfere with radio frequency signals to make it difficult to detect the signals from RFID tags, we then excluded items such as canned beverages, products in tin boxes, and juice bottles. In addition we excluded loose products sold by weight (such as spices), to arrive at a list of “taggable” products from among the top 30 most sold items. On average 69% of the top 30 most sold products were taggable. Then based on the stock take and this listing of which were taggable, we printed tags for the firm in our field office, and returned the next day to apply the tags to this selection of products from the firm. On average we tagged 282 products per firm. At this time we also scanned the RFID tags to provide information on the baseline stock level of the selected products in these firms. A detailed multi-media appendix provides a visualization of this process. This includes a flowchart of the set-up process (appendix 1), photographs showing products tagged in the store (appendix 2), and a video of the scanning process taking place (appendix 3). 9 3.3 Scanning, Re-tagging, Surveying, and Physical Stock Count Each product was given a unique 12 digit ASCII string identification number. This is then converted into a 24 digit hexadecimal string that is encoded on the tags. The resulting xml database is uploaded to the memory of the rAgent software on the RFID reader, and bifurcated into 22 “picklists” by the unique store identifiers. Then the field team would go to the store and hold the reader in the vicinity of the tagged goods to scan these items. For each tag detected, the reader runs through the picklist to attempt to find a match, and then records the number of unique RFID tags detected, time-stamping the counter. This was then stored in the reader database, and extracted each day in our field office. Our field team would go to the microenterprise each day during selected weeks to carry out these scans.. Information on new incoming stock was provided to the field team during these daily visits, allowing retagging to take place as required. In our initial design, we had expected the RFID reads to provide an accurate measure which we could then compare survey responses to. We therefore had firms rotating between different types of survey questions: a one week recall which asked item by item for inventory levels and sales of all products tagged, and then one day recall which asked about the three highest selling products (appendix 4 provides the questionnaires). However, it became apparent that the RFID reads were significantly lower than reported in the surveys, and so in order to have a reliable gold standard to assess which was correct, we also implemented physical stock takes. We then use the data for the 14 days for which we have both RFID reads and stock-take data for the firm. 3.4 Field Office Trial of Additional Products We supplement the proof of concept trial in actual firms with testing in the Kandy Consulting Group (KCG) field office. This was done for several purposes. First, it enabled us to test whether the read efficiencies obtained in firms would be higher in a more controlled environment. Second, it enabled us to test a much broader range of products. This included paper and other stationary products, fruits and vegetables, footwear, and higher value items such as a laptop computer, fans, compact discs, and cellphones. Appendix 2 provides photographs of these products. 10 4. Results 4.1 More difficult to use than expected Since this is a proof of concept trial, the first set of results concerns the feasibility of implementing this process. Our (naïve) prior based on online descriptions of printers as easy to use, and of standard desktop printers, was that this should be simple plug-and-play technology that could be easily set up within a day or two. In addition, we were under the impression that since the RFID reader did not have to physically scan a barcode item by item, it would be able to quickly scan the entire tagged inventory. In practice the process turned out to be much more difficult to set up and more time consuming to employ than expected. We purchased the printer, reader, and tags in Washington, D.C., and shipped them to Colombo, Sri Lanka. The printer is large (10.9” width x 13.3” height x 18.7” depth) and heavy (32.4 pounds, shipping weight of 49 pounds). The equipment was held up in customs for over two weeks due to the size of the package. The size of the printer also makes it impractical to take from microenterprise to microenterprise and print tags on location. The set-up process required trial and error to correctly calibrate the printer to correctly print the tags, to figure out which memory bank on the RFID tag to store the product information on, and to configure the software correctly for both printing and reading purposes. Appendix 5 describes this process in more detail. Ultimately we were able to print and read the tags. We did two trials on successive days. In the first we printed 70 tags and were able to successfully read 69 of them, and on the second day we printed 200 tags and were able to read 198 of them. These tags were not attached to any product, so merely were a test of whether the tags were being printed and then read correctly. In addition to being more time-consuming to set-up than anticipated, the time taken to read the scanned product information was much longer than expected. When the reader scans the tags, it searches through a picklist to find each one, looping through each time. It took one to two hours to generate a new picklist each evening to use the next day, and then averaged 10-15 minutes to scan the selected 280 or so items in a firm, and another 15 minutes to process the tags by finding matches against the database on the reader (this was usually done while travelling from one store to another). We were able to scan 20 firms in a day, but this took the entire day. This has obvious 11 implications for the cost of employing such technology. At 22 cents per tag, tagging lots of products per firm can add up, and it clearly would not be cost effective for these business owners to employ this technology. Nevertheless, we could see this being used in some impact evaluations to obtain an objective measure of stock turnover at a tagging cost of perhaps $80-100 per firm that may not greatly exceed the cost of a survey round in some contexts. However, given the high cost ($5,200) of the reader and reader software, if a reader can only manage 20 to 22 firms in a day, then the cost of using this technology on a large number of firms becomes more prohibitive. 4.2 Accuracy of the RFID measurement We present results for the days for which we have RFID reads, physical stock-takes, and survey reports. Table 2 provides the results of the field trial at the product level, while Table 3 aggregates by product category. We see the main items tagged were packets of biscuits, plastic items, clothing, soap and washing powder, and clay and china pots. In total out of 4,773 tagged items physically counted in the stores by our field teams, the RFID reader was only able to read 1,210 items, or 25.4 percent. The highest read efficiency (defined as percent of tags counted which were actually read) is for plastic basins, where we were able to read 76.1 percent of the tagged items, while there are five items for which we were unable to read any product tags at all. Moreover, there is considerable variation in the read efficiency for the same product over different days, as indicated by the standard deviation, and for the same type of product over different brand/product size/store combinations. For example, we see one type of soap bars (sunlight soap small) had 52.5 percent average read efficiency, although with a standard deviation of 18 percent across days. Moreover, five other types of soap bars had read efficiencies below 10 percent. As a result, soap as a category has the second lowest read efficiency in Table 3. If the RFID read efficiency was always the same fraction, then one could re-scale the number of products read to get a more accurate estimate of the true stock on hand. However, as Table 2 shows, the read accuracy varies across products and for the same product over different days. As a result, when we aggregate up, the overall read efficiency varies from day to day. This is shown 12 in Figure 1, where the aggregate read efficiency varies between 7.0 percent and 43.8 percent across days. We then turn to results from our KCG office trial. Our first trial tested similar products as those tested in firms in the field trial. To simulate a firm-like environment some of these products were moved around to a different location within the room, or taken out or added from one day to the next. Appendix 6 reports the results. In total we read 485 of 2,004 tags (24.2 percent), which is very similar to our read efficiency in the field of 25.4 percent. This suggests it is not the field setting that was leading to the low read efficiency. Our second field office test worked with stationary items and office products that we expected to work better with the RFID technology. Again reading took place over several weeks. Table 4 reports the results. In total we were able to read 2,111 tags out of 2,586 (81.6 percent). Figure 2 shows that this greater read accuracy over the standard microenterprise inventory products occurs on every day, and ranged between 70 percent and 90 percent. Importantly it also shows that there is no tendency for performance to worsen over time. Finally, we test a broader range of goods, including 118 products over 6 or more days. Photographs of all these products are provided in Appendix 2, and item-by-item read efficiencies in Appendix 7. We aggregate the results by category in Table 5. Overall we read 1,343 out of 3,609 tags, for a 37.2 percent read efficiency. Footwear was the category with highest overall read efficiency, with us able to read 82.1 percent of tags on average, including 100 percent of the gents’ slippers we tagged. We also had 100 percent read efficiency on our risograph (a high- speed digital printing machine used for printing our survey questionnaires for other projects), and 89 percent efficiency (reading the item on 8 out of 9 days) for our photocopy machine. In sharp contrast, the read efficiency was only 6.4 percent for fresh fruits and vegetables, including zero tags being read for mango, watermelon, coconut, pears, apples, and papaya. It is also worth noting that stationary performs much worse in this last trial than in Table 4. The main stationary items in our last trial were notebooks, which were stacked in a pile with the tags affixed to the covers. The RFID reader appears unable to read the tags of items stacked under several other books, likely reflecting the shadow tagging phenomenon. 13 4.3 Survey accuracy In contrast, simply asking firm owners to report how many of each item they had in stock appears vastly more accurate than using the RFID technology. Table 2 compares the number of items reported by the owner to the number subsequently counted by our field team in the physical stock count. On average the survey measure is 99.4 percent of the actual enumerated amount. Moreover, when we consider this item by item, the median item has survey response exactly equal to the count, and 50 percent of the items have a survey to stock-take ratio between 91.3 and 104.0 percent. Figure 1 shows the survey responses dominate the RFID reads in terms of accuracy on every day for which we have both measures. We acknowledge here the possibility that firm owners may have been paying more attention than usual to these items in their inventory because we had applied tags to them, and because we returned to ask about these items on multiple days. Nevertheless, the results demonstrate that surveys can obtain accurate information on inventory levels, and it would be of interest in future studies to test this further. 4.4 Discussion: Why did RFID underperform and when does it work best? These results demonstrate very disappointing overall performance of RFID tags. A first potential explanation is that we used them incorrectly. Certainly we experienced a lot of difficulties getting this technology up and running, but we were able to print tags and read almost all of them before they were applied to any product, and to get very high read efficiencies on particular products. So we do not believe this can be the main explanation for the poor performance. The more likely explanation appears to be that the technology does not work very well with certain types of products. The literature had pointed to the possibility of interference from liquids and metals. This may have been the problem with some of our products – for example, the packaging on some of the packets of biscuits contains a thin metallic layer, while some of the fruits have high water content. But it also appears that the technology does not work very well when items are stacked in a pile on top of each other, removing line of sight between the tag and the reader. But having goods stacked like this is very common in a microenterprise setting (see videos of the store settings in appendix 3), and if one needs to physically pull out each item and scan it one by one, then RFID offers little advantage over just reading a bar code or physically 14 counting items. These issues may be less severe if more powerful antennae are used. However, we did not want to use larger RFID tags because we did not want the tags to take up too much of the packaging and lead to the microenterprise owners or their customers complaining, and the tags we had were still relatively large in size compared to the products they were being placed upon. As technology improves, presumably it will be possible to use small tags with more powerful antenna. Based on our results, the current technology works best with a few large items like photocopiers for which there is clear line of sight and no other tags, and for stationary items so long as they are not stacked up. This accords with the current most common use of RFID tags in large stores, which is to track large boxes of goods in warehouses – here the tags would be on large paper items, with clear line of sight in reading. 5. Conclusions The fact that most microenterprise owners in the developing world keep no records makes it difficult for researchers to measure inventory levels, profits, and sales. RFID in theory offers a potential way for researchers to overcome this problem and obtain objective measures of stock flow. However, our proof of concept trial finds that currently this technology performs very poorly in practice with the types of goods sold by many microenterprise retailers. Moreover, the technology is relatively complicated and expensive to set up and use. As such, we do not see RFID as a solution to this measurement problem in the near future, despite news accounts of its increasing use in large retailers in developed markets. As a silver lining, our analysis finds that simple survey questions asking microenterprise owners how much they have of each item seem to do very well on average at matching the amount measured by physical stock counts. This will not alleviate all concerns about deliberate or systematic misreporting in response to an intervention or because of tax concerns, but does suggest that fact that owners do not keep records need not itself be a large barrier to obtaining reasonably accurate measures of stock. The challenge then remains for future work to develop and test other objective measures of inventory and sales turnover. 15 References Collins, Daryl, Jonathan Morduch, Stuart Rutherford and Orlanda Ruthven (2009) Portfolios of the Poor: How the World’s Poor Live on $2 a day. Princeton University Press, Princeton, NJ. Daniels, Lisa (2001) “Testing Alternative Measures of Microenterprise Profits and Net Worth”, Journal of International Development 13: 599-614. De Mel, Suresh, David McKenzie and Christopher Woodruff (2010) “Wage Subsidies for Microenterprises”, American Economic Review Papers & Proceedings 100(2): 614-18 De Mel, Suresh, David McKenzie and Christopher Woodruff (2009) “Measuring Microenterprise Profits: Must We Ask How the Sausage Is Made?”, Journal of Development Economics, 88(1): 19-31. Fafchamps, Marcel, David McKenzie, Simon Quinn and Christopher Woodruff (2012) “Using PDA consistency checks to increase the precision of profits and sales measurement in panels”, Journal of Development Economics, 98(1): 51-57. Gaukler, Gary and Ralf Seifert (2007) “Applications of RFID in Supply Chains”, pp. 29-48 in Hosang Jung, F. Frank Chen, and Bongju Jeong (eds.) Trends in Supply Chain Design and Management: Technologies and Methodologies, Springer-Verlag, London. Gillmore, Dan (2011) “RFID in CPG to Retail - What Really Happened?”, Supply Chain Digest, March 10, http://www.scdigest.com/ASSETS/FIRSTTHOUGHTS/11-03-10.php?cid=4298 [accessed August 1, 2014]. Holloway, Simon (2006) “RFID: An Introduction”, Microsoft White Paper, http://msdn.microsoft.com/en-us/library/aa479355.aspx [accessed August 1, 2014]. Lee, Youbok (2003) “Antenna Circuit Design for RFID Applications”, Microchip Technology, http://ww1.microchip.com/downloads/en/AppNotes/00710c.pdf [accessed August 1, 2014]. McKenzie, David and Mark Rosenzweig (2012) “Preface for symposium on measurement and survey design”, Journal of Development Economics 98(1): 1-2. McKenzie, David and Christopher Woodruff (2014) “What are we learning from business training and entrepreneurship evaluations around the developing world?”, World Bank Research Observer 29(1): 48-82 Roberti, Mark (2013) “How Can We Minimize the Failure Rate of RFID?”, RFID Journal, 6 September, http://www.rfidjournal.com/blogs/experts/entry?10719 [accessed August 1, 2014]. Vijverberg, Wim P.M. (1991) “Measuring Income from Family Enterprises with Household Surveys”, LSMS Working Paper no. 84. 16 Violino, Bob (2005) “What is RFID?”, RFID Journal, January 16, http://www.rfidjournal.com/articles/view?1339/ [accessed August 1, 2014]. Williams, Simon (2008) “Zebra RZ-400 – RFID Printer”, http://www.itpro.co.uk/609036/zebra- rz400---rfid-printer#ixzz38sAzWLtx [accessed August 1, 2014]. 17 Figure 1: Comparison of the Accuracy of the RFID Reads and of Survey Measures 100 Percentage 500 0 5 Days 10 15 RFID Read Efficiency Survey Efficiency Notes: RFID Read Efficiency defined as total number of tags read by RFID reader as a percentage of number of tagged products counted in physical stock-take; Survey Efficiency is total number of items of tagged products reported by owner in survey as percentage of number counted in physical stock-take. Results are aggregated across all firms in the field trial. 18 Figure 2: Comparison of the RFID Read Accuracy of Office Products and Microenterprise Inventories 100 80 Percentages 60 40 20 0 10 Days 20 30 Office Products Microenterprise Inventory Items Note: Results from KCG Field Office Trial. 19 Table 1: Baseline Characteristics of Pilot Firms Percentiles Mean StdDev 25th 50th 75th Maximum Business Characteristics Value of Stock on Hand 84143 70512 50000 60000 100000 300000 Number of items for sale among top 30 products 579 395 316 484 730 1820 Proportion of top 30 products that are taggable 0.69 0.15 0.57 0.63 0.80 0.95 Number of items for sale that are taggable 290 187 168 243 365 765 Number of items for sale actually tagged 282 118 184 262 367 543 Weekly revenue from top 30 products 22073 18444 5550 15934 32300 66200 Percent of sales from top 30 products 75.9 21.2 58 85 91 100 Monthly profit 16786 11589 7000 15000 20000 48000 Owner Characteristics Owner is Female 0.24 0.44 0 0 0 1 Age of owner 52.62 11.65 47 53 61 75 Years of Education 10.24 2.59 9 11 13 13 Age of business (years) 14.85 15.33 4 8.5 23 53 Keeps no business records 0.95 0.22 1 1 1 1 Number of paid employees 0.05 0.22 0 0 0 1 Notes: Data is for 22 small firms used in pilot. All amounts expressed in Sri Lankan Rupees (1 USD = 130 SLR). 20 Table 2: Comparison of Accuracy of RFID Reads and Survey Measures in Pilot Firms by Product RFID Accuracy Survey Accuracy Aggregate Aggregate Aggregate Percentage of Std Dev of % Survey to Stocktake Std Dev of Survey to Product Name Product category RFID count Survey Count Stock count Tags Read Tags Read % Ratio Stocktake % ratio Plastic Basin Plastic Items 51 56 67 76.1 40.7 83.6 36.1 Munchee Marie Biscuits Biscuits 176 259 234 75.2 34.1 110.7 16.0 Tikiri Marie (Small) Biscuits 66 104 101 65.3 38.6 103.0 14.5 Rice Sieve Clay (Large) Clay or China 28 33 43 65.1 91.9 76.7 17.5 Dustbin (Small) Plastic Items 266 466 467 57.0 222.6 99.8 7.6 Sunlight Soap Small Soap 42 79 80 52.5 18.0 98.8 2.6 Maliban Chocolate Cream Biscuits Biscuits 84 185 173 48.6 38.1 106.9 11.6 Gold Marie Biscuits Biscuits 73 169 158 46.2 37.9 107.0 44.1 Uniform Whitening die Liquid 17 39 39 43.6 71.9 100.0 0.0 Clay Pot Cover (Small) Clay or China 28 36 69 40.6 36.0 52.2 25.1 Bread Pack Plastic packed food 2 5 5 40.0 50.0 100.0 0.0 Lifebuoy Soap (Large) Soap 9 24 23 39.1 28.7 104.3 11.2 Baby Bathtub Plastic Plastic Items 14 43 42 33.3 18.8 102.4 7.6 Plastic Basin (Small) Plastic Items 87 247 272 32.0 23.3 90.8 19.9 Bucket (Small) Plastic Items 21 69 68 30.9 20.5 101.5 20.6 Cream Cracker Biscuits 150g Biscuits 6 22 22 27.3 31.9 100.0 0.0 Batik Sarong Clothing or Other 54 235 221 24.4 13.1 106.3 12.5 Indian Sarong Clothing or Other 25 96 115 21.7 14.5 83.5 18.4 Surf Excel Washing Powder (Small) Washing powder 20 82 111 18.0 8.2 73.9 39.6 Tea Leaves 100g pack Tea 10 54 60 16.7 17.7 90.0 18.5 Sanitary Towel Clothing or Other 30 170 183 16.4 25.0 92.9 15.1 Diana Biscuits Biscuits 31 203 196 15.8 18.1 103.6 6.3 Lifebuoy Red Soap 9 70 60 15.0 14.6 116.7 20.8 Clay Pot (Small) Clay or China 16 112 145 11.0 10.6 77.2 21.2 Printed Sarong Clothing or Other 14 220 181 7.7 9.9 121.5 68.6 Munchee Bourbon 100g Biscuits 1 14 14 7.1 0.0 100.0 0.0 Bun Plastic packed food 9 138 142 6.3 17.9 97.2 19.7 Baby Soap Soap 5 100 109 4.6 9.0 91.7 25.6 Sunlight Yellow Soap 14 853 807 1.7 2.4 105.7 40.1 Lifebuoy Soap (Small) Soap 1 65 64 1.6 5.8 101.6 8.6 Lux Soap Soap 1 54 66 1.5 3.4 81.8 12.2 Tipitip Pack (Small) Plastic packed food 0 80 80 0.0 0.0 100.0 0.0 Sunlight Washing Powder Small Washing powder 0 100 106 0.0 0.0 94.3 25.6 Green Sunlight Soap (Large) Soap 0 3 3 0.0 n.a. 100.0 n.a. Cream Cracker Biscuits 0 112 112 0.0 0.0 100.0 11.8 Rock Salt Packet Plastic packed food 0 145 135 0.0 0.0 107.4 13.6 AGGREGATE 1210 4742 4773 25.4 99.4 Notes 21 Standard deviation is the standard deviation across days. Data is aggregated over all days and products for which we have survey, stock count, and RFID measures. n.a. denotes no standard deviation available as product only present for one day. Table 3: Accuracy of RFID Reads and Survey Measures in Pilot Firms by Product Category RFID Accuracy Survey Accuracy Survey to Std Dev of Survey Aggregate Aggregate Aggregate Percentage of Std Dev of % Stocktake % to Stocktake % Product category RFID count Survey Count Stock count Tags Read Tags Read Ratio ratio Plastic Items 439 881 916 47.9 34.0 96.2 10.5 Liquid 17 39 39 43.6 71.9 100.0 0.0 Biscuits 437 1068 1010 43.3 27.0 105.7 8.7 Clay or China 72 181 257 28.0 17.7 70.4 19.1 Clothing or Other 123 721 700 17.6 9.9 103.0 17.1 Tea 10 54 60 16.7 17.7 90.0 18.5 Washing Powder 20 182 217 9.2 9.8 83.9 21.8 Soap 81 1248 1212 6.7 5.1 103.0 26.0 Plastic Packed Food 11 368 362 3.0 11.0 101.7 13.4 Table 4: Read Efficiency for Office Products in KCG Field Office Trial Aggregate RFID Aggregate Percentage Tags Std Dev of % Tags Product Name count Stock count Read Read Computer Chairs (Old) 13 14 92.9 8.2 Magazine Files 3 inch 109 122 89.3 24.1 Lever Arch Files 3 inch 60 68 88.2 22.2 Computer Chairs (New) 59 68 86.8 26.2 Duplicating Paper Packet with Printed cover 624 724 86.2 19.0 Paper Bundle (Cat A) 425 494 86.0 20.5 Standing Fans 5 6 83.3 17.7 Stationery pack 78 95 82.1 39.4 Paper Bundle (Cat B) 126 161 78.3 20.4 Magazine Files 4 inch 231 305 75.7 18.8 Duplicating Paper Packet without Printed cover 127 170 74.7 24.9 Paper Bundle (Cat C) 254 359 70.8 24.4 AGGREGATE 2111 2586 81.6 22 Table 5: RFID read percentages for final KCG office trial, by products category Category Name # of tags read total # of tags % tags read std dev(in %) Footwear 133 162 82.10 8.69 Other 62 81 76.54 17.95 Spice packets 61 81 75.31 4.90 Toothbrush and paste 80 108 74.07 10.58 Clothing 53 81 65.43 10.31 Washing powder 49 81 60.49 18.52 High-end items 197 387 50.90 7.02 Soap 79 162 48.77 8.23 Eggs 42 90 46.67 22.36 Clay and China 37 81 45.68 8.69 Plastic items 56 135 41.48 8.68 Office Supply Items 121 324 37.35 4.63 Liquid 56 162 34.57 5.40 Tea 35 108 32.41 2.78 Plastic packed food 143 477 29.98 3.94 Stationary 95 324 29.32 13.11 Fresh Fruits and Vegetables 40 585 6.84 5.86 Biscuits 4 162 2.47 4.04 Dry fruits 0 18 0.00 0.00 Broom 0 18 0.00 0.00 AGGREGATE 1343 3609 37.21 Results represent aggregate numbers for 6 days. For individual products refer to Appendix 7. 23 Online Appendices Radio Frequency (Un)Identification: Results from a Proof-of-Concept Trial of the Use of RFID Technology to Measure Microenterprise Turnover in Sri Lanka Appendix 1: Printing and Reading Tags Appendix 2: Photographs of Products Tested Appendix 3: Video of Scanning Process in Operation Appendix 4: Example of Survey Questions Appendix 5: Discussion of Issues in the Set-up Process Appendix 6: Results from KCG Field Office Trial of Similar Products to Field Trial Appendix 7: Item-by-item read efficiencies from Field Office Trial 24 Appendix 1: Printing and Reading Tags RFID Printer NiceLabel Software • Installation: RFID Testing & • Installation: NiceLabel printer setup softrware software is installed on Calibration instrallation on the PC the PC for the installed • Test trials: Trial runs RFID printer to calibrate the • Configuration: Printer is database association calibrated; RFID read • Configuration: between the printer and write power Software is calibrated and the NiceLabel adjusted, in accordance to align with the size software and also with the RFID tag in use and tyoe of the RFID check the read and tag in use write settings of RFID printer Encoding of data and batch-printing of tags Testing and Calibration RFID Reader RFID reader • Test runs on reading software sample tags encoded • Configuration: • Installation: rAgent using the uploaded Caibration of RFID software installed in database in the read settings to order to store and reader ensure data encoded extract the RFID data • This was done to on tags can be read • Configuration: As a ensure that the first step, RFID tag encoding used by the database is uploaded NiceLabel is in xml format to the compatible with the reader reader 25 Appendix 2: Photographs Selected Field Images Nadu Rice 5kg Packet Rigam Soya 50g Cup, Flower Basket and other plastic items Bat Bags 26 Bun packets Crackers Surfexcel 20g Washing Powder Dustbin Large, Plastic Basin Small Buckets, trash cans and other plastic items Gents Belt (Large) 27 Hats Munchi Gold Mari (Biscuits) Rin Washing Powder Lifebuoy Soap Large (Soap) Signal Large (toothpaste) Clogard Large (toothpaste) 28 Lifebuoy Soap, Surfexcel washing powder Signal Tooth Brush, Comb Large Diva Packet and other items KCG Field Office Products Sorted in Descending Order of Percent Read Risograph – 100% Flat Silipers Gents - 100% White Board – 89% Signal Tooth Brush - 89% 29 Photo Copy Machine - 89% White Board Marker – 89% Sanitary Towels – 87% Marina Cooking Oil – 85% High Heel Shoes for Women – 85% Nestomalt 400g - 83% Ninja Mosquito Coils Packet– 81% Computer Chairs - 81% Ruhunu Chilli Powder 100g- 80% Prime Soya - 78% Lifebuoy Soap 100g - 67% La-O-Jee Tea Leaves 100g- 65% Computer Tables - 65% Twine Cord - 61% Signal Large 70g - 59% CDs with case/cover - 59% 30 Rin washing powder 200g - 59% Telephones - 50% Pears Soap 75g - 48% Spice Packets - 43% Finger Bowls Small - 43% Bun Packets - 39% Fan - 39% Opera Table Salt 400g - 39% Calculator - 37% Super Glue (small) - 37% Raththi Milk Powder 400g - 37% CFL White Bulb - 37% Stapler machines – 35% Sunlight Washing Powder 20 g - 35% Water Pitcher Clay (large) - 33% Egg - 31% Bat - 31% Sunlight Soap 90g - 31% Sliced bread - 30% Rice Sieve Clay (Large) - 28% 31 Drinks Mega Bottle – 26% Atlas pen Blue - 26% notebooks - 23% Punchers - 22% Beetroot - 19% Pumpkin - 19% Maggi Deviled Chicken Noodles - 17% Guava - 17% Kekiri - 17% CD Packs - 17% Dates - 15% Water Pitcher Clay (Small)- 13% Lemon/Lime - 13% Gherkin - 13% Chinese cabbage - 13% Broom - 13% Onion - 13% Beans - 11% Batik Sarong - 11% Mobile phones - 11% 32 Rice five kilo packets - 11% Star Salt Cristal - 9% Eggplant - 7% Cucumber - 7% Tomato - 7% Pomegranate - 7% Wood apple - 6% Pineapple - 6% Carrot - 6% Munchee Super Cream Cracker 190g - 6% Cashew nuts - 6% Blue Uniform die - 6% Strawberry - 4% Potato - 4% Orange - 4% Coconuts - 4% Cabbage - 4% Banana - 4% Mushroom - 4% Avocados - 2% 33 Munchee Hawaian Cookies 100g - 2% Asamodagam Herbal Tonic - 2% Capsicums – 2% Papaya - 0% Meliban Lemon Puff 200g - 0% MDK Coffee 50g - 0% Mango - 0% Laptops – 0% Jumbo Peanuts - 0% Bogawantalawa Tea Leaves 20g - 0% Apple - 0% Samaposha 200g - 0% 34 Appendix 3: Video Video of the RFID scanning process can be found at: https://drive.google.com/file/d/0B9C9RwWKZrUNTWx6SUdnUDVFUXM/edit?usp=sharing 35 Appendix 4: Example of Survey Questions Weekly Recall Questionnaire 36 Daily Recall Questionnaire 37 Appendix 5: Discussion of Issues in the Set-Up Process for Printer and Reader Several weeks of time was spent on getting the encoding of the tags right. We describe some of the key issues here in order to give a sense of the set-up process required and to help guide other researchers considering using RFID. Calibrating the printer: the printer has to be configured according to the types of tag being used. The read and write power of the printer varies with the type of tag, size of the antenna, and its offset. The default settings for read and write power are set to 16, after a series of calibration trials we anchored these at 18 and 26 for read and write respectively. The position of the sensor which detects the IC embedded in the RFID tag is the other major calibration that needs to be done in the printer. This is critical as without the correct positioning the RFID tags would be ‘voided’ or not printed correctly, and the data would not get encoded. For our tags, the sensor was set to forward motion at 24mm. These calibrations were achieved by sending commands using the Zebra Programming Language (ZPL – II), using the communication interface offered by the printer software. Figuring out where to store product information on the tag: A standard RFID tag consists of 5 memory banks or partitions – Electronic product Code (EPC), Access, Kill, User memory and Tag Identification (TID). The user memory bank offers the maximum amount of space, therefore allowing for more variables to be encoded onto the tag. In our first attempt, we therefore used the user memory bank to store the data on the tag in the form of a hexadecimal string. However, we subsequently discovered that the rAgent software (recommended by Motorola for use with its reader) is designed to read the EPC memory bank and not the user bank. Obtaining the correct software: the reader does not come with the software capability to allow extraction of the data being read off the tag. This resulted in further delays while we contacted the manufacturer, obtained the recommendation for the rAgent software, and then procured and installed this software on the reader. Picklist reading: In the initial stages of the experiment, we also encountered some technical difficulty with the recording of processed tags due to picklists not getting updated. Based on the information provided with the rAgent software, the time counter corresponding to a tag in the picklist was supposed to update itself every time that particular tag was read and processed. For example: On day 1 when a tag is processed at, e.g. 11 am, the time counter is updated to ‘day1, 11am’ from its previous value. When the same tag is read by the reader on day 2 at 10 am, the time counter is update to `day2,10am’ from `day1,10am’. However due to some technical issues, the time counter did not get updated resulting in inaccurate RFID reads for the initial stage of the experiment. During this period, we used the same picklist that was made on day one, by reloading prior to scanning at firm on a daily basis. As a result, on any given day the only tags which the reader could process, were those which had not been read even once till that day, 38 leading to a sharp decline in percentage of tags read over time, until we could not read even a single tag. 10 To understand why the same tags that could be read once could not be read again, we consulted with Zebra Printers, Barcoding – the company that provided us the RFID tags, and rAgent. After consultation with the rAgent software technical support, we were advised to circumvent this problem by making a fresh picklist for all 22 firms prior to every field scan. This fixed the problem associated with the resetting of time counter, and allowed us to correctly measure the RFID tags. However, making 22 picklists on rAgent software is a time consuming process, and as a result this increased the time required for the field operations by approximately 1.5 – 2 hours per day. 10 Note we only use data from the period after these issues had been sorted out. 39 Appendix 6: Results from KCG Office Trial of Similar Products to Field Trial The table below shows the read efficiencies from our KCG office trial for similar products as were tested in stores in the field trial. A total of 27 separate read attempts were made: one per day for the first 11 days, and then 2 reads per day for the remaining 8 days. Appendix 6: Read Efficiencies in Field Office of Similar Products to Firm Field Trial Aggregate Aggregate Stock Percentage Tags Std Dev of % Product Name Category RFID count count Read Tags Read Cloth Bags Clothing 42 49 85.71 26.58 Rin Powder 200g Washing powder 53 77 68.83 6.92 Maggi Devilled Chicken Noodles Plastic packed food 72 127 56.69 14.64 Captain Soyameat Chicken Flavour 50g Plastic packed food 28 52 53.85 10.76 Prime Soya Plastic packed food 27 52 51.92 15.15 Ruhunu Chilli Powder 100g Spice packets 27 52 51.92 11.10 Harvest Noodles Plastic packed food 27 52 51.92 30.69 Raththi Full Cream Milk Powder 400g Plastic packed food 63 127 49.61 14.40 Opera Table Salt 400g Plastic packed food 22 52 42.31 25.96 Lifebuoy Soap 100g Soap 22 52 42.31 16.67 Lux Soap Soap 21 52 40.38 21.02 Pears Soap 75g Soap 19 52 36.54 26.14 Porcelain Mugs Clay and China 11 49 22.45 35.67 Star Rock Salt packet Plastic packed food 8 52 15.38 22.22 Marina Cooking Oil 500ml Liquid 7 52 13.46 27.52 Signal Tooth Brush Toothbrush and paste 7 52 13.46 27.52 La-O-Jee Tea Leaves 100g Tea 7 52 13.46 27.52 Maliban Chocolate Cream Biscuits 100g Biscuits 7 52 13.46 27.52 Signal Large 70g Toothbrush and paste 6 52 11.54 22.09 Sunlight Soap 90g Soap 7 127 5.51 12.13 Munchee Hawaian Cookies 100g Biscuits 1 52 1.92 6.67 Bogawantalawa Tea Leaves 20g Tea 1 102 0.98 5.00 Tikka BBQ flavour 6g Plastic packed food 0 127 0.00 0.00 Drinks Mega Bottle (Cream soda+Orenge Crash+EGB) Liquid 0 52 0.00 0.00 MDK Coffee 50g Plastic packed food 0 52 0.00 0.00 Maliban Lemon Puff 200g Biscuits 0 52 0.00 0.00 Samaposha 200g Plastic packed food 0 52 0.00 0.00 Munchee Super Cream Cracker 190g Biscuits 0 52 0.00 0.00 Snack Crackers 30g Biscuits 0 127 0.00 0.00 Munchee Lemon Puff 100g Biscuits 0 52 0.00 0.00 AGGREGATE 485 2004 24.2 Notes Percentages and std deviation are calculated for number of tags read as a percentage of stock-take count, agggregated over 27 reads - one per days for the first 11 days, and two per day for the next 8 days. Except for cloth bags and porcelain mugs which have 26 reads. 40 Appendix 7: Item-by-Item Read Efficiencies in Field Office Trial Appendix 7: Read Efficiency of Individual Products in KCG Field Trial # of tags read total # of tags % tags read std dev(in %) Flat Slippers Gents with box Footwear 27 27 100.00 0.00 Risograph Machine Highend 9 9 100.00 0.00 Signal Tooth Brush Toothbrush and paste 48 54 88.89 18.63 White board marker Office Supply 24 27 88.89 23.57 Photo Copy Machine Highend 8 9 88.89 33.33 White boards Highend 8 9 88.89 33.33 DSI Bata with box Footwear 24 27 88.89 16.67 Female Sanitary Towel Fems (Single) Clothing 47 54 87.04 18.22 Marina Cooking Oil 500ml Liquid 46 54 85.19 10.02 High Heel Shoes Ladies Footwear 23 27 85.19 29.40 Nestomalt 400g Plastic packed food 45 54 83.33 0.00 Ninja Mosquito Coils Packet Other 44 54 81.48 26.93 Computer Chairs Highend 44 54 81.48 17.57 DSI Bata w/o box Footwear 22 27 81.48 17.57 Ruhunu Chilli Powder 100g Spice packets 43 54 79.63 7.35 Prime Soya Plastic packed food 42 54 77.78 8.33 High Heel Shoe Ladies with box Footwear 21 27 77.78 16.67 Lifebuoy Soap 100g Soap 36 54 66.67 14.43 Spice Packets Spice packets 18 27 66.67 0.00 CFL White Bulb Other 18 27 66.67 0.00 Clay Pot (Small) Clay and China 18 27 66.67 0.00 Computer Tables Highend 35 54 64.81 10.02 La-O-Jee Tea Leaves 100g Tea 35 54 64.81 5.56 Finger Bowls Small Plastic items 17 27 62.96 11.11 Sunlight Washing Powder 20 g Washing powder 17 27 62.96 11.11 Twine Cord Stationary 33 54 61.11 8.33 Signal Large 70g Toothbrush and paste 32 54 59.26 12.11 Flat Slippers Gents w/o box Footwear 16 27 59.26 14.70 Rin Powder 200g Washing powder 32 54 59.26 22.22 CDs with case/cover Office Supply 32 54 59.26 29.00 Plastic Chair Highend 51 90 56.67 46.90 Eggs in a box Eggs 27 54 50.00 40.82 Telephones Highend 9 18 50.00 0.00 Rice Sieve Clay (Large) Clay and China 13 27 48.15 24.22 Bat Plastic items 13 27 48.15 17.57 Pears Soap 75g Soap 26 54 48.15 13.03 Opera Table Salt 400g Plastic packed food 12 27 44.44 28.87 Egg Eggs 15 36 41.67 17.68 Bun Packets Plastic packed food 7 18 38.89 41.67 Fan Highend 10 27 37.04 26.06 Calculator Office Supply 20 54 37.04 30.93 Super Glue (Small) Plastic items 20 54 37.04 11.11 Stapler machines Office Supply 22 63 34.92 25.86 Blouse Clothing 6 18 33.33 25.00 Rice Pack (5kgs) Plastic packed food 3 9 33.33 50.00 Sunlight Soap 90g Soap 17 54 31.48 15.47 Raththi Full Cream Milk Powder 400g Plastic packed food 17 54 31.48 5.56 Desktop Highend 17 54 31.48 21.15 Permanent Marker Office Supply 8 27 29.63 11.11 Punchers Office Supply 5 18 27.78 36.32 Guava Fresh Fruits and Vegetables 5 18 27.78 44.10 Notebooks Stationary 62 270 22.96 15.22 Drinks Mega Bottle (Cream soda+Orenge Crash+EGB) Liquid 6 27 22.22 16.67 Cucumber Fresh Fruits and Vegetables 4 18 22.22 26.35 Dustbin Large Plastic items 6 27 22.22 16.67 Water Pitcher Clay (Small) Clay and China 6 27 22.22 16.67 DVDs with case/cover Office Supply 10 54 18.52 5.56 Star Rock Salt packet Plastic packed food 5 27 18.52 17.57 41 # of tags read total # of tags % tags read std dev(in %) Pomegranate Fresh Fruits and Vegetables 3 18 16.67 25.00 Pineapple Fresh Fruits and Vegetables 3 18 16.67 25.00 Beetroot Fresh Fruits and Vegetables 3 18 16.67 25.00 Beans Fresh Fruits and Vegetables 3 18 16.67 25.00 Sliced bread Plastic packed food 3 18 16.67 25.00 Maggi Devilled Chicken Noodles Plastic packed food 9 54 16.67 14.43 Eggplant Fresh Fruits and Vegetables 3 18 16.67 25.00 Ladies Fingers /Okra Fresh Fruits and Vegetables 3 18 16.67 25.00 Mobile phones Highend 6 54 11.11 11.79 Lemon/Lime Fresh Fruits and Vegetables 2 18 11.11 22.05 Pumpkin Fresh Fruits and Vegetables 2 18 11.11 22.05 Chinese Cabbage Fresh Fruits and Vegetables 2 18 11.11 22.05 Blue Uniform die Liquid 3 54 5.56 8.33 Tomato Fresh Fruits and Vegetables 1 18 5.56 16.67 Mushroom Fresh Fruits and Vegetables 1 18 5.56 16.67 Munchee Super Cream Cracker 190g Biscuits 3 54 5.56 8.33 Onion Fresh Fruits and Vegetables 1 18 5.56 16.67 Bellpepper Fresh Fruits and Vegetables 1 18 5.56 16.67 Carrot Fresh Fruits and Vegetables 1 18 5.56 16.67 Gherkin Fresh Fruits and Vegetables 1 18 5.56 16.67 Avocados Fresh Fruits and Vegetables 1 27 3.70 11.11 Asamodagam Herbal Tonic Liquid 1 27 3.70 11.11 Munchee Hawaian Cookies 100g Biscuits 1 54 1.85 5.56 Cabbage Fresh Fruits and Vegetables 0 18 0.00 0.00 MDK Coffee 50g Plastic packed food 0 54 0.00 0.00 Luffa Fresh Fruits and Vegetables 0 18 0.00 0.00 Papaya Fresh Fruits and Vegetables 0 27 0.00 0.00 Orange Fresh Fruits and Vegetables 0 18 0.00 0.00 Apple Fresh Fruits and Vegetables 0 18 0.00 0.00 CD Packs Office Supply 0 9 0.00 0.00 Kekiri Fresh Fruits and Vegetables 0 18 0.00 0.00 Wood apple Fresh Fruits and Vegetables 0 9 0.00 0.00 Strawberry Fresh Fruits and Vegetables 0 9 0.00 0.00 Capsicums Fresh Fruits and Vegetables 0 18 0.00 0.00 Maliban Lemon Puff 200g Biscuits 0 54 0.00 0.00 Batik Sarong Clothing 0 9 0.00 0.00 Coconuts Fresh Fruits and Vegetables 0 9 0.00 0.00 Banana Fresh Fruits and Vegetables 0 18 0.00 0.00 Pears Fresh Fruits and Vegetables 0 9 0.00 0.00 Jumbo Peanuts Plastic packed food 0 54 0.00 0.00 Bogawantalawa Tea Leaves 20g Tea 0 54 0.00 0.00 watermelon Fresh Fruits and Vegetables 0 18 0.00 0.00 Samaposha 200g Plastic packed food 0 54 0.00 0.00 Atlas Pen Blue Office Supply 0 9 0.00 0.00 Cashew nuts Dry fruits 0 9 0.00 0.00 Broom Broom 0 18 0.00 0.00 DVD Packs Office Supply 0 9 0.00 0.00 Laptops Highend 0 9 0.00 0.00 Mango Fresh Fruits and Vegetables 0 18 0.00 0.00 Dates Dry fruits 0 9 0.00 0.00 Plums Fresh Fruits and Vegetables 0 9 0.00 0.00 Potato Fresh Fruits and Vegetables 0 18 0.00 0.00 42