Introduction

Agriculture is facing profound challenges in many parts of Europe. The need for technological change to address these challenges has often been highlighted (Cuadros-Casanova et al., 2023; MacPherson et al., 2022; Mizik, 2023; Pinke et al., 2022). For example, negative environmental impacts could be minimized through more precise management of arable fields, or Global Positioning System (GPS) guidance technologies could help farmers carry out cultivation measures and minimize environmental stress (Batte & Ehsani, 2006; Chivenge et al., 2021; Robertson et al., 2012; Tey & Brindal, 2022). But these benefits remain hypothetical in large parts of Europe due to the low uptake of precision farming technologies.

Overall, European farmers tend to be critical of digitization, especially information-intensive technologies. Lowenberg‐DeBoer and Erickson (2019b) show that only 8–14% of farms in Europe use variable rate fertilization. In smallholder regions such as Bavaria, where only about 5% of farmers use this technology, the adoption rate falls even further below European standards (Gabriel & Gandorfer, 2023). According to Gabriel and Gandorfer (2023) there are even cases of farmers returning to non-digital farming after adopting digitization, which raises the question of why these circumstances have not been explored yet. In contrast, farmers in regions with large agricultural structures (e.g. USA, Australia) have a clear advantage due to economies of scale (Robertson et al., 2012; Mizik, 2023; Say et al., 2018). The focus of this work is therefore placed on small-scale European structures.

With the help of qualitative research on technology adoption, some resistance factors have been uncovered. Factors such as the lack of user-friendliness of technologies, small farm sizes and associated economic challenges, and lack of support for implementation have been identified as important barriers (Busse et al., 2014; Eastwood et al., 2017; Kernecker et al., 2020; Pathak et al., 2019). However, research has focused mainly on the factors that influence the decision to adopt a technology; little has been done to examine the entire process of adoption with all its relevant economic and technical implications (Tey & Brindal, 2022).

There are notable exceptions. A case study by Batte and Arnholt (2003) provides insight into the problems of adopting precision agriculture technologies, with the steep learning curve being cited as a source of farmer frustration. Mizik (2023) adds to this by identifying the high cost of adoption as a major barrier; Adrian et al. (2005) also show that the steep learning curve and initial high investment work as barriers to deploying precision agriculture, while Kernecker et al. (2020) note that users face connectivity and equipment complexity barriers upon adoption.

In none of these studies are the pathways and associated economic impacts described in sufficient detail to explain the low adoption rates in smallholder regions. Pathak et al. (2019), Barnes et al. (2019b) and Knight et al. (2003) assert that risk-taking in the adoption of innovations is the critical factor. But are farmers in small-scale regions more risk averse than other farmers? Isn't it far more likely that potential risks and challenges for farmers in small-structured regions have been overlooked to date, and that farmers are therefore justifiably hesitant?

As research has too often overlooked the entire deployment process from investment decision through adoption to farm use, insight into farmers in small-scale regions is missing; in particular, the process of implementation has been neglected. Since theoretical models of technology adoption have underpinned most prior research, important practice-relevant questions remain unanswered (Chavas & Nauges, 2020; Pathak et al., 2019). Montes de Oca Munguia et al. (2021) emphasize that the pragmatics of adoption need to be measured to capture farm realities in practice.

Following this suggestion, this study was undertaken to investigate through an application of precision agriculture technology (i) whether there are differences in implementation effort due to different initial mechanization conditions, (ii) how entry and exit affect farm profitability, and (iii) how a farmer can find the right time to exit if technology use seems unprofitable. To do this, a conceptual framework is first established, through which an economic evaluation of smallholder technology use can be made based on empirically available numbers.

Conceptual framework

The implementation process is analyzed using a defined precision agriculture technology (variable rate fertilizer application, online approach with map overlay), a defined farm (average German arable farm), and defined scenarios (described below). To derive the economic impacts along the implementation process, decision paths were formed similar to the proposal of Triantafyllou et al., 2020 (see Fig. 1). The underlying data, e.g. on necessary steps, investments, and working time requirements, were taken mainly from KTBL information materials (In German: Kuratorium für Technik und Bauwesen in der Landwirtschaft e.V.; In English: the Association for Technology and Construction in Agriculture) (Drücker, 2016; Hufnagel et al., 2004; Hüter et al., 2005; KTBL e.V. 2022; Noack, 2007; Reckleben et al., 2007; Treiber-Niemann et al., 2013; Achilles et al., 2018).

Fig. 1
figure 1

Different pathways for entry into precision agriculture based on the current mechanization of the farm with choices of M1 (existing machinery obsolete and cannot be retrofitted), M2 (existing fleet can be retrofitted), M3 (existing fleet is state of the art), and M4 (engaging a service provider)

Full size image

Data were supplemented with the help of the author's practical experience and information available online (e.g., user reports, manufacturer information). Resulting estimates were then validated by experts, who concluded that the individual decision paths, with their associated workloads and investment costs reasonably represent the pragmatics of technology adoption by a small farm business and can serve as a basis for economic analysis.

Technology selection

Variable rate fertilizer application (online approach with map overlay) was selected as the technology to be studied because its complexity (information-intensive technology), high initial investment cost, and difficult to determine return on investment may cause farms to consider returning to non-digital farming (Fabiani et al., 2020; Gabriel & Gandorfer, 2023; Lowenberg-DeBoer & Erickson, 2019a; Miller et al., 2019). Variable rate fertilizer (VRF) technology enables application of different amounts and types of fertilizers (here, nitrogen) to different parts of a field, with the goal of increasing productivity of the cultivated land (Heege, 2013; Meyer-Aurich et al., 2010; Pedersen & Lind, 2017). VRF is a sophisticated technology (Miller et al., 2019; Ofori et al., 2020) that is highly developed, well-researched, and among the most widely used precision agriculture technologies by farmers in Europe (Gabriel & Gandorfer, 2023; Lowenberg-DeBoer & Erickson, 2019a; Nowak, 2021; Sishodia et al., 2020).

The VRF technology model includes four components: a tractor (1) with a control unit that can adjust the target amount of fertilizer according to the application map and the collected information from the attached nitrogen sensor (2), interfaced through an ISOBUS (standardized agricultural BUS system) or proprietary solution to a variable rate applicator (3) that can deliver fertilizer according to commands from the control unit. The fourth component is a Global Positioning System (4) that communicates spatial position information to the tractor’s control unit. The four components work together with a fertilizer application map created through software (Pedersen & Lind, 2017; Sawyer, 1994).

A centrifugal spreader was chosen as the applicator because it is one of the most commonly used technologies for fertilizer application (Pedersen & Lind, 2017; Lowenberg‐DeBoer and Erickson 2019b).

Farm model

As little precedent exists for study of technology adoption in small-structured areas (Gabriel & Gandorfer, 2023), an average German arable farm was selected as a representative farm model for European conditions. Other small-structured areas in Europe should face similar implementation challenges, with farmers confronting similar investment decisions.

As digitization entails whole-farm changes (Shockley et al., 2017), all calculations were performed on a whole-farm basis (McBratney et al., (2005) using agricultural statistics published by the Federal Statistical Office of Germany (Statistisches Bundesamt, 2017). This yields an average farm size of 89.85 ha, on which wheat, barley and rapeseed are grown at the average cropping ratios and yields listed in Table 1.

Table 1 Farm specifications and technology-induced changes in revenues and direct costs of an average arable farm in Germany (farm size 89.85 ha)
Full size table

The power of the tractor used for fertilization was adjusted to the farm size and assumed to be 83 kW. Likewise, the size of the attached centrifugal spreader was chosen to be 4000 l.

The machine data, prices and direct costs were taken from the KTBL database, which is regularly updated with the help of experts (Heinrichs et al., 2021) with planning data for (mainly) German farms. The database is considered a reliable source by scientists in agriculture research (Heinrichs et al., 2021; Reichardt et al., 2009; Shang et al., 2023).

As shown in Table 1, cereal crops dominate the model farm’s crop rotation. The prices represent a multi-year average and are intended as a benchmark. In order to be able to answer the research questions without the influence of fluctuating yields, product prices and consequent benefits, constant values were assumed over all years of the useful life, following Shockley et al. (2017).

To calculate additional benefits, values were taken from the literature on a crop-specific and technology-specific basis and averaged (see Tables 3, 4, and 5 in the Appendix) and then applied to the cost structure of the crops. The change in revenue results from a change in yield and a change in product price. The change in direct costs results from savings in seed, fertilizer, and pesticides.

The number of fertilization measures and the amount of fertilizer applied aligns with the expected yield. Values were taken from the recommendations of the KTBL (KTBL e.V., 2023).

Entry and exit scenarios

In order to simulate the differences that exist in practice with regard to the initial conditions on the farms, four variant entry scenarios were derived. Each entry scenario has a corresponding exit scenario that recreates as closely as possible the farm condition prior to entry into precision agriculture. The four variants are presented Table 2 and Fig. 1 and are described in detail in the following section.

Table 2 Combinations of the entry scenarios with the exit scenarios used, with the intention of restoring the initial state as closely as possible
Full size table

The first variant, M1, represents farms that use their existing technology far beyond its usual economic utilization potential (usually 8–10 years) and therefore face a higher barrier to entry due to their use of outdated technology. Variant M1 therefore assumes that the existing mechanization (tractor and applicator) cannot be retrofitted, so an investment in new technology would have to be made (Achilles et al., 2018; Barnes et al., 2019b; Hufnagel et al., 2004; Robertson et al., 2012). In M1, retrofitting either may not make sense (e.g. if obsolete technology would be replaced soon anyway) or may be precluded by technical limitations.

The second entry level variant, M2, represents farms whose existing mechanization can be retrofitted, as suggested e.g. by Robertson et al. (2009) or Lowenberg-DeBoer et al. (2021). M2 assumes that the necessary conversion of the existing machines can be carried out by the farmer himself.

The third entry level variant, M3, represents farms that have current technology (ISOBUS-capable, guidance system), but do not utilize its technical potential (Gabriel & Gandorfer, 2023). In this case, only minor adjustments (e.g. activation of specific ISOBUS functionalities) have to be made.

If fertilization is carried out with self-mechanization (M1–M3), it is assumed that the nitrogen sensor must be purchased.

A fourth variant, M4, represents contracting for site-specific fertilization as a service, rather than via self-mechanization. This variant was suggested e.g. by Kutter et al. (2011) as an option that places no requirements on the existing farm mechanization.

Application maps must be created by the farmer in all entry scenarios.

In the entry scenario M1, a complete exit involves selling all newly acquired machines and returning to non-digital cultivation after buying used machines similar to those used before. In the case of M2, only easily removable retrofit parts (including the nitrogen sensor) can be sold. In this case, the farmer must remove and sell the parts, taking into account the amount of work required to do so. In M3, the previous, only slightly modified mechanization can simply be retained (except nitrogen sensor) and the site-specific fertilization discontinued. Since the guidance system is not removed, the advantages of this technique can be maintained even after the variable application rate phase-out. In the case of M4, no changes were made to the existing mechanization anyway. The farmer will therefore simply not renew the contract and continue with the existing mechanization. In all exit scenarios, the investment in the software to create the application maps is considered a sunk cost, since a sale is not possible. Data in the supplementary information (SI) provide more details on the decision paths and their economic impacts.

Economic evaluation

Investing in digital technology entails both one-time and recurring costs. To evaluate these, the partial budgeting approach has been widely-used, considering only those changes directly related to the investment that deviate from the status quo (Lowenberg-DeBoer et al., 2020; Shockley et al., 2017). For M1, M2 and M4 the status quo is defined as no precision farming technology is in use yet. For M3, GPS guidance technology is already in use, i.e. savings from this technology are not additionally credited to the farm for site-specific fertilization.

As rendered by Eq. (1), the result \(R\left(n\right)\) for each year n is generated through cash flow analysis. Indexing each year allows the method of Shockley et al. (2017) to be used to identify cash-flow bottlenecks. To show financial impacts up to a point of exit, the result for each year \(R\left(n\right)\) is aggregated over the duration of use.

$$ R\left( n \right) = \mathop \sum \limits_{n = 1}^{t} B\left( n \right) - C\left( n \right) - IE\left( n \right) - EC\left( t \right) - RC\left( n \right) $$
(1)

In Eq. (1), the time of investment is 0, and the time of exit is t. The years of use are described by n, meaning n = 1 is after the first year of use. \(B\left(n\right)\) is the annual benefit, \(C\left(n\right)\) is the annual added cost of the technology, \(IE\left(0\right)\) is the Implementation Effort and represents the one-time cost in the year of investment, \(EC\left(t\right)\) is the Exit Cost and is the one-time cost in the year of exit, and \(RC(n)\) is the annual cost of hiring a fertilizing service contractor. For the entry scenarios M1–M3, \(RC\left(n\right)=0\). The years of potential economic utilization is set to 10 in these calculations and described by \(N\).

The additional benefit \(B\left(n\right)\) is calculated over the useful life according to Eq. (2):

$$ B\left( n \right) = \mathop \sum \limits_{n = 1}^{t} B_{GPS} \left( n \right) + B_{SSM} \left( n \right) $$
(2)

where \({B}_{GPS}(n)\) represents the value-added benefit of the GPS steering system (direct cost savings) and \({B}_{SSM}(n)\) represents the value-added benefit of site-specific fertilization (direct cost savings, increase in yield and increase in product price). A subdivision was made here because in the M3 exit scenario, the benefit of the steering system may be retained. The other three scenarios forfeit all benefits upon exit.

The additional annual costs are aggregated according to Eq. (3):

$$ C\left( n \right) = \mathop \sum \limits_{n = 1}^{t} VC_{T} \left( n \right) + VC_{A} \left( n \right) + VC_{SE} \left( n \right) + VC_{SO} \left( n \right) + D_{T} \left( n \right) + D_{A} \left( n \right) + D_{SE} \left( n \right) $$
(3)

The annual cost \(C\left(n\right)\) refers to self-mechanization only and consists of the added variable costs for the tractor \({VC}_{T}(n)\), the applicator \({VC}_{A}(n)\), the nitrogen sensor \({VC}_{SE}\left(n\right)\),and the software \({VC}_{SO}(n)\). The variable costs for the tractor include, for example, the costs for the transmission of the Real Time Kinematic (RTK) correction signal (mobile phone contract) and the working time required for additional work steps (e.g. transfer of the application map to the tractor terminal). It is assumed that the variable costs of the newly purchased, converted or available applicator (M1–M3) do not change significantly, so that \({VC}_{A}\left(n\right)\) can be neglected. The variable costs of the nitrogen sensor \({VC}_{SE}\left(n\right)\) include, among other costs, the effort required to calibrate the sensor before each fertilization measure. The amount of \({VC}_{SO}(n)\) is determined by the number of fertilization measures, the working time for creating the application maps, the annual license costs for the software, and the documentation effort. In M4, the variable machinery costs (of farm tractor and applicator) are credited to the farm, because by hiring the contractor, no machinery owned by the farm has to be used, and so the variable machinery costs for fertilization can be saved over the status quo.

The annual costs also include the depreciation of the newly acquired technologies or components. For depreciation, the method of double declining-balance (depreciation rate 20% per year) is used, since in reality a greater loss of value can be assumed in the first few years and the calculation model should be as realistic as possible. Thus, upon exit, the farmer loses the difference between the acquisition price in year 0 minus the selling price of the technology in year \(t\). \({D}_{T}(n)\) includes the loss in value of the tractor for M1, and for M2 for the retrofitted components. For M3, only the costs for activating functionalities are subject to depreciation. In addition to this, M1–M3 takes into account the depreciation of the nitrogen sensor \({D}_{SE}\left(n\right)\) purchased for fertilization. The annual interest costs for the machines are included in these three items so are not listed separately. M4 does not require any new hardware to be depreciated on part of the farm.

One-time expenses at time 0, the year of investment, are calculated by Eq. (4):

$$ IE\left( 0 \right) = IE_{T} \left( 0 \right) + IE_{A} \left( 0 \right) + IE_{SE} \left( 0 \right) + IE_{SO} \left( 0 \right) $$
(4)

The implementation effort of the tractor \({IE}_{T}\left(0\right)\) includes the labor cost for the steps shown in Fig. 1. This effort varies depending on the initial situation and the selected entry scenario. The same applies to the implementation effort for the applicator \({IE}_{A}\left(0\right)\) and nitrogen sensor \({IE}_{SE}\left(0\right)\). In the case of \({IE}_{SO}\left(0\right)\), the one-time investment in the software application, unlike an investment in machinery, also counts as an implementation cost and not as an annual cost \(C\left(n\right)\) subject to depreciation. As already mentioned in the previous section, the software implementation effort is considered a sunk cost as it cannot be recovered upon exit.

Equation (5) shows the calculation of exit costs in exit year t:

$$ EC\left( t \right) = EC_{T} \left( t \right) + EC_{A} \left( t \right) + EC_{SE} \left( t \right) $$
(5)

Here, \({EC}_{T}\left(t\right)\) stands for the costs incurred by the exit. In the case of M2, this can be the labor costs for removing saleable retrofit parts; in the case of M1, the purchase of a used tractor (comparable to the sold tractor of status quo) is included here. The same applies to the cost of the applicator \({EC}_{A}\left(t\right)\) and nitrogen sensor \({EC}_{SE}\left(t\right)\). There are no exit costs for the software, as its use is simply discontinued. A sale is not an option for the reasons mentioned above.

In addition to calculating economic changes in the course of implementation (Eqs. 1–5), the following section formulates decision rules that provide guidance in the event using technology proves unprofitable.

Decision rules for retaining the investment

The conditions put forward here rest on the assumption that a farmer decides to retain an investment according to the principle of benefit maximization, meaning as long as the additional benefits exceed the additional costs. This is the approach of Khanna (2001), Watcharaanantapong et al. (2014) and Kolady et al. (2021). Ideally, this would be the case from the beginning of use, but this is not always so on a farm, as one-time costs incurred at the beginning of use (e.g. \(IE\left(0\right)\)) cannot be fully recovered in the first year. But in any case, the principle should hold over the duration of technology use. If profitability is not possible because, say, investments have already been made, an attempt should be made to keep financial loss to a minimum.

In formulating the decision rules, a distinction must be made between linear and nonlinear aggregated result functions. This distinction determines the form of the resulting decision rules, as explained below.

Decision rules for linear result functions (M4)

Since \(B\left(n\right)\) and \(RC(n)\) are relatively constant variables from year to year, and \(C\left(n\right)\) at low investment levels consists mainly of variable costs \(VC(n)\), whose level is also relatively constant year to year, a linear relationship can be observed in the result functions \(R\left(n\right)\) for the M4 scenario. In this case, the following conditions (6)-(8) apply for maintaining site-specific fertilization:

$$ R^{\prime}\left( n \right) < 0 \;{\text{and}}\;R\left( N \right) \ge 0 $$
(6)

Condition (6) checks whether the investment is profitable over the entire economic utilization potential N. If the result function has a negative slope, meaning costs exceed benefits, but the function does not intersect the x-axis until the economic utilization potential is reached or later, the investment can be retained.

$$ R^{\prime}\left( n \right) = 0\;{\text{and}}\;R\left( 0 \right) \ge 0 $$
(7)

If the benefits of precision agriculture are exactly offset by the costs, the slope of benefits vs. costs remains flat and condition (7) applies. In this case, a positive result must have already been established at the beginning of the investment.

The case of condition (8), however, is more likely and most desirable. The benefits increase over the costs during the entire utilization period:

$$ R^{\prime}\left( n \right) > 0\;and\;R\left( N \right) \ge 0 $$
(8)

The rule here is simple: if the result function has a positive slope and a positive result is achieved at or before the economic utilization potential is reached at N, the investment can be retained.

Decision rules for parabolic result function (M1–M3)

In the scenarios M1–M3, the high investment requirement (degressive trend for \(D(n)\)) leads to the result function \(R\left(n\right)\) behaving in the nonlinear manner of a second-degree polynomial. In this case, the following condition (9) applies to the retention of the technology:

$$ \left( {R^{\prime}\left( n \right) = 0} \right) \le \frac{N}{2}\;{\text{and}}\;R\left( N \right) \ge 0 $$
(9)

If the local minimum of the result function is at or less than half of the economic utilization potential N and the aggregated result reaches at least a value of zero at the end of the useful life, the investment should be maintained. If condition (9) cannot be fulfilled, but condition (10) is fulfilled, it is also advisable to continue site-specific fertilization:

$$ R\left( N \right) \ge R\left( t \right) $$
(10)

Condition (10) says the aggregate result by the end of the economic utilization potential is greater than the aggregate result in the chosen exit year \(t\) (\(t<N\)) it makes most sense from an economic point of view to keep the technology. If an investment turns out not to be profitable, the farmer would either have decided not to invest before time 0, or he would pursue (after entry) the strategy from condition (10). This does not exclude the possibility that a time before t might have made more sense.

Note that in exit year t, the amount of one-time exit costs \(EC\left(t\right)\) and, if applicable, the update of \(C\left(n\right)\) after exit must be considered in all scenarios.

Consideration of uncertainty in the phase-out of site-specific management

If negative economic changes occur in the first few years after the start of site-specific variable rate fertilization, it has to be decided whether withdrawing from the investment is the economically better alternative.

While it would be possible to perform a detailed analysis in advance of an investment (ex-ante) using multi-year averages and to decide against an investment in time, i.e., before purchase and implementation, as Swinton and Lowenberg-DeBoer (1998) show, it is not possible after implementation to reliably determine the profitability for future years (after entry) in light of volatile markets and changing growing conditions (Munz & Schuele, 2022). However, the future trend of the aggregated result function \(R\left(n\right)\) can be predicted with the help of regression lines based on the data collected from previous years, as shown for example by Al-Gaadi et al. (2016). The shape of the regression functions is fit based on the trend of the previous data points (after the third year, the polynomial form is chosen for M1–M3 and the linear form for M4). Using these regression lines, an advisor can judge whether it is advisable to keep the technology by applying the decision rules (6)–(10). The complexity of economic forecasting points to the urgency of close cooperation between farmers and consultants, as Barnes et al., 2019a also recommends.

It should be noted that a linear regression line is also indicated for M1–M3 at the end of the second year, since there is no other estimate when only two data points are available. Moreover, in all cases (M1–M4), at least two years are needed to determine a regression line, so even if it would make economic sense to exit after the first year, such a decision could not be made by a regression estimator.

Results

Operational changes: cash flow analysis

Changes by year of operation in the arguments of the result function R(n) given by Eq. 1 are depicted in Fig. 2 through Fig. 5 for all entry scenarios with values given per hectare. For the entry scenarios M1–M4, \(t\) was set to 1–9 under the assumption that occurring costs and benefits are known for all years. An exit at the end of the tenth year was not calculated since the full economic utilization potential period of 10 years had already been reached. All values were calculated based on the average farm defined in Table 1, assuming the decision to start site-specific fertilization had already been made under each entry scenario, regardless of whether it made economic sense to do so. To provide a better overview, all values shown in Figs. 2, 3, 4, and 5 are presented in the Appendix (Tables 7, 8, 9, and 10). Individual values from these tables are given in the following section. More detailed information can be found in the supplementary materials.

Fig. 2
figure 2

Development of the variables of Eq. 1 (Additional Annual Benefit B(n), Implementation Effort IE(0), Exit Costs EC(t), Additional Annual Costs C(n) and Annual Contractor Rate RC(n)) over time (top) and aggregated result function R(n) (bottom) for the entry scenario M1 (Existing fleet is outdated, no retrofitting) and exit years 1–9

Full size image
Fig. 3
figure 3

Development of the variables of Eq. 1 over time (top) and aggregated result function R(n) (bottom) for the entry scenario M2 (Existing fleet can be retrofitted) and exit years 1–9

Full size image
Fig. 4
figure 4

Development of the variables of Eq. 1 over time (top) and aggregated result function R(n) (bottom) for the entry scenario M3 (Existing fleet is state of the art) and exit years 1–9

Full size image
Fig. 5
figure 5

Development of the variables of Eq. 1 over time (top) and aggregated result function R(n) (bottom) for the entry scenario M4 (Engaging a service provider) and exit years 1–9

Full size image

To illustrate the effects of time on the individual variables, the annual results in the upper part of Figs. 2, 3, 4, and 5 are presented individually and not aggregated over the useful life. The lower part of the figures then shows the aggregation of the individual variables according to Eq. 1. Each exit year was marked by a different color with a different symbol, starting with a light shade (yellow) for the first year of exit, to a dark shade (dark purple) when not exiting.

In the entry scenario M1 (Fig. 2), an additional benefit \(B\left(n\right)\) per hectare and year of 70.11 € is achieved. In the first year of use, the sale of the old, previously used machines (tractor and applicator) results in a negative value for \(IE\left(0\right)\) of -114.32 € ha−1 despite additional costs e.g. for familiarization. \(IE\left(0\right)\) always occurs at the beginning of use, regardless of the time of exit. Since in the case of an exit the original state is to be almost restored, i.e. previous non-digital machines must be purchased again, costs \(EC\left(t\right)\) amounting to 150.34 € ha−1 are incurred for the exit in year \(t\). Like the implementation costs, the exit costs are incurred only once, but depend on the year of the exit. Additional annual costs \(C\left(n\right)\) are mainly influenced by tractor, applicator and sensor depreciation and therefore takes a degressive course overall. \(C\left(n\right)\) can be neglected from exit onward, since all purchased mechanization can be resold. \(RC(n)\) is not considered for M1–M3 because no contractor is hired.

Looking at the aggregated result, the farm with M1 characteristics should not have entered at all based on the outcome (lower part of Fig. 2 and Table 7). However, should the decision have already been made, as assumed in this analysis, it is important to exit as soon as possible, preferably in the first year. Farms that do not exit by the fourth year should maintain the investment until the end of the useful life, since exit would firstly incur costs \(EC\left(t\right)\) and secondly, in this case, the benefits would be completely lost, which would result in the overall outcome function ending at a lower level when \(N\) is reached than would be the case if exit were not made.

Farms with a retrofittable fleet (M2, Fig. 3) achieve the same annual benefit as in M1 and M4 of 70.11 € ha−1 until exit. Since in this case the existing machines continue to be used after retrofitting, the initial setup costs \(IE\left(0\right)\) do not include any sales revenue. The effort for e.g. conversion, familiarization with the functionalities and one-time investments cause costs per hectare of 43.92 € in this case. At exit in year \(t\), costs \(EC(t)\) amounting to 8.85 € ha−1 are incurred due to the labor required to remove the saleable retrofit parts and nitrogen sensor. The degressive pattern of \(C(n)\) is caused by the depreciation of the retrofitted components and nitrogen sensor and starts at 133.82 € ha−1 in the first year and decreases to 23.19 € ha−1 in the tenth year if no exit occurs. Since not all retrofitted components can be removed and sold at a reasonable cost in the event of an exit, the loss in value of these components is extrapolated to leave €9.21 ha−1 as depreciation at \(C(n)\) in the first year if an exit occurs. This value is reduced by decreasing depreciation to 1.54 € ha−1 in the tenth year.

For M2, decision rule (9) is satisfied, so it is advisable for farms with these characteristics to hold on to the investment until the end of its useful life (lower part of Fig. 3 and Table 8). Since the break-even point is not reached until the last year of use, it can be a challenge to remain patient. However, if the farmer exits between the first and seventh year, this is associated with a loss of 115.13–195.65 € ha−1. Farmers in scenario M2 are thus in a lock-in situation, since an exit without major losses is only possible in the last year. This is then obsolete anyway, since a clearly positive trend can be seen in the aggregate result function.

For farms with state of the art equipment (M3, Fig. 4) the annual additional benefit amounts to 47.06 € ha−1 and not to 70.11 € ha−1 as in the other scenarios. This is due to the assumption made in the scenario that a GPS guidance system is already available and used and therefore the savings potential of the guidance system is not taken into account when comparing the digitized variant to the status quo. Since M3 only requires ISOBUS activation for site-specific fertilization, purchase and familiarization with the creation of application maps and nitrogen sensor, the expenditure at the beginning of use \(IE\left(0\right)\) amounts to 33.63 € ha−1. When phasing out site-specific fertilization, only the creation of application maps is then discontinued and the nitrogen sensor is sold, resulting in low exit costs \(EC\left(t\right)\) of 0.96 € ha−1. There are annual variable costs associated with the production of application maps and calibration of the nitrogen sensor. Combined with the depreciation of the investments (activations and nitrogen sensor), a slightly degressive trend of \(C\left(n\right)\) can be seen. After an exit, depreciations of the activations simply continue, since they cannot be sold. For this reason, the aggregate result function after exit is not parallel to the x-axis, but has a slightly degressive course that follows the loss in value of the investments in activations made. Since for M3 the additional benefit is not sufficient to cover the additional costs incurred from the beginning, an exit until the end of the fifth year will have a negative impact on the final result (loss of 8.47 € ha−1). Only from the sixth year onwards is it possible to exit without loss. However, since decision rule (9) is fulfilled from the beginning and a positive result can already be achieved shortly after the sixth year, the farm in scenario M3 should stick to site-specific fertilization over the entire useful life. In this way, a better result of 151.33 € ha−1 can be expected over the entire economic utilization potential (in case of no exit) compared to the status quo.

For farms that hire a service provider (M4, Fig. 5), the annual benefit is €70.11 per hectare. In this scenario, no changes have to be made to the own mechanization. Therefore, the implementation costs \(IE(0)\) are limited to the purchase and training of the software to create the application maps and amount to 22.25 € ha−1. No exit costs \(EC\left(t\right)\) are considered in M4, since no conversion measures of the own mechanization are necessary for the exit. Annual costs \(C(n)\) are only incurred for the preparation of the application maps in the amount of 4.20 € ha−1. Since the number of fertilization measures and thus the expenditure for the preparation of the application maps remain the same over the years, a clear linear trend can be seen here. The same applies to the contractor costs, which remain constant at 44.46 € ha−1 per year and cease to apply with immediate effect in the event of exit.

When summing up the annual result according to Eq. (1), it is noticeable that the result function has a linear course due to the variables \(B\left(n\right)\), \(C\left(n\right)\) and \(RC(n)\) remaining constant. Looking at the aggregated result function, it is noticeable that the resulting additional benefit in the first year almost compensate for the additional costs incurred, causing only a small loss of € 0.81 per hectare. A loss-free exit would be possible in the following years, but is not advisable due to the clearly positive development of the aggregated result function. If site-specific fertilization is maintained until the end of the useful life, an advantage of 192.17 € ha−1 can be achieved compared to the status quo. Therefore, it is advised to maintain the investment.

Comparison of the different scenarios

As shown in Figs. 2, 3, 4, and 5, there are distinct differences in the implementation effort \(IE\left(0\right)\) between the entry scenarios M1-M4. It is interesting to note that in M1, \(IE\left(0\right)\) has a negative value despite the great effort made to restructure the machinery. This is due to the sale of the old tractor and applicator and hides the fact that the implementation of the technology in M1 involves a significant effort in terms of working time and financial resources. Comparing M2, M3 and M4, it is noticeable that the cost of implementation are almost halved from 43.92 € ha−1 in M2 to 22.25 € ha−1 in M4, since in M4 no changes have to be made to the farm's own machinery by hiring a contractor.

In no entry scenario do the benefits achieved in the first year cover the entry costs. This leads to the conclusion that an assessment of economic viability should never be made after the first year. In the first year, the additional costs either exceed the benefits of the first year (M1–M3), or completely consume the additional benefits (M4). Further, each scenario incurs the highest cost outlay \(C\left(n\right)\) in the first year of operation. This leads, at the end of the first year, to a loss of 373.46 € ha−1 for M1, 116.49 € ha−1 for M2, 46.10 € ha−1 for M3 and 0.81 € ha−1 for M4. The financial burden due to the introduction of site-specific fertilization is so high for M1 that, despite the rapid fall in costs\(C\left(n\right)\), the farm cannot turn a profit over the entire period of potential economic utilization\(N\). Costs in M1 only fall below the economic benefits accrued in the last year of use. In M2, the annual costs fall below the annual benefit shortly after the fourth year. This means that the additional costs incurred up to that point can still be offset until the end of the useful life. However, it should be noted that only a few farms have the liquidity to sustain multiple years of unprofitable operations. This can lead to farms in M2 no longer using the technology despite a favorable starting position, especially if the exit costs \(EC\left(t\right)\) are not taken into account.

Farms in scenario M3 and M4 enjoy (almost) cost-free exit conditions (\(EC\left(t\right)=0\)). In M3, non-removable components (ISOBUS activations) continue to be depreciated, but these do not have a decisive negative influence on the result \(R\left(n\right)\) due to the low amount, so that an exit is possible without much of a financial hit. In contrast, an exit from M1 and M2 is significantly more costly. In the year of exit, M1 incurs costs of 150.34 € ha−1 for the purchase of conventional technology comparable to the status quo, with the assumption that these costs will remain fixed over time. In M2, the farmer must first remove saleable components and then incur conversion costs of 8.85 € ha−1 and is then subject to additional depreciation of the non-saleable components. These results make clear that farms with obsolete technology (M1/M2) not only face high costs when converting to digitization, but also face major costs when exiting. The extent to which this circumstance influences the timing of the exit is described below.

In summary, farms that can enter site-specific fertilization (M3/M4) with little effort (especially in terms of changes in their own mechanization) are clearly favored. A similar pattern emerges for exit. If exit can be done with little effort and ideally without updating \(C\left(n\right)\), as is the case for M4, the investment is less risky. Furthermore, it was shown that farms with start condition M1 should never have entered site-specific fertilization (by means of self-mechanization). Farms with the starting condition M2 can reach the break-even point at the end of the useful life (see Fig. 3), but run the risk of abandoning the technology due to the long period until profitable use or not reaching the specified 10 years due to the still limited experience regarding the length of the useful life.

However, if a farmer wants to exit (especially in M1/M2), there is only a narrow time window for an optimal exit, which should not be missed under any circumstances. Whether this point in time can be determined by the farmer is analyzed next.

Exit decision: what can a farmer know vs. what does a farmer need to know?

Figure 6 compares a static financial analysis, including the aggregated result function \(R\left(n\right)\) after exit in different years, to a dynamic forecast of the result function for the entry scenarios M1–M3. The static analysis assumes all future costs and benefits are already known, including exit costs \(EC\left(t\right)\)), while the dynamic forecast is used to model whether in reality a farmer can make a decision to exit or remain in site-specific fertilization at the optimal time. Omitted from the presentations is M4, contracting fertilization as a service, because linearity of its result function makes the static and dynamic analyses converge.

Fig. 6
figure 6

Comparison of R(n) of the static analysis (left, including the aggregated result function after exit in different years) with the dynamic forecasting of the result function R(n) (right, aggregated result function after exit is not shown here) for the entry scenarios M1–M3. The first linear regression line of M1 and M2 is truncated (detailed results can be found in Table 8)

Full size image

The future development of the aggregated result function \(R\left(n\right)\) is predicted using regression lines based on the data collected from previous years (right-hand side of Fig. 6). The previously developed decision rules are then applied to these predictions and the result is then compared with the actual result function (left side of Fig. 6). All values are given for the entire farm.

The presentation of the further course after the exit (see left side "Static Analysis") was omitted for the right side (“Dynamic Forecasting”) for better clarity. The further course of the result function after the exit would then correspond to the curve of the respective year on the left-hand side.

For the entry scenario M1, based on the static analysis and applying conditions (9)-(10), the recommendation is to exit immediately after the end of year one. For the dynamic estimation, however, the requirement for another data point makes it necessary to wait at least until the end of year two. In this linear regression line, condition (6) recommends immediate exit. In this case, if a farm follows this advice, the business would exit site-specific fertilization with a loss of €56,658.67. If a decision were put off until the end of year three, it would be guided by the nonlinear regression line that had emerged. On the basis of condition (10), continuing until the end of the economic utilization potential would be the best alternative. Following through on this alternative, the farm will then have lost € 93,173.41 at the end of the period of potential economic utilization \(N\) compared to not starting site-specific fertilization. This means that with the forecasts and decision rules, an M1 farmer can never make an optimal decision, unless it is made immediately after the second year.

In the entry scenario M2, the aggregate result function after exit declines slightly due to the non-saleable retrofit parts. Together with the overall trend of no exit (peaking around the fourth year), sticking with site-specific fertilization over the entire useful life is recommended. Even exiting after the first year of use would result in a lower final result than exiting at year 9 or remaining with the technology until the tenth year (satisfying condition (10)).

Comparing this with the dynamic prediction, a noticeably strong linear decreasing trend is predicted after the second year (Fcst. year 3). Since condition (6) is not fulfilled and no improvement can be expected due to linearity, the farmer would be advised to quit immediately after the end of the second year. Only at the end of the third year, with the possibility of non-linear estimation, does the forecast advise the farmer to continue with site-specific fertilization. With Fcst. Year 4 to Fcst. Year 9, condition (9) can be fulfilled. Although this forecast proves to be too positive in the further course (years 4–9), it does not change the fact that condition (9) is fulfilled and it is therefore advisable to retain the technology for all further years. In conclusion, it can be said that based on the available knowledge (dynamic forecasting), a farmer would be led to a wrong decision to exit at the end of the second year. Later years with non-linear estimation lead to the same decision resulting from static analysis.

Since lower investment costs are incurred for M3 compared to M1 and M2 and the resulting annual depreciation only accounts for a small proportion of the result function, an almost linear relationship can be observed. Overall, condition (9) is fulfilled, so that the continuation of site-specific fertilization over the entire useful life is recommended. In any case, a farmer must stick with the investment at least until shortly after the fifth year in order to be able to cover the costs incurred up to this point. Also, when estimating the result function from the end of the third year (Fcst. Year 4) to the end of the ninth year (Fcst. Year 9), condition (9) is always fulfilled and thus the result of the static observation is confirmed. Only at the end of the second year (Fcst. Year 3), when a slight linear declining trend is predicted, would the farmer be advised to exit the investment if condition (6) is applied. It can therefore be concluded that in the case of M3, the right decision to remain in the investment can only be made from the end of the third year.

Reliable forecasts can be made from the end of the second year onwards with clearly linear result functions, as is the case with the entry scenario M4 (not shown in Fig. 6). Here, there is no deviation between the estimated result function and the actual result function (see Tables 6, 11). That a linear regression requires at least two years of data, however, means a farmer has to wait at least this long in each scenario before making an economic evaluation.

In summary, nonlinear outcome functions lead to considerable uncertainties regarding the correct exit timing and, in the case of M1 and M2, wrong decisions can be made even with the help of the estimation curves and the decision guidelines from section 0. In reality, a farmer cannot predict the financially optimal time to exit. This can lead to increased losses if the exit is immediate (see M2) or if the farmer waits, misses the right exit time and thus has to accept greater losses than necessary (see M1). The main focus should therefore be on the misguided decision to enter. In both M1 and M2, a better alternative would have been to keep the current cultivation method of homogeneous fertilization, or to take advantage of the profitable outlook offered by M4 and contract for the service.

Discussion

Despite research on the diffusion and functionality of precision agriculture technologies, especially site-specific fertilization, there has been a lack of an analysis into the adoption process that would consider different conditions of mechanization existing on farms, and potential ways farms could exit unprofitable precision agriculture investments. The present study was undertaken to provide such analysis.

Four digitization scenarios were considered. In M1, the farm has outdated equipment and would have to start precision agriculture from scratch; in M2, existing equipment can be retrofitted with precision technology. In M3, the existing equipment is state-of-the-art and needs only be set up, while in M4, the farm contracts for site-specific fertilizer service. Each scenario represents conditions on a real-world smallholder European farm.

To highlight challenges and barriers these farm businesses face in introducing digitization, a cost vs. benefit analysis was done that considers annual economic benefits, one-time implementation costs, ongoing operating costs, and real-world economic constraints for each scenario. The assumption was made that the farm in each scenario had already made the decision to adopt variable rate fertilizer technology, although the analysis herein reveals this to be an unprofitable decision for farms in M1.

Analysis further indicates that there are significant differences between scenarios M1–M4 in terms of the financial outlay for the introduction of digital technologies, leading to the conclusion that farms with outdated equipment in small-scale agricultural regions are not yet ready to implement site-specific fertilization due to the given cost structures. While farms with retrofittable equipment (M2) may reach profitable use near the end of their useful life, it is likely that these farms could not accept such a long period of unprofitable use, especially if the assumed benefits cannot be achieved or the useful life is shorter than assumed in this study. Since outsourcing this measure (M4) was found to be the best option in this study, these disadvantaged farms could move in this direction.

Furthermore, it was shown that the clear economic boundaries that exist in theory have limited applicability in practice. Applying these boundaries to future predictions of the overall outcome led to incorrect decisions in the cases of M1–M3 with a non-linear outcome function. Only in the case of linear trends, as observed in M4, did predictions and the application of economic boundaries lead to reasonable advice for sticking with the technology despite a negative outcome in the first year of use.

Calculations were based on assumptions from agriculture statistics; these were used to work out clear trends that could be separated from market volatilities. But the author is aware of dynamics in the pricing of agricultural commodities, and these can lead values to deviate from the assumptions made, meaning:

  1. (1)

    The yields and product prices of the individual crops, which are constant over the useful life in this analysis, could be replaced by realistic time series – the approach of Diebold et al. (2006) – or by using historical time series data from an agricultural statistics bureau. This would likely affect the dynamic analysis in the determination of the optimal exit time.

  2. (2)

    Use of a single farm model precludes a mapping of economies of scale. For example, it could be assumed that the economies of scale mentioned by Mizik (2023) would have a negative impact on very small operations, which would face an insurmountable hurdle in combination with the entry scenarios M1 and M2.

  3. (3)

    Exit scenarios could be adjusted in M1 and M2 according to the results of Gabriel and Gandorfer (2023), i.e. an exit scenario could be formed, in which the purchased mechanization is maintained and only the site-specific fertilization is stopped.

  4. (4)

    Since the focus of this study was on the economic aspects of the implementation process and rigorous economic optimization functions were applied to determine exit/no-exit decisions, socioeconomic factors were not considered. Non-economic variables may have a promoting or inhibiting influence on the implementation process (e.g. Kernecker et al., 2020; Pathak et al., 2019; Weersink & Fulton, 2020). It would be interesting to find out if the economic loss or gain shown in this study is included in farmers' ex ante decision on adoption (perceived benefit) or if it cannot be estimated due to the high complexity and number of steps.

  5. (5)

    In individual years (especially the entry year and exit year), liquidity shortfalls could occur. This was not investigated in detail in scenarios M1–M3, but may be the subject of further studies as suggested by Shockley et al. (2017).

  6. (6)

    Since the determination of working time requirements for familiarization and regularly recurring tasks is associated with uncertainties, is linked to a large number of factors, and lacks empirical foundations, reasoned assumptions have been made. These assumptions are based on KTBL literature, other practice-oriented sources, the author's experience, and expert assessments. Even though practice-oriented literature such as the KTBL materials are considered trustworthy by farmers and experts (Reichardt et al., 2009), a quantitative study with a representative sample would be necessary to determine more accurate values. Since depreciation effects, as relatively accurately predictable quantities, have a much more pronounced effect on the course on the results function \(R\left(n\right)\) due to their magnitude than is the case with the assumed labor time expenditures, the assumptions made can be used without any problems until more accurate values are available.

In addition, it is conceivable that the time required in initial effort, exit conditions and operating conditions varies with the ability of the farm manager, so that, for example, farmers who are not familiar with digitization have to invest considerably more time than farmers with the corresponding knowledge (a conjecture by Ofori et al., 2020 and Marra et al., 2003).

The entry scenario M1 was derived from both the author's experience and practice-oriented literature. It would be interesting to conduct a quantitative study to find out what share of farmers would assign themselves to this category. The results calculated here could well match a farmer’s own business sense, which would explain the low adoption rates in small-scale agricultural areas that Lowenberg‐DeBoer and Erickson (2019b) and Gabriel and Gandorfer (2023) highlighted. As the results from this study clearly show, the M1 entry scenario is not only inferior to all others, but offers a farmer no option for profitability. Until the farm upgrades its equipment, digitization for this group makes no sense. Depending on the size of this group, measures must be derived to support these farmers if the goal is digitizing their farms.

A further quantitative study of interest could determine the rate at which farmers leave precision agriculture. In combination with a qualitative research approach, similar to the case study conducted by Batte and Arnholt (2003), motivations that lead farmers to abandon this farming practice could then be analyzed. If support from digitized systems is to be relied upon in Europe of the future for more sustainable management of agricultural land, as proposed for example by MacPherson et al. (2022), these open questions urgently need to be clarified and suitable measures elaborated.

Dynamic analysis: prediction of the future course of the result function

To predict the future course of the aggregated result function, it was assumed that a farmer has the data used (in particular Tables 6, 11) and on this basis is able to form (independently) a regression line and to apply decision rules (6)–(10) to it. This does not always apply, especially in the case of increased complexity with a non-linear course of the result function. In addition, when a decision is made to exit, \(EC\left(t\right)\) must be added to the actual result function and this adjusted result function must then be updated and compared to the final results \(R\left(N\right)\) at the end of useful life. In addition to the solutions discussed by Busse et al. (2014) and Mizik (2023), it is therefore advisable not only to seek professional help during implementation and in the early stages, but also during its use to have the (forward-looking) economic viability of the system assessed by a consultant or expert.

Conclusions

Researchers have puzzled over the slow adoption of precision agriculture by small-scale farmers. By recording the decision paths such farmers have to make along the implementation process of deploying for the first time site-specific fertilization, it was possible to show, using a prototypical small farm model, that farms with outdated, non-retrofittable technology are effectively excluded from digitization, as it will never be profitable over the lifetime of the precision technology. This results primarily from the high financial outlay for new equipment, and far less from the working time required for implementation.

Furthermore, it became evident that even when retrofittable mechanization is available, the investment is at the limit of a profitable use at the current costs and the model farm size of 90 ha.

If a farm already uses digital technology that can be used for precision agricultural management at modest cost, or if a farm contracts out the site-specific fertilization program, then a positive linear effect in the operational result can be observed, delivering from 151.33 to 192.17 € ha−1 of accumulated added revenue at the end of useful technology life.

Most importantly, it was shown that under uncertainty, inefficiency of site-specific fertilization, and non-linear behavior in the aggregated overall return of an investment, it is difficult for any farm to find the right time to exit from unprofitable use of the technology. Decisions on the termination of unprofitable technology use (after introduction) cannot be made according to rational criteria in the first two years. However, an immediate decision would be necessary, e.g. in M1, which puts farmers in a predicament. They are faced with the dilemma of either exiting too late (on the basis of rational considerations) or exiting too hastily and relying on their intuition. This means the initial decision—before investing—for or against the technology has the greatest impact on operational results.

It has been speculated that the overall low adoption rates of precision agriculture technology by small-scale farms can be explained by the increased complexity of entry and use compared to existing outdated mechanization. Yet this analysis herein shows a more fundamental reason. Farmers using traditional equipment may be reluctant because they recognize the unlikelihood of recovering the costs of digitization. In any case, the scenario of acquisition and non-adoption described in the literature can now be understood from an economic point of view.

The present work should be seen as a starting point for a more detailed economic investigation of the precision agriculture implementation process that might clarify questions relevant to practice. The results herein underscore the importance of professional support for any proposed implementation, especially during early feasibility studies. Broader questions to take away from this study suggest further quantitative work. What is the share of farms in small-structured areas where precision agriculture does not make business sense for the farmer? Should farms with obsolete technology be supported financially? Answers to these questions will help determine the future of precision agriculture in small scale farms.