This project report was written as a formal Deliverable (D8.2) of the ICEDIG Project and was previously made available to project partners and submitted to the European Commision as a report. While the differences between these versions are minor the authors consider this the definitive version of the report.

The following text is the formal task description (Task 8.4) from the ICEDIG project's Description of the Action (workplan):

This task will gather the complete costs of constructing and operating DiSSCo as a distributed infrastructure for digitisation. The costs of different methods of digitisation must be identified (i.e. per design alternatives described by task 8.3) and entered in a ‘ Costbook ’ (D8.2). The costs of constructing the infrastructure must be itemised. A basic principle is that the full costs of all construction and operations activities must be itemised, irrespective of any expectation that these elements are already available or could be offered for free (or with a reduced price or as in-kind contribution). Output: ‘ Costbook ’, itemising costs of construction and operation of the infrastructure (D8.2). Services as input material to design the business model (task 8.5).

A variation (narrowing) of the scope of the task description was agreed with the project Coordinator (January 2019), focusing only on the costs of approaches to mass digitisation as practised across multiple museums and avoiding unnecessary overlaps with the work to be done in the DiSSCo Prepare project. This aligns with the objectives of ICEDIG to concentrate on looking at innovations/efficiencies of digitisation, whilst the broader costs of building/operating DiSSCo are better dealt with in the DiSSCo Prepare project; where there is a whole work package (WP4) on financial readiness, including costing of construction and operation. The present task must contribute what DiSSCo Prepare needs for its work on achieving financial readiness.

Following basic cost accounting principles, we identify several components of costs:

Capital costs: Capital costs are fixed, one-time costs incurred on the purchase of equipment, buildings, construction to be used for digitisation. In other words, it is the total cost of bringing a digitisation facility to operational readiness. If in doubt about what to count as capital, a general rule is that if an asset has a useful life of more than one year, it is a capital cost.

While outright purchase of equipment and space is most common, it is sometimes possible to lease assets for a period. The terms of any lease – in particular, whether there is an option to acquire the asset e.g., at the end of the lease – affect whether the cost is treated as capital or as an operating cost.

Operating costs: Sometimes known as running costs or revenue costs; operating costs are the ongoing expenses related to carrying out business, in this case digitisation. Operating costs can be fixed or variable. Fixed costs are unrelated to the volume of specimens digitised. No matter how high or low are the rates of digitisation, fixed costs remain the same. Variable costs, on the other hand, show a relationship (normally linear) between the volume of specimens digitised and total variable costs.

Fixed operating costs: Fixed operating costs are expenses incurred for operating a digitisation facility that are not dependent on the level of usage. These costs are incurred for as long as a facility is operational (but not necessarily operating). No matter how high or low are the rates of digitisation, costs remain the same. Fixed costs can be non-recurring (one-off) expenses, such as replacement parts, or recurring expenses, such as monthly maintenance contract, salaries, building/floor rental, heating and lighting, etc. Sometimes, fixed costs are split into direct fixed costs i.e., those costs that can be easily and directly associated with the facility itself, and indirect or overhead costs (normally, costs of space, electricity, heating, lighting, general administrative staff, etc.) that are incurred by an institution as a whole but which cannot be directly attributed to specific activities, Indirect or overhead costs are normally apportioned on a percentage basis to different departments, facilities, etc.

Variable operating costs: Variable costs are recurring expenses incurred only when digitisation is taking place. They include rated labour costs (i.e., per hour costs of staff carrying out digitisation tasks, who don’t work, or who work on other tasks, when digitisation is not taking place) and consumable materials used during the digitisation process, such as barcode labels. The amount of these costs depends upon the scale of the digitisation activity. The level and type of digitisation affects variable costs. Costs may depend on the amount of data to be recorded, the difficulty of working with that data (e.g., in transcription), and the number of images to be made. Recording just the unique code and taxon name of a specimen takes less time than recording all information available for a specimen. Some specimen categories take longer to process than others.

It can be helpful to consider the marginal costs associated with digitising one additional specimen (or collection). Understanding these costs can be helpful for comparisons between approaches digitising single or small numbers of specimens, mass digitisation and digitisation-on-demand.

When an additional specimen can be digitised for less than the average cost of all previous digitisations of specimens, economies of scale are being achieved. The aim of introducing automation, for example is to force the marginal cost below the long-run average cost, so that the latter eventually falls. Conversely, there may be approaches to digitisation – for example dealing with special requests - where marginal cost is higher than average cost. In this case, a consequence of handling increasing numbers of special requests is potentially higher average costs overall.

Costs of digitisation divide naturally into: i) establishment costs, meaning the upfront costs of building and equipping a digitisation facility, ii) costs of digitising specimens, and iii) costs of preserving that digitised data and making it findable, accessible, interoperable and re-usable (i.e., ‘FAIR’). A cost model identifying the main cost elements within each of (i) – (iii) (explained below) helps us to understand where the significant costs lie.

Nevertheless, different scenarios of digitisation, largely determined on whether digitisation is carried out in-house or outsourced, and at small versus large scale lead to different costs.*1 Differences among scenarios make it hard to collect, generalise and to compare costs, and more so when different modes of preservation are introduced. Thus, when stating costs, it’s essential to clearly state the digitisation scenario to which they relate.

Currently, most known digitisation initiatives fall into the in-house category, incurring capital costs for establishment and operating costs for running the facility. Some digitisation projects are undertaken on an outsourced/contract basis where a per item or total negotiated price is paid to cover the variable costs of digitisation, recoupment of contractor’s capital and fixed costs and provide a profit margin.

For the purposes of the present task we are mainly interested in the costs of establishing and operating in-house facilities but where possible to collect, it is also interesting to gather costs of outsourcing.

As noted, different types of collections have different requirements in terms of handling procedures and technical approaches to digitisation.

Initially we considered to adopt the storage classification proposed by van Egmond et al. (2019), but this was considered to have too many categories of collection type to ask costs for. On the other hand, Cocks et al. (2020) collected a shorter list of collection types (Table 2) that is easier to work with. Most institutions will only be able to provide costings for a few different collection types anyway, and then probably at only a coarse level. This suggests proceeding with the shorter, coarser list, which also allows us to complement the analysis already done and reported by Cocks et al. (2020).

To complement work carried out on present technical capacities of digitisation centres within ICEDIG participating institutions (Cocks et al. (2020)), it is helpful to gather historical costs of digitisation as a baseline for further planning in DiSSCo, moving towards mass digitisation. These are most helpful when gathered on an annual and per item basis related to specific categories of specimens in specific facilities, incurred potentially over several years.

A template for gathering information has been designed (Suppl. material 2). ICEDIG partners having collections holdings (i.e., APM, LUOMUS, MNHN, Naturalis, NHM, RBGK and UTARTU) were asked to complete a template for each of the digitisation facilities they have, and for a representative sample of different specimen/collection types. RBINS completed the template as part of the SYNTHESYS+ Project work on "Report on the cost models for digitisation on demand" to extend the evaluation and cost gathering.

Gathered costs are adjusted to take account of the different purchasing power of money in different economies and represented for the EU as a whole. This adjustment is done using the Eurostat Purchasing Power Parity (PPP) exchange rates to convert costs to an artificial currency called a Purchasing Power Standard (PPS) with which someone could, in theory, buy the same amount of goods and services in any economy. By convention, one PPS is equal to one euro (€) on average for the EU as a whole.

Several approaches to implementing and maintaining the costbook have been considered, including:

• Use of Excel spreadsheets;

• Google Sheets; and,

• Another tool, like Airtable.

In the first instance, gathering of costs has been carried out with a small number of collection-holding institutions that are beneficiaries in the ICEDIG project using an Excel spreadsheet template as first designed (Suppl. material 2). The template was emailed to each of the participating institutions with instructions to complete a worksheet for each digitisation workflow to be costed.

Alternative approaches such as Airtable can be adopted when either a larger number of institutions are asked to provide costs, and/or for budgeting purposes. To test this premise, a pilot workspace was set up in Airtable. The flat Excel template was partially normalised into a relational data structure, and calculated fields added to mirror the calculations in the Excel costbook. A small set of test data was entered into the Airtable tables, and results checked against the Excel template to confirm that calculations had been accurately replicated.

Data were originally received in the form of 22 completed template worksheets (Suppl. material 4) with an additional 13 provided at a later date (Suppl. material 5). To support analysis and visualisations, these data had to be aggregated into a common flattened table structure. The template worksheets were copied manually from their existing files into a new Excel workbook, to simplify the code required for data aggregation and to provide a collated set of the templates in the same file (Suppl. material 3) and in their original format for reference (Suppl. material 4).

A manual process was also used to create a set of descriptive field names for the 82 data fields in the template and to map each field to the row and column of the relevant cell in the template. For future reference, allocating named ranges to the cells when creating the original template would have negated the requirement for this manual step. This is a modification that we propose should be made before the templates are used again.

A short Visual Basic for Applications (VBA) procedure was written and executed to extract the data (Suppl. material 1). This code iterates through the field list for each of the completed templated worksheets, locating the relevant cell according to the mapped coordinates and copying the data into a single worksheet containing a grid of the combined data for all workflows and institutions.

The data were manually transposed into a standard table format, with one column per data field. A pivot table was created using the flattened table as the data source to provide some support for dynamic analysis and visualisations.

The Purchasing Power Standard (PPS) artificial currency has been used throughout analysis to facilitate comparisons.

As illustrated in Table 2, establishment costs for herbarium sheet and pinned insect digitisation lines are different from one another. Even within each category, the gathered data exhibits a range of costs from a few thousands to several tens of thousands of PPS.

Establishment costs are highly variable as is their effect in overall annual digitisation costs. Detailed breakdowns and descriptions of equipment purchased were not given for most of the costbooks, whereas in several cases some additional information was given indicating that costs also included computers, printers and other ancillary equipment. This makes it hard to understand what the costs really cover and the variations between institutions. Because of this the numbers mask differences in the kinds of equipment purchased so comparisons can be made only cautiously.

In the case of herbarium digitisation, the gathered costs mainly relate to equipping a single workstation; yet in one case it is known that an automated conveyor system was included, and in another case, it is known that a high-capability/resolution scanner was purchased. Nevertheless, the average and median costs are similar, with a range of €26,000 – €38,000 PPS as a typical workstation cost. When an integral conveyor system is included, the cost is higher.

Pinned insect lines show a greater variability across the range of reported establishment costs. Insect lines are one area subject to much recent innovation in attempts to increase throughput, and thus a greater variety of novel equipment solutions have been purchased and tried. It’s not possible to give a typical cost for establishing a pinned insect line, except to say that for static (low throughput) solutions the equipment costs are typically low – basically a few thousand PPS for camera(s) and lighting, whereas introducing automation via a conveyor system for higher throughput substantially increases costs (by an order of magnitude).

For several digitisation capabilities, insufficient data was returned to give any credible picture of establishment costs for other collection categories. One outlier worthy of note is a setup composed of a specialised fluorescence/brightfield slide scanner and research microscope for digitisation of microscope slides. This cost more than €150,000 PPS.

In common across all institutions and regardless of digitisation workflow/capability is the observation that establishment costs focus almost solely on equipment purchase and to a lesser extent on costs of acquisition and upgrade. Few non-equipment elements of the expected costs of establishment – such as building/workspace renovation costs, new furniture, electrical work, etc – were reported. This suggests either that such costs are not frequently incurred or (more likely) that such costs are unknown or cannot be accurately accounted for after the fact.

Space requirements for equipment range from 10m² – 65m² with average and median of 29m² and 25m² respectively. 15m² – 20m² seems to be a typical amount of space needed for these kinds of digitisation facilities, with conveyor systems needed larger spaces.

Finally, depreciation periods for such equipment are typically stated as 5 or 7 years, indicating that respondents consider this to be a reasonable lifetime for such investments (even if actual lifetimes are sometimes longer).

Establishment costs are one-off costs, normally funded out of capital budget, infrastructure development or project grants. Depreciation is therefore used as an element of the fixed costs calculation to give a truer reflection of the actual cost of digitising specimens. Depreciation costs vary, depending on the original establishment cost and the chosen depreciation period.

Fixed costs are unrelated to the volume of specimens digitised. No matter how high or low are the rates of digitisation, fixed costs remain the same. Table 3 shows the fixed costs of each institution as a proportion of their overall annual digitisation costs for the relevant line, while Table 4 shows how the various components of fixed costs contribute to the total fixed costs across the institutions.

Fixed staff cost made up the largest percentage of total fixed costs. Some institutions factor staff into fixed costs (e.g. NHMUK where digitisation staff are largely on long term contracts) while others consider it a variable cost depending on the finance structure that supports the role. Every institution reported fixed-term staff except for RBINS and every institution reported variable cost staff except for the NHM. Among the institution that report fixed cost staff, the average number of staff was 0.84 with a maximum of 2.5 and the total annual labour cost ranged from €1,798 – €124,025 PPS, the highest case of which was MNHM’s outsourced workflow for ReColNat.

Labour was considered a factor in both fixed and variable costs. When considering the impact of staff costs on overall annual cost, it is important to note that some institutions may have entered the same staff member across multiple sheets, thus ‘double counting’ both the number of staff and the cost associated with that staff. This should be taken into consideration when considering institution-level costs and, in future developments of this analysis, should be re-assessed.

There were two sources of variable costs that were measured in this analysis – variable cost labour and the cost of consumables. Table 5 shows the variable cost of each institution as a percentage of total annual costs.

Where labour is considered a variable cost, it makes up a significantly larger percentage of variable costs than consumables (although the potential for double-counting should be taken into consideration). Labour costs were calculated by number of staff, their average gross monthly salary and the length of their working week. The average number of variable-cost staff (excluding the NHM who reported none) among the remaining workflows was 1.54, with a maximum of 4, indicating that it may be more feasible for many institutions to employ variable-cost staff than a team of full-time fixed-cost staff.

Using and treating labour as a variable cost implies that the cost of digitisation can more easily be pushed downwards, as these costs are only be incurred when digitisation is taking place – unlike where labour is treated as a fixed cost, meaning that the labour is being paid for even when no digitisation is taking place.

However, in practice labour is rarely fully ‘elastic’ and unless an institute can easily switch staff between digitisation and other tasks there are costs in redeployment, recruitment and training.

For the fix institutions with variable-cost staff, total annual fixed labour cost ranged from €18,727 – €123,264 PPS. One of RBGK’s workflows included national insurance payments and superannuation into their calculations and was removed from this analysis due to its incomparability to other workflows.

The cost for consumables per batch of 100 objects (single specimens or containers) ranged from zero to €54.49 PPS. The specific consumables used for each project were not named in every case, so it is not possible to identify precisely what the costs are or the reason for this wide range in consumables cost. The two reported cases of fungarium had a much higher cost for consumables than other specimens (Fig. 1). There is a significant variance in the median consumable cost between herbarium (€7.55 PPS, n = 8) and pinned insects (€0.04 PPS, n = 5). Herbarium has a much wider range than pinned insects.

Fig. 2 and Table 6 illustrates the proportions of fixed and variable costs that make up the overall cost of each workflow. The split between fixed and variable costs depends primarily on how staff is funded. While the costs for labour may be inflated in this analysis, labour can generally be considered to make up the largest percentage of overall cost for an institution’s digitisation efforts, whether funded through fixed or variable budgets.

Direct comparison of the reported rates of digitisation between institutions is not possible as each has different setups and team compositions, as illustrated in Table 7 where the different types of workflow and numbers of staff operating them are summarised. One assumption is that images are captured by all workflows, however the number and type of images is not known. Secondly, it cannot be assumed that label data capture occurs or is the same across all workflows. In some cases, it does. In others it doesn’t, and in yet other cases it happens at a higher level (i.e., folder, container) than individual specimens. Label data capture depends on institutional needs and can vary even from one project to another.

These differences in workflow and the level of capture can be seen in the throughput within specimen groups. After removing the single case of automated outsourcing due to its exponentially higher throughput, the remaining 22 workflows showed a wide range of throughputs where more than one case was reported, particularly for microscope slides and pinned insects (Fig. 3). Because of these wide ranges and a lack of data on certain collections types, it is difficult to assess an expected mean throughput.

Institutions also vary in the number of staff dedicated to digitisation, ranging from 0.1 to 4.8 people. As labour makes up the largest percentage of digitisation costs, it is important to understand labour’s impact on throughput. Contrary to expectations, a larger staff did not necessarily result in a linear increase in throughput. (Fig. 4).

Herbarium specimens showed a slight association between team size and throughput. However, the throughput of pinned insects varied widely on teams of one from 1,737 to 114,700 specimens annually, with the largest team of 3.8 returning the smallest throughput. While semi-automated processes did tend to show a higher throughput, the two cases of manual processes for pinned insects showed a throughput of 21,818 and 1,736. While the one case of an herbarium semi-automated workflow did yield one of the highest throughputs (52,800), the highest was a manual workflow (62,400).

These differences may be due to the depth of information collected in the digitisation process. While it is hard to make direct comparison with workflows, both LUOMUS and NHMUK have developed high throughput workflows for pinned insects (Wu et al. 2019, Dupont and Price 2019) which have annual throughputs that are more than double that of other workflows using a similar staff count. There is less variation among the herbarium sheet workflows when taking when considering staff count.

The time required to digitise a batch of 100 objects (single specimens or containers) is affected by multiple factors, including: layout of the institutions, storage facilities, equipment available, etc. There were 18 reported cases of time spent across all specimen types –NHM and RBINS did not provide any time data. The median hours spent digitising 100 objects was 9.88 and ranged from 2.10 to 217.67. RBGK’s microscope slides, the high outlier, are exponentially more time consuming than any other specimen type and was removed from further analysis.

The two palaeontological cases had a wide range, with one requiring 41.67 hours per 100 objects and the other double at 83.33 (Fig. 5). Pinned insects took an average 7.18 hours per 100 objects and herbarium required an average 9.39 hours. The much lower times per 100 items for these two collections is due to their relative homogeneity and more mature workflows.

Time was also estimated for each stage of the digitisation process – curation, image capture, image processing, data capture and preservation. In general, curation was the most time-consuming step in the process across most projects and specimen types (Table 8).

In order to assess the cost per item, an RBGK project that included national insurance and pension payments in their cost analysis and their case of microscope slide digitisation which had an exponentially higher cost per item than all other cases (€381.26 PPS) was excluded, as well as an UTARTU case that did not provide cost data. This left 19 cases.

The median cost per item across all cases was €2.10 PPS, ranging from €0.53 PPS to €34.22 PPS. Again, the range between the two cases of palaeontological digitisation proved to be the widest while pinned insects and herbarium were relatively consistent. The median cost per item for herbarium was €2.78 PPS and for pinned insects was €1.06 PPS (Fig. 7)

Figure 7.  

Cost per Item (€ PPS).

In the two cases where the digitisation process was fully automated – MNHN’s outsourced ReColNat workflow and UTARTU’s palaeontological collection – cost per item was reduced considerably (Fig. 8) and was accompanied by a near-equivalent increase in productivity (Table 9). Semi-automation saw even further reduced cost which may be a function of an increase in productivity without the additional costs of an outsourced service.

Figure 8.  

Cost per item by digitisation method.

While six out of seven institutions returned costbooks categorized by specimen type, RBINS return costbooks categorized by method of digitisation and size of the item being digitised. While this makes it difficult to compare with other institutions, it does provide insights into different aspects of digitisation costs by showing which methods of digitisation are more costly than others.

For example, 3D imaging is the most expensive digitisation method and with a very low throughput offset by the quality of the image captured. Transcribing metadata and 2D photo captures of insect boxes are the least expensive and have the highest throughput. Interestingly, µCT scanning has the highest annual total cost because of high fixed depreciation costs for X-ray equipment (€63,571 a year). However, the average cost per item remains relatively low because µCT achieves a throughput that offsets the increased costs. Fig. 9

Figure 9.  

Average cost per item (€ PPS) compared to total annual throughput per person (Items).

In conjunction with this costbook analysis, Walton et al. (2020) analysed the cost of transcribing label data in the digitisation process. Aside from RBINS, who reported an average cost per item for transcription of €4.85 PPS, no other institutions have broken out transcription costs as a separate line item. In the transcription analysis, case studies were collected that cover a range of collections types, transcription methods and data inputs which were analysed in lieu of cost data.

The minutes per item for transcription ranged significantly from ~30 seconds to up to 41 minutes to fully transcribe label data on a specimen. This large is due to the range of information that is included in the transcription process, the method used and the amount of quality assurance required. For example, georeferencing adds significantly to the time required for transcription, particularly if the label includes only vague location description. Some case studies reported that they did not include georeferencing because of limitations on either time or funding.

Some of the case studies provided examples of either outsourcing transcription to a service like Alembo, using a crowdsourcing platform like DigiVol or testing an automation tool like Google Vision. In each of these cases, staff resources were saved by not requiring museum labour resources for the actual transcription. However their were, in each case, time and money trade-offs for the increased need for project management, volunteer recruitment, quality checks and/or development resources needed to carry out the project.

The analysis showed that, consistent with the other digitisation components studied in this report, time and cost can vary significantly depending on collection type, staff resources and method deployed.

Anthropological, Palaeontological, Mineralogical and non-insect Invertebrates collections were not included in the scope of ICEDIG digitisation research. While non-insect invertebrates are a major collection type, they were erroneously omitted from the scope of Cocks et al. (2020) which was used as the basis of this report.

The costbook work in ICEDIG will be inherited and expanded upon by the DiSSCo Prepare project, specifically in Tasks 4.1 and 4.2, the “Costbook for DiSSCo” and “Cost model for charging services”, and their corresponding reports.

While not directly working on a costbook, SYNTHESYS+ will be gathering and assessing cost data as part of the new Virtual Access workpackage (Hardy et al. 2020,Smith et al. 2019).

In the subsections that follow, we offer some further considerations that other projects in the DiSSCo Programme portfolio should take into account but they apply to any organised large scale digitisation of collections.