Automation in public sector jobs and services: a framework to analyze public digital transformation ’ s impact in a data-constrained environment

Purpose – This study aims to estimate the value of the impact from digital transformation (DX) focusing on its automation effect, looking at the time and cost savings coming from the substitution effect with an adoption of digital technologies. For example, cloud and arti ﬁ cial intelligence technologies such as ChatGPT have the potential to change ways of working, substituting and replacing several of the tasks that are currently carried out by public administration (PA) employees and labor processes underpinning PA services. Design/methodology/approach – The paper outlines a new framework to estimate the potential impact of DX on the public sector. The authors apply this framework to estimate the value of the impact of DX on the Italian PA, de ﬁ ning the latter by the collection of the value of its labor (i.e. PA workforce salaries) and by the collectionof thevalue ofitsoutputs (i.e.publicservices ’ costs). Findings – This study ultimately maps out the magnitude and trends of how likely the PA occupations and services could be substituted in a wider process of DX. To do this, the authors apply their framework to the Italian PA, and they triangulate secondary data collection, from of ﬁ cial accounts of the Italian Ministry of Economics and the National Statistical Institute, with methodological antecedents from the UK Of ﬁ ce for National Statistics and experts ’ insights. Results provide a snapshot on the type and magnitude of PA jobs and services projected to be affected by automation over the next 10 years. Originality/value – To the best of the authors ’ knowledge, this paper provides for the ﬁ rst time an approach to estimate the value of the impact of DX on the public sectorin a data-constrainedenvironment – or in the lack of the required primary data. Once applied to the Italian PA, this approach provides a granular map of the automatability of each of the PA occupations and of the PA services. Finally, this paper mentions preliminary insights on potential challenges related to equity in public sector jobs and implications on recruitmentprocesses.


Introduction and background
The issue of efficiency, effectiveness and productivity is not a new topic in public digital transformation (DX) research.The adoption of DX carries great promises to improve government efficiency in terms of lower costs or increased revenues and effectiveness, to improve citizen service delivery, to create new types of services that were not previously available, to increase the accessibility of those services and to transform government itself through automation.The search for efficiency has been one of the most important drivers of ICT use in government, and most national strategies specifically address this goal.The pursuit of efficiency sees as its objective not only the reduction of overall spending but also the allocation of funds to higher priority areas (Lee and Perry, 2002;Brown, 2001;Moon and Norris, 2005;OECD, 2003;Luna-Reyes et al., 2012).
However, although many DX programs have been started under this effigy, and measurement becomes fundamental in a model of continuous improvement, today little is still known about the impact in terms of efficiency of DX initiatives, partially because of a lack of effective measures to evaluate these aspects (Carbo and Williams, 2004;Kunstelj and Vintar, 2004;Esteves and Joseph, 2008).Intelligent automation and artificial intelligence (AI) technologies such as ChatGPT are an example of DX adoption that policymakers are addressing across the public administration (PA), despite a significant lack of evidence about their expected impact on resources and processes (Campion et al., 2022;Kuziemski and Misuraca, 2020).
There is a multitude of definitions for DX: it encompasses both process digitization with a focus on efficiency, and digital innovation with a focus on enhancing existing physical products with digital capabilities (Berghaus and Back, 2016) as "a process that aims to improve an entity by triggering significant changes to its properties through combinations of information, computing, communication, and connectivity technologies" (Vial, 2019).
Significantly, technological advances are leading to the development of a spectrum of digital workforce tools that organizations can use to automate their processes.At one end of the digital workforce spectrum is basic automation, which uses technology to manipulate existing software to complete a process.At the other end of the spectrum is AI, which is software able to learn by analyzing data and then refining future performance (Cooper et al., 2019).
According to Kaplan and Haenlein (2020), AI can be understood as "a system's ability to interpret external data correctly, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation." Despite their undeniable relevance, the analysis of impacts that DX-enabled automation brings on the workforce, organizational structures, economy, government or society remains largely incomplete (de Sousa et al., 2019;Alon-Barkat and Busuioc, 2023).
More specifically, in the public sector, automation through DX can support the decisionmaking processes of public employees through the application of pre-established rules to existing data (Kuziemski and Misuraca, 2020).However, DX and the consequent automation cannot entirely replace civil servants' work, as the reality of the individual citizen who requests a service has specificities from time to time, that will never fully correspond to the general law or regulation.Therefore, human judgment remains essential to delivering public services (Borry and Getha-Taylor, 2019;de Boer and Raaphorst, 2023).The above should not lead to believe that DX does not bring with it enormous opportunities for change, along with associated risks, for the public sector.
The focus of our work is on measures such as efficacy, efficiency, productivity and savings that DX generates.On one hand, services are expected to be more effective and efficient, and on the other hand, the application of DX opens up problems such as the right to TG 18,1 equal access, fair treatment and privacy for citizens, and on the organizational side, it brings back the known issues of DX and unemployment risks, and the question of accepting farreaching changes in organizational structures and culture for employees and users to accept innovation (Willems et al., 2022;Neumann et al., 2022;Andersson et al., 2022).
The radical effects of DX on employment could take fundamentally two forms: a negative and substitution effect on labor demand due to technological replacement of human activities and skills, and a positive and complementary effect due to the creation of new industries and jobs (Frey and Osborne, 2017).
Based on these premises, the aim of this paper is to offer a new methodology to map out the immediate automation effects from DX on the public sector.In the existing literature, we observe a significant gap on how public management could measure the value of DX impact (i.e.immediate effects based on DX automation effects, interpretable as opportunities and downsides) based on an evidence-based approach, especially in a data-constrained environment.
A growing interest to have a human centric and inclusive DX adoption is observable both in theory and practice, yet current literature does not highlight how the distributional impact might be mapped across the PA.This methodology might be useful for policymakers to map out where most support is needed in terms of digital skills and competencies, and to better prioritize investment decisions and appraise potential opportunities and risks.
The paper is organized as follows.Section 2 analyzes the existing literature on the effects of DX-enabled automation in the public sector, and the main antecedents on automatability estimation.Section 3 illustrates the data and the different methodologies used in our analysis, distinguishing between estimates based on professions and on public services.Section 4 illustrates the findings.Section 5 provides a discussion of the findings and policy implications, whereas Section 6 concludes, also suggesting avenues for further research on the topic.

Literature antecedents on automatability or probability of automation from public digital transformation processes
In the literature, it is possible to distinguish between two main effects from a widespread DX adoption in organizations (Brynjolfsson and McAfee, 2014;Daugherty and Wilson, 2018;Davenport and Kirby, 2016): automation, that implies machines taking over human tasks, and augmentation, in which humans interact with machines to perform tasks.An intrinsic tension among these two effects of DX must be addressed, as DX automation effect would substitute human activities and skills for process rationalization and efficiency reasons (Davenport and Kirby, 2016), whereas DX's augmentation effect would enhance employees' effectiveness and productivity.
Despite a broad consensus on this tradeoff, these two effects cannot be neatly separated from a managerial perspective.Organizations adopting digital technologies have to be aware of these tradeoffs between complementarity and substitution, and between automation and augmentation effects.Following this theoretical approach, Raisch and Krakowski (2021) have argued that automation and augmentation effects from DX technologies are simultaneously contradictory but interdependent.
In 2017, Frey and Osborne (2017) implemented a novel methodology estimating the probability of computerization on jobs in the labor market, based on an algorithm able to identify the automatization or computerization of a specific job within the next 10-20 years, on the grounds of whether the overall job could be automated thanks to a widespread DX.

Automation in public sector jobs
The authors could find, using their algorithm, the probability of automation based on a series of computerization bottlenecks, across all occupation categories in the US labor market.
Digitalization or computerization bottlenecks are indeed those job's attributes that would be unlikely to be computerizable or automatable by emerging digitalization in the near future.In a nutshell, these bottlenecks include those variables that require the following: High level of perception and manipulation -For example, when sophisticated nonrepetitive, manual dexterity is required or those jobs where cognitive skills are needed with non-machine-readable data, e.g. in terms of problem solving.Creative intelligence -For example, when a specific job requires a high level of original, unusual ideas about a given topic or situation, e.g. in fine arts.Social intelligence -All those jobs that require a high level of social perceptiveness, negotiation, persuasion and assisting and caring for others, e.g.nurses and doctors.
Applying such analysis on the US labor market, they looked at the substitution effect of future computation on the mix of jobs across sectors distinguishing between high, medium and low risk occupations, based on their probability of computerization (thresholding at probabilities of 0.7 and 0.3).
The Frey and Osborne (2017) approach for probability of automation assumed that whole occupations rather than single job-tasks are automated by technology.This might have led to an overestimation on probability of automation, and hence risk of displacement.
For this reason, we looked at the OECD study by Arntz et al. (2016) that considered the heterogeneity of workers' tasks within occupations.As said, as occupations usually consist of performing a bundle of tasks, not all of which may be easily automatable, the potential for automating entire occupations and workplaces may be much lower than suggested by the approach followed by Frey and Osborne.
Based on such adjustments, the OECD paper was then able to estimate what jobs would be most/medium/least at risk of automation (using the same thresholding at probabilities of 0.7 and 0.3), based on their probability of automation applied to jobs for 21 OECD countries using a task-based approach.
Arntz et al. predict that more jobs are likely to experience change than be automated.In the UK, this is 25% and 10%, respectively.Change is likely when 50%-70% of tasks are automatable, whereas automation is likely when more than 70% of tasks are automatable.
Relying on such approach, recently the UK Office for National Statistics (ONS, 2019) produced estimates on occupations at risk of automation based on their underpinned probability of automation, based on the nature of their tasks and in a context of increasing DX.
The ONS analysis looked at the tasks performed by people in jobs across the whole UK labor market, to assess the probability that some of these tasks could be replaced through automation.
They identified that probability of automation tends to be higher for lower-skilled roles involved in routine and repetitive tasks, that can be carried out more quickly and efficiently by an algorithm written by a human, or a machine designed for one specific function.The three occupations with the highest probability of automation were estimated to waiters and waitresses, shelf fillers and elementary sales occupations, all of which are low skilled or routine.

TG 18,1
The automation estimates do not have a time element attached to them, but they can be interpreted as "the probability that this occupation will be partially, or fully, automated in the future."The ONS approach used the OECD task-based approach to obtain these estimates; the probabilities of automation can be not strictly interpreted as the percentage of tasks to be automated, rather than an overall probability of automation, calculated by looking at what impact each task within an occupation has on this figure.
Few studies have sought to apply these estimates to a specific national context.Using a measure of automation susceptibility, Adamczyk et al. (2021) show that approximately 20% of Brazilian Public Sector employees work in jobs with a high potential of automation in the coming decades.
Still, so far, none of these antecedents have attempted to apply these estimations to a measure of value of PA employment or PA services, to assess the magnitude of effects by DX-enabled automation.A point of novelty of our study is that we suggest using these automatability estimates with a precise purpose: to measure the actual economic value of the automation impact of DX, based on the aggregated displacement effect on the PA workforce or services.

Data and methods
In this section, we outline the main research methods and data used in this paper to appraise the scale and distribution of impacts from DX in the public sector, with a specific focus on its immediate automation effects.
Despite an accelerated uptake of emerging technologies and an increased interest from governments for DX across the whole public sector, most policymakers and public managers are not actually able to target public policies and investments, given they lack evidence to appraise which are the high-impact areas (OECD, 2020).Given the lack of "bottom up" primary data on what are the direct and indirect benefits, costs and risks from the adoption of digital technologies across the PA, governments might tend to draw their interventions based on anecdotal evidence from the private sector or based on off-the-shelf strategic priorities.
While there is not a sound and agreed approach on measuring the impact of DX in terms of augmentation and productivity effects, in recent years a consolidated body of work across the economics and management literatures has been formed to measure the automation effect of DX on organizations' business models and workforce.
Indeed, our paper proposes a new framework for public managers to map out the value of the impact from a widespread DX on the PA, focusing only on the automation effects (automatability), disregarding for now the augmentation and productivity effects.
3.1 A new approach to seize the value of the impact from digital transformation on public administration Focusing only on the automation effect, public managers might be able to estimate the impact of DX by applying the estimates on the probability of automation for PA occupations or PA services to the PA value, proxied by the cost of PA services or by the cost of PA workforce.
Given most OECD countries have developed a PA annual accounts data set on workforce and services, its public managers might be able to overcome the challenges from lack of primary data on the impact of digital technologies uptake using instead this new suggested methodology.
First, our approach would require policymakers and public managers to estimate the value of a country's PA or public sector; this can be based on its internal and external impact.

Automation in public sector jobs
Overall, we can identify two different approaches: an accounting-based input-output approach (where its outputs equal the costs of inputs and intermediate-inputs) and a qualityadjusted approach (where experimental analysis attempts measuring PA's outcomes adjusting the value of outputs).For simplicity and the purpose of this paper, we suggest public managers could use the cost of PA's inputs to proxy the value of a PA.
Given the commonality across countries' accounting systems, we suggest the following measures for public sector's inputs: the cost of PA workforce; and the cost of PA services.
Second, to implement such an approach, public managers would need to identify the probability of automation for PA workforce or for PA services.
As mentioned in the previous section, significant work has been done to estimate the probability of displacement or automation of jobs based on the mix and nature of its tasks, but we propose that a similar framework can be used also to estimate the impact on PA services.
The automatability estimates can be gathered based on secondary data, such as publicly available external estimates, or primary data, such as estimates obtained directly by experts with a theory-based structured survey consistent with the approach of previously mentioned studies.
Main studies such as Arntz et al. (2016) and ONS (2019) consider highly automatable public services or jobs when automation probability is or higher than 70%, based on the underpinning tasks/processes.
Similarly, they consider medium automatability with a probability of automation between 30% and 70%, based on underpinning tasks.Low automatability is defined when the automation probability is less than 30%, having considered its tasks.
These estimates are based on an average of scale-based scores where each expert consulted in those studies was requested to appraise the automatability of a job or a service based on the three already mentioned main cluster criteria: (1) extent of level of perception and manipulation; (2) creative intelligence; and (3) social intelligence requested to deliver the mix of tasks or processes underpinning a PA job or a service.
Immediate automation effect of DX from substituting jobs' tasks and services can be interpreted in a twofold manner: the uptake of digital technologies promises to lead to significant savings in terms of time and resources, but at the same time such displacement effect might need to be interpreted as a risk of job displacement, that could lead to increased income inequality.This latest aspect will be further expanded in the discussion section.
To condense what said so far, our new framework in seizing the automation effects from DX on a PA can be outlined as below: The total value of the impact of DX on a PA or public sector would be the result of a twostage process.
TG 18,1 First, public managers would need to proxy the organizational input value of a PA.As a starting point and based on Governments' accounting systems, we suggest this could be built on the aggregation of the PA workforce's costs (e.g. total PA employees' salaries) or based on the PA services' costs.
Secondonce a baseline for the value of a PA or public sector is obtained, this should be multiplied by the probability that such asset could be substituted or automated in a process of DX.In practice, the automatability of an entire PA or public sector would be defined as a weighted average of the automatability of all public occupations or public services.
While estimates could be based on secondary datacoming from existing estimates, or primary databased on directly measured with experts via a survey, the automatability of jobs' tasks or services could be assessed on the basis of the above mentioned three main computation bottlenecks that can be also defined as: the intensity of cognitive or non-cognitive tasks, the level of originality, dexterity required and presence of machine-readable data.

Data sources for methodological application to the case of Italy
For illustrative purposes, we aim in this paper to adopt the above mentioned methodological framework to map out the impact from a widespread DX across the Italian PA.Such approaches, by their nature and data required, are replicable to most of other OECD countries' PA.
We summarize in Table 1 the main data sources and specific data used.
The following paragraphs describe the methods used for each of the thematic focuses.

Seizing the digital transformation impact on Italian Public Administration, based on probability of automation of public administration employees' tasks
First, we map out the DX impact on Italian PA, based on external estimates on the probability of automation of PA occupations' tasks from the latest UK Office for National Statistics report (ONS, 2019).Building up on previous methodologies, the ONS study has For the application to Italian PA, we assume that such automatability effect from DX would be realized in its entirety within the next 10 years.
The time horizon of 10 years is a middle ground between ONS and OECD estimates which do not express a timeline (implicitly suggesting immediate or upcoming realization of the probability of automation) and estimates from Frey and Osborne, which indicate a time span of 10-20 years.
Once appropriately converted the 370 labels of UK to Italian occupational categories (e.g."Top-tier executives," "Healthcare professionals") based on same international ISCO codes, these estimates were applied to the Italian PA occupations, which are detailed in the MEF (Italian Ministry of Finance) Annual Account.
Once done, to apply the UK estimates to the Italian PA occupations outlined in the MEF data set, we converted the professional categories from Italian Institute for National Statistics (ISTAT) CP2011 with the taxonomy presented by the MEF Annual Account.To enable a robust conversion and comparison, we matched each of the 109 MEF's professional categories with a group of ISTAT occupational categories based on MEF and ISTAT guidelines, and following a discussion and validation by face-to-face and electronic correspondence with experts at MEF and ISTAT.This required, for example, matching job categories based on the information provided by the account and for a similar package of complexity and nature of tasks, knowledge and experience required.
As a result, we obtained for the first time in the Italian context a new dataset with each of the PA occupations' information (e.g.salaries, number of employees in that occupation) attached to a ISCO code and an estimate on the probability of automation coming from a secondary source.
As condensed by the below equation, once obtained the probability of automation for each of the 109 PA occupations, we would be able to measure the aggregated economic value from the weighted displacement effect of DX on each of the occupations' tasks: Value of the automation impact of DX on a PA For the sake of illustration, if a Health and Safety Officer occupation has an annual salary of e45,000 and based on the nature of the tasks, a probability of being automated of 35.8% within the next 10 years; the value of the automatability from DX reflects the portion of tasks displaced.
The value can easily be obtained by multiplying 3.58% (given we consider a 10-year horizon) for the associated salary.The aggregated result across all the 109 professions would produce an annual estimate of the value of the DX's displacement effect.

Seizing the digital transformation impact on Italian Public Administration, based on probability of automation of public administration services
Second, we also applied our new approach to estimate the immediate DX impact on Italian PA based on the automatability of public services, and their underpinning processes.
Given the absence of current estimates produced on the DX automatability effect on services, we conducted structured questionnaires on a sample of experts to gather their estimates on the probability of automation of PA public services.

TG 18,1
While under the first approach we estimated the value of the potential DX impact on PA by multiplying the probability of automation by occupation for the associated cost (i.e. the PA occupations' salary), under the second approach we estimate the potential DX impact's value by multiplying the probability of automation by PA public service for the associated cost to deliver the service.
Despite we do not have at this stage the break-down of costs by Italian PA public service, we used the associated cost of labor across different types of PA services e.g."benefits, grants and loans" or "issuing authorizations."The approach can be summarized by the following formulation: Value of the automation impact of DX on a PA ¼ e X PA occupationsg x f% PA 0 s job n o Hence, beyond the list of public services delivered by the PA, ISTAT provided information on the % of staff involved on a specific PA activity or service ("Employed staff dedicated to direct management services/employed staff").
ISTAT grouped such statistics at aggregate level by macro-category of institutions, e.g.regions, municipalities and ministries.These statisticsin conjunction with the information on the cost of labor across several PA institutions provided by the MEFwere relevant for us to provide an initial estimate of the labor costs spent to deliver each of the 53 registered public services.
For the sake of illustration, if the public service "benefits, grants and loans" or "issue authorizations" has an annual associated labor cost to deliver it of e45m and based on the nature of the tasks, a probability of being automated of 45.5% within the next 10 years; the value of the automatability from DX reflects the portion of specific processes that can be displaced.
The value can easily be obtained by multiplying 4.55% (given we consider a 10-year horizon) for the associated production cost.The aggregated result across all the 53 types of services would produce an annual estimate of the value of the DX's displacement effect.
Following we describe more in depth the design and how we obtained these set of estimates grounded on experts' feedback, based on semi-structured questionnaires.
To identify a framework to seize the impact of DX, a content analysis method (Krippendorff, 1980;Bardin et al., 2010) was applied to analyze the surveyed literature indepth, associated to main studies such as those by Frey and Osborne (2017), Arntz et al. (2016) and ONS (2019).
After deciding on the purpose of the study, the first step entailed validating the correctness and completeness of such framework.To this aim, this was sent to five between academic and practical experts in Emerging Disruptive Technologies and Public Policy Innovation in the UK and Italy, who provided comments that were incorporated to refine the terminology, to make it more generalizable.These were selected based on their fit and track record on the subject as well as on their background, to make sure to include experts from the industry as well as from the government.
As a result, we could then validate the three main clusters of computation bottlenecks: extent of level of perception and manipulation, creative intelligence and social intelligence requested to deliver the mix of tasks or processes underpinning a PA job or a service.
After validation of the framework, in a second step, a questionnaire was developed asking for an estimate on the "Probability that the PA public service could be automated within next 10 years" for each of the 53 labels of public services as defined by ISTAT PA.

Automation in public sector jobs
The questionnaire was developed alongside an explanatory note designed to provide the purpose of the study, the context and necessary background information to properly conduct the exercise.Hence, the note outlined the theoretical framework with the definition of the three bottlenecks, associated research findings and how to score for the automatability of the services.
In the note, we also outlined insights from recent ONS research studies ( 2019) on what words from job adverts were associated with occupation at high or low risk of automation (e.g."machine," "operate," "check" versus "control" and "plan," "research" and "advise").
For each of the PA services, we asked for the interviewee's view, to provide a score based on the same scale mentioned (0%-100%) once considered the presence of computation bottlenecks.
Once identified the research questions in the questionnaire and explanatory note, three experts were selected, two senior managers from a large digital technology company and from an IT strategic consultancy, and one senior manager from the UK Government expert on public programs on digital technologies.
The identification of the participants was based on their expertise as well as on their background.One common standard in defining the qualitative sample size was considering when reaching thematic saturation, which refers to the point at which no new thematic information is gathered from participants.Given the diverse background and specific competences of the interviewees as well as the number of their colleagues involved for validation, we consider at this stagegiven the illustrative purposes of our framework's applicationenough information power was reached.
Once identified the interviewees, we had to consider the best ways to contact them, obtain informed consent, arrange interview times and locations.An initial contact was made through email and followed up with more details so that the individual can make an informed decision about whether they wish to be interviewed.The participants were also informed that they can refuse to answer questions or can withdraw from the study at any time, including during the interview itself.
An online conversation with each of the three experts was held to explain the purpose and the process of the exercise and helping the experts with any doubt.Ahead of the meeting, the explanatory note and the questionnaire in excel were sent by email to the preselected experts.
In terms of the design of the actual interview, we decided to carry out an online survey providing two weeks of time to send it back with the actual responses rather than running a face-to-face survey.The main advantages were that this allowed the respondents to take their time to study the material and provide the best answer as possible after a deep-dive research and an internal consultation.In the meantime, several interactions with the experts were conducted to clarify issues concerning concepts and completion of the questionnaire.
All these things taken into account, we considered these benefits outweighed the inherent limits from online questionaries, such as not being able to record verbal and non-verbal cues.
While our methodological approach was based on gathering information from these three experts only, we are aware that each of these respondents followed consulting and validating on the research questions in their own organization with practitioners and colleagues with a strong experience on evaluating automation technologies and their impact.
The number of practitioners reached by each expert ranged between 3 and 8; the estimates were provided following an internal consensus (at least 51% agreement).
Ultimately, as a result, each of the 3 experts returned the questionnaire with associated estimates of automatability for each of the 53 public services.For each of the estimates, experts also provided examples on how to tackle upcoming rounds of refinement of such TG 18,1 research.Further details on the estimates from the three experts can be found in the Appendix 1 (Table A2).
We acknowledge some of the limitations this approach presents: for instance, due to the COVID-19, travel restrictions meant that it was not possible to have face-to-face interactions as initially planned and conversations with the experts and between the experts and their teams were run remotely.
Additionally, while the estimates collected by the three experts reflect the views of a much larger audience based on the size of their respective teams, we aim to further increase the scope and range of the panel of experts involved in future estimations.Figure 1 shows the obtained probability of automatability across PA occupations based on their tasks.We provide the granular representation of these estimates in the Appendix 1 (Table A1).
Overall, our analysis estimates an average probability of 32.3% that PA jobs within the next 10 years have the potential to be automated, once considered their underpinning tasks.Without a qualification, manual and office-based occupations tend to be the PA occupations with the highest rate of automation.
Hence, the occupations with a higher percentage of manual, administrative and routine tasks that require low competency tend to be the ones most likely to be automated within the next 10 years.
Professions such as doctors, directors and teachers tend to be the ones least likely to be automated, due to their high percentage of specialized cognitive and non-cognitive capabilities.

Probability of automation by profession in the
Italian PA

Automation in public sector jobs
Once we identified the automation probability for each PA occupation category, we could seize the total absolute value of the impact from DX considering the individual automatability of each occupation based on its tasks, the salary and the volume of employees per organization.Taking a conservative approach, assuming that the value of the impact should be proportionate to how widespread is the level of DX technologies adoptionwe identified a baseline value of annual displacements in the PA in the next 10 year for e3.45bn.It is worth flagging how these potential "time/cost savings" estimates are associated to the actual widespread uptake of digital technologies.A successful scale up of DX technologies would depend by multiple exogenous and controllable factors such as the ability of the PA's change management to implement the right enablers and delivery mechanisms to maximize DX opportunities.
That said, as it is possible to see in Figure 2, in absolute terms the greatest automation opportunities come from those occupations in the PA which have a high intensity of administrative and machine-readable routine tasks.These include civil servant and public servant roles such as Categories C and D and Areas A and B as well as "Assistants" and "Personal A.T.A." (school non-teaching staff).
4.1.2Seizing the digital transformation impact on Italian Public Administration, based on probability of automation of public administration services.Based on the methodological framework mentioned and primary data, with estimates from a pool of industry and academic experts, we could then estimate the automatability for each PA's public service, and whether these might be more or less likely objects of automation within a widespread DX.
As mentioned above, first, we estimated the total value of PA public services, based on the labor costs allocated to each public service.
Applying previously mentioned methodology and experts' assessment on the probability of automation, we were able then to map out low, medium and high automatable public services, and consequently, the associated value of processes displaced from automation with DX.
Based on this second methodological strand, based on the probability of automation of PA public services, overall our analysis estimates the baseline annual value of displacement in the PA for e3.05bn over the next 10 years.This value does not differ significantly from previous estimations based on the probability of automation of occupations.As shown in Figure 3, results suggest the highest impact from DX automation effects to come for transactional activities such as copy of judicial documents, issuing authorizations and certifications and other administrative office activities.
The PA public services that would involve a high intensity of dexterity, original tasks and that require sophisticated cognitive capacities would tend to be less automatable.These include, for example, PA public services such as "activities of nurses and physiotherapists (other human health activities)" or "teaching activities (tertiary education: graduate or postgraduate level; arts schools)." On the other hand, public services can be defined as highly automatable when these involve a low level of cognitive (such as judgment and computation) and non-cognitive (such as caring) skills, and when these involve mostly repetitive, manual and machine-readable tasks.
These include for example, PA public services such as "copy of judicial documents (law and justice)" and "issue of authorizations, of statements, of certifications and of appraisals (law and justice)."

Discussion and policy implications
The application of our novel methodological framework to the Italian PA highlights how this could hand over to public managers a useful tool to anticipate and map out the potential impacts and risks from a widespread uptake of digital technologies across a public organization.This framework would become more and more valuable in contexts where primary data are not available, and more generally in a data-constrained environment.
The underpinning framework is replicable, and the same approach could be applied to other countries' PA.In terms of next iterations, we aim to enhance our framework to ensure a more granular and detailed distributional analysis and map out the DX impact across the PA also by type of organization and across geographies.
In relation to the mentioned limitations of our methodology, the work could certainly be improved in terms of the robustness of the estimates obtained on the DX automation effect, relying on a broader pool of experts and triangulating the results with emerging bottom-up evidence.Furthermore, we are conscious the paper applies estimations based on a different context to the Italian sector and does not factor in the cost of new technologies in the Automation in public sector jobs projections provided.These limitations could be addressed in future versions of the paper, with a wider level of data availability.This new methodological approach would also enable to look for the first time at the distributional impact of digital technologies across the PA workforce and on how automatability would relate with the average salaries.
As shown in Figure 4, carrying out a correlation analysis, initial results seem to show a negative relationship with a Bravais-Pearson correlation coefficient of À0,52, suggesting that on average PA employees that have a lower salary tend to have more automatable tasks, leading to relevant consequences in terms of policy implications.
That said, further iterations of this approach could be aimed to strengthen and test these insights carrying out a more comprehensive econometric analysis to test the statistical significance between these variables, and appropriately treat factors affecting causation such as the endogeneity problem.
More generally, we consider that these and previous results should be interpreted with caution.As suggested by Arntz et al. (2016), the methodological approach on estimating the displacement impact from DX based on its automation effect, still reflects technological capabilities rather than the actual utilization of such technologies, which might lead to a further overestimation of the occupation automatability.For this reason, we attempted mitigating these by spreading the opportunities over a longer period of time.
In addition, even if the new DX technologies are increasingly adopted across the PA, the prospects of task substitution depend on whether PA workplaces adjust to a new division of labor, as workers may increasingly perform tasks that are complementary to new technologies.
Other limitations include that our approach considers only existing jobs, although new digital technologies are likely to create new jobs and associated tasks.
All these things considered, the study reinforces, in line with extant literature (Adamczyk et al., 2021;Arntz et al., 2016) how low educated workers across the PA likely will bear the These findings point toward the need to focus more on the potential inequalities and requirement for the re-training arising from technological change and on how public management would be able to rely on DX to ensure inclusive income growth, i.e. reducing the gap between high and low salaries across the public sector through higher labor productivity boosted by DX.
Furthermore, this framework could assist policymakers' choices on public employment.By enabling informed, data-driven analysis on the professions and services most exposed to automatability, such analyses could help singling out areas which are more prone to substitution or complementarity, thus informing the prioritization of workers' profiles and competences to be included in recruitment processes.

Conclusions
Based on lessons learned from close disciplines, such as economics and automation engineering, our paper proposes a new framework for policymakers and public managers to disentangle and measure the impact from digital technologies uptake on the public sector.
Building on this, our paper suggests a new and replicable framework for public managers to measure the value and risks from DX investment on PA, which are adopted in part from existing best practices in other disciplines and applied originally to a public management setting.
The new framework for policymakers estimates the economic value of the automation impact of DX on the public sector using the total earnings of public employees and the total costs of public services.
Having caveated the assumptions and its limitations, these new tools would be ultimately useful as a basis for policymakers and public manager to make more targeted evidence-based DX investments' decisions in a data-constrained environment, and to better target new public policies to support digital inclusion, such as training and re-skilling, and recruitment.
For illustrative purposes, applying the new approach to the Italian PA, our analysis highlighted initial findings on the overall value of the impact on PA as a whole and by type of PA profession and by PA service.
Results show a strong expected impact from DX on Italian PA revealing the sign and the magnitude of promises of efficiency gains as well as raising concerns in terms of potential risks of job displacement and of arising income inequality across the PA workforce.Further applications of this framework could vertically target specific service chains or geographical clusters, to provide more qualitative and nuanced evidence on the automation/augmentation tradeoff and the associated change management challenges in specific administrative settings.
methodological application to Italian Public Administration 4.1.1Seizing the digital transformation impact on Italian Public Administration, based on probability of automation of public administration employees' tasks.Matching the UK with the Italian occupational categories, we were able to use the latest ONS study estimates on the probability of automation for the Italian PA professions.
Figure 1.Probability of automation by profession in theItalian PA Figure 2. Value of the impact from digital transformation considering the individual % automatability of each occupation based on its tasks, the salary and the volume of employees per organization in the Italian PA

Figure 3 .
Figure 3. Value of DX impact on PA public services, based on the labor costs allocated to each public service in the Italian PA Figure 4. Distributional impact of digital technologies across workforce in the Italian PA the jobs of 20 million people in the UK in 2017 and has labeled each of these occupations with a probability of automation.For example, Health and Safety Officers have a 35.8% probability of being automated, while waiters and waitresses a 72.81%. analyzed

Table A2 .
Corresponding authorAndrea Bonomi Savignon can be contacted at: bonomi.savignon@economia.uniroma2.itFor instructions on how to order reprints of this article, please visit our website: www.emeraldgrouppublishing.com/licensing/reprints.htm Or contact us for further details: permissions@emeraldinsight.comAuthors' and experts elaboration based on ISTAT PA and MEF occupational classifications