Feasibility and safety of the practical functional assessment at scale: findings from a multi-state ABA organisation

The functional analysis (FA) has been central to applied behaviour analysis (ABA) since the seminal work of Iwata et al. (1994), providing a systematic method for identifying the environmental variables that maintain behaviour of concern. By isolating the reinforcement contingencies responsible for behaviour of concern, FA enables clinicians to develop individualised, function-based interventions rather than relying on topography alone (Beavers et al., 2013). The Behavior Analyst Certification Board designates FA as a “precise, effective, and client-centred approach” (BACB, 2014, p. 11), and it remains the gold standard for behavioural assessment in research and clinical practice alike.

Despite its scientific rigor, traditional FA has proven difficult to implement broadly in applied clinical settings. The methodology can require specialised training, extended session durations, a high degree of environmental control and the deliberate evocation of behaviour of concern, factors that collectively discourage widespread adoption (Hanley, 2012; Oliver et al., 2015; Roscoe et al., 2015). Many clinicians report avoiding FA due to concerns about safety, resource demands and insufficient training opportunities (Roscoe et al., 2015). Proponents of trauma-informed care have further cautioned that repeated exposure to aversive conditions may itself constitute a harm for some clients (Rajaraman et al., 2022). The practical functional assessment (PFA) represents a direct methodological response to this concern. Rather than treating the evocation of severe behaviour as a procedural requirement, the PFA builds the assumption of distress into its design: the test condition terminates at the first sign of early behavioural escalation and clients spend the majority of each assessment in an enriched environment where preferred activities and social interaction are freely available. In this way, the PFA reframes functional assessment not as a process done to the client but as a collaborative process structured around the client’s comfort and safety at every stage (Rajaraman et al., 2022; Iovino et al., 2022). As a result, despite its status as an ethical and clinical standard, FA remains underused at scale and evidence documenting its large-scale application within service organisations is limited (Oliver et al., 2015; Roscoe et al., 2015).

It should be noted that throughout this paper and the broader literature, the terms “Practical Functional Assessment” (PFA) and “Interview-Informed Synthesised Contingency Analysis” (IISCA) are used interchangeably to refer to the same assessment process. The PFA was developed to address the barriers described above by offering a streamlined, clinician-friendly assessment process that can be feasibly embedded within routine service delivery (Hanley et al., 2014). The version of the PFA used in this study incorporates several features that distinguish it from earlier FA methodologies, including those using synthesised contingencies (Hanley et al., 2014). The concept of synthesised contingencies refers to the combination of multiple suspected contingencies rather than isolating each one individually, into a single condition during a FA (Jessel et al., 2016; Iovino et al., 2022). Rather than alternating conditions that test individually isolated behavioural functions, the PFA uses a synthesised test condition, in which identified triggers and motivating conditions are present to make behaviours of concern more likely, constructed from open-ended interview data (Jessel et al., 2016). This is contrasted with a synthesised reinforcement (SR) control condition in which all identified reinforcers are freely available, approximating a “happy, relaxed, and engaged” (HRE; Gover et al., 2022) baseline state. A critical procedural safety feature is that the establishing operation (EO) condition terminates and SR access is provided immediately contingent on either precursor behaviour or behaviour of concern, allowing the analysis to proceed without requiring escalation to high-intensity responses. The PFA responds to the full targeted response class, including precursors, rather than waiting for severe behaviour (Warner et al., 2020). Together, these features represent a meaningful departure from traditional multielement designs in both their theoretical framing and clinical application (Jessel et al., 2018; Gover et al., 2022).

Iovino et al. (2022) introduced a performance-based adaptation of the PFA that incorporated two procedural features with direct implications for safety and implementation accessibility. Firstly, rather than waiting for escalation to behaviour of concern, clinicians provided SR contingent upon the first instance of any member of the open-contingency response class, including associated low intensity behaviours of concern, using an open-response class framework informed by Warner et al. (2020). This allows the analysis to capture functional control without requiring exposure to dangerous topographies. Secondly, the EO was not re-presented until the client had been calm, relaxed and engaged (HRE; Gover et al., 2022) for a minimum of 30 consecutive seconds, ensuring that the next EO presentation was contingent on the client’s behavioural state rather than the passage of time alone. Together, these features distinguish the performance-based PFA from its session-based predecessor (Hanley et al., 2014) and provide a concrete, performance-driven structure that clinicians can implement without requiring escalation to severe behaviour. The version of the PFA used in the present study operationalised these performance-based procedural standards: EO progressions were contingent on client behavioural state rather than the passage of time and any member of the open-contingency response class, including precursor topographies, triggered immediate termination of the EO and return to the SR condition.

Although the PFA has demonstrated utility as a pragmatic assessment model in small-scale research contexts (Coffey et al., 2019), its feasibility within large, geographically distributed organisations, those serving hundreds of clients across multiple states and settings, has not been systematically evaluated. For the purposes of this paper, “large-scale implementation” refers to the coordinated delivery of a standardised assessment protocol across a multi-state organisation involving multiple levels of implementers (technicians, clinicians and coaches), diverse service settings (clinic, home and community) and varied coaching modalities (in-person, telehealth and on-site observation). Implementing the PFA at this scale requires operationalising its core features in ways that allow for clinically necessary flexibility while preserving procedural integrity, and establishing a training and support infrastructure capable of sustaining consistent implementation across implementers and sites.

Standardisation is a necessary precondition for this kind of scaled implementation. Within any ABA service organisation, but particularly in large ones, standardised protocols ensure that consistent practices are delivered across all clients and sites, prevent drift from core clinical expectations and enable meaningful measurement of outcomes using common metrics (Silbaugh and El Fattal, 2022; Bakkeli and Breit, 2022; Townsend et al., 2024). Without standardisation, variability in care delivery compromises both data integrity and the organisation’s ability to evaluate its own clinical effectiveness (Committee on Quality of Health Care in America, 2001; Rosoff, 2004). Yet standardisation alone does not guarantee adoption. At the organisational level, successful scaling also depends on values alignment, evidence-based training systems, data-based decision-making and structures that support sustained implementation over time (Biglan, 2015; Horner and Kittelman, 2022). Clinicians responsible for applying evidence-based strategies in dynamic, real-world settings must also have access to assessments that are safe, efficient and immediately actionable (Slocum et al., 2014), criteria the PFA was specifically designed to meet.

The present study describes the implementation of the PFA across a large, multi-state ABA organisation serving clients in clinic, home and community settings. Specifically, we report on the safety and functional control outcomes of 313 PFAs conducted by 251 clinicians supported by 23 coaches, examining whether the PFA can be implemented with high rates of safety and clear functional control when delivered at this scale. The study also documents the training and coaching model used to support implementation, including variation in training format (10-h course vs 2-h webinar) and coaching modality (in-person, telehealth and on-site observation). The findings contribute to an emerging empirical base examining the conditions under which function-based assessment can be feasibly and safely disseminated within large service organisations.

Participating clients were 314 individuals receiving insurance-funded ABA services. Ages ranged from 2 to 22 years (M = 7.2, SD = 4.6, mode = 4), reflecting the breadth of age ranges served within this organisational context. All participants had a diagnosis of autism spectrum disorder. Clients were selected by their supervising clinicians to receive a PFA based on clinician judgment informed by a structured clinical decision aid that evaluated multiple behavioural and contextual factors. These included the frequency and intensity of behaviours of concern, the degree to which the client demonstrated difficulty tolerating delays or denials of reinforcement, the reliability of the client’s functional communication and the effectiveness of previously implemented behavioural interventions. Clients were considered appropriate candidates when behaviour of concern was clinically significant, characterised by high-intensity topographies lasting more than 30 s or behaviours posing imminent safety risks to the individual, others or property and when existing interventions had produced limited or variable behaviour change. Selection reflected a higher composite clinical concern rather than a single threshold criterion; no formal cutoff score was applied. Referral was based on clinician judgment following review of existing behavioural data and caregiver report. The inclusion of clients across multiple contingency profiles, including those with automatically reinforced components, indicates that the sample was not limited to individuals presenting with difficulty tolerating delays or denials of reinforcement and supports the generalisability of the PFA procedures across a broader range of functional presentations. Of the 314 referred clients, 313 completed a PFA; one client did not complete the assessment due to safety concerns identified during the design meeting and is excluded from all outcome analyses.

Participant characteristics are summarised in Table 2. Multi-tiered treatment teams composed of technicians, clinicians and coaches (who included residents and directors) supported each PFA. A description of each role is provided in Table 1. In total, 92 technicians and 251 clinicians participated in implementing the PFAs; 23 Coaches monitored the PFAs, consisting of 7 directors and 16 residents. All Coaches were board-certified behaviour analysts (BCBAs) with an average of five years of certification among the directors and an average of 2.8 years among the residents.

PFAs were conducted during regularly scheduled ABA sessions across various settings. One hundred and seventy-four sessions occurred in a clinic setting, 136 were conducted in the participant’s home and four were conducted in community settings. Data were collected across 12 US states, with the most considerable proportions coming from Michigan (40.8%) and Arizona (17.5%). The remaining PFAs were conducted in California, Georgia, Indiana, Massachusetts, Maryland, Minnesota, New Mexico, Oregon, Texas and Virginia. The setting data are summarised in Table 2.

The IISCA + app (Link to garage94Link to the cited article) was developed to record and analyse behavioural responses and environmental events associated with running a performance-based IISCA, as Iovino et al. (2022) outlined. Data collection during coaching sessions and PFA implementation was performed in real time using the IISCA + app.

Materials needed to create an SR (control condition) and EO (test condition) were individualised for each client. Consideration for inclusion came from the open-ended interview with parents or staff and reported preference or non-preference for the items and activities.

All telehealth and video conferences took place over Health Insurance Portability and Accountability Act (HIPAA)-compliant platforms. The implementer and/or coach used a laptop with internet access and wore Bluetooth earbuds to communicate during the analysis. Zoom, a virtual meeting software, was used to allow team members to join remotely and maintain real-time audio access throughout the session.

These documents summarise the order of events, including the analysis process and the necessary components required to implement a safe analysis. Categories for these steps include preparation, SR condition, EO condition and procedural considerations. Each section was scored using a rating scale from 0 to 5, ranging from 0 = “significant coaching required across all components” to “5 = implemented all components without coaching.” The score was then calculated using percent (total score obtained divided by total possible score multiplied by 100).

Artificial intelligence writing assistance (Claude, Anthropic) was used during the preparation of this manuscript to support structural editing, prose revision and formatting of the reference list in accordance with Emerald Harvard style. AI tools were not used in data collection, data analysis or the interpretation of findings. All AI-assisted content was reviewed, edited and approved by the authors, who take full responsibility for the accuracy and integrity of the submitted work.

The dependent variables in the analysis were the frequency and topographical class of behaviour of concern, the safety of the session, the control and duration of the conditions and the sessions.

Behaviours of concern were classified into two response classes, identified as low-intensity and high-intensity behaviours of concern (Iovino et al., 2022). Low-intensity behaviours were all behaviours reported during the interview with duration under 30 s that reliably occurred before behaviours of concern (e.g. tensing, whining, crying, pouting, ignoring and turning away, vocal protest or pulling items away from others (Heath and Smith, 2019; Iovino et al., 2022; Najdowski et al., 2008). These behaviours were not disruptive to the environment or others and did not pose a risk to the client or others in the room. High-intensity behaviours were identified during the interview as behaviours with a high magnitude or duration over 30 sec (e.g. tantrum, crying, screaming) or any attempted or completed behaviour that would result in imminent risk of harm to the client or the implementers, including but not limited to aggression (e.g. hitting, kicking, scratching, pinching), severe environmental destruction and self-injury (e.g. head hitting with fist or open hand, head banging on objects, arm or finger biting). It should be noted that this definition aligns with more recent literature by Jessel et al. (2023) and Iovino et al. (2022). By expanding the response class definition and including less obvious topographies of lower-intensity behaviours (e.g. flat affect, facial grimace, attempted flopping, turning away from the Implementer, physically resisting, etc.), the implementers aimed to prioritise the safety and dignity of the client and the safety of the Technician and Clinicians that were involved in each analysis. Responses were recorded using the IISCA app as “R1” (high intensity behaviour of concern) or “R2” (low intensity behaviours of concern); the durations of the SR and EO contexts were also recorded as components of the total duration (Iovino et al., 2022).

A session was considered safe if all instances of behaviour of concern ceased after presenting all relevant known reinforcers by the end of the session and no behaviour resulted in injury or risk of injury to the client or Implementer. Sessions were terminated if behaviour of concern did not cease after the presentation of all known reinforcers or if there was an injury or risk of injury.

Control in the analysis was defined as the differentiation between the occurrence of an R1 or R2 behaviour during the presentation of the EO and a consistent absence of those same behaviours and the presence of HRE during the SR condition across a minimum of three consecutive replications.

The duration of analysis was measured from the onset of all available ecologically relevant reinforcers being presented to the client and ended after a minimum of 30 s was established in the control condition, where the child was observed to be happy, relaxed and engaged following the third or subsequent presentation of the EO condition.

The degree to which an implementer independently and accurately conducted the components of the PFA was measured using a rubric of target behaviours required to implement a safe and accurate analysis. This rubric was scored independently by the coach during the analysis.

Retrospective social validity was used to establish the potential differences between the perceived social validity score before the process and the score provided after the process (Hill, 2020; Little et al., 2020). Traditional pre/post social validity may have resulted in response shift bias, where throughout the process, participants may have calibrated their prior understanding of the processes, resulting in lower post-test scores than their pretest scores, even though their knowledge increased (Hill, 2020). Although this may not have always occurred, the retrospective pretest (RP)–post-test social validity provided a quantitative measure of the degree to which a clinician may have been affected by a program (Hill, 2020). At the end of the process, clinicians were allowed to report their current perception, attitude or belief, then similarly retrospectively think back to the onset of the process, providing a consistent frame of reference to respond from (Little et al., 2020).

The organisation is a large, multi-state provider of insurance-funded ABA services operating across 12 US states, serving more than 3,000 autistic clients supported by over 350 supervising clinicians and approximately 4,000 behaviour technicians. Its clinical model is structured as a tiered continuum of care. All clients receive a universal, tier-one behaviour support plan grounded in principles of assent, enriched environments and the development of functionally equivalent skills, an approach analogous to a standard behaviour intervention plan, but explicitly oriented around safety, dignity and the client’s quality of life. PFA is reserved for a subset of clients, estimated at 15%–20% of the total caseload, who present with clinically significant behaviour of concern that requires more intensive functional assessment. Importantly, PFA was not an optional clinical tool: its use was embedded as a professional expectation for clinicians whose clients met referral criteria for intensive behavioural support. This organisational mandate reduced the self-selection bias common in voluntary adoption studies, where implementation rates may reflect clinician confidence or motivation rather than the protocol’s actual feasibility under routine conditions.

The decision to systematise PFA across the organisation was preceded by deliberate values work at the senior clinical leadership level. The organisation’s clinical framework is anchored in three guiding principles, safety, dignity and trust, and the adoption of PFA was driven in part by recognition that traditional FA methodologies posed barriers to realising those values consistently across a large, geographically dispersed workforce. Conducting lengthy or aversive assessments that repeatedly evoke behaviour of concern was viewed as inconsistent with a client-centred, trauma-informed approach to care. Rather than treating clinician avoidance of functional assessment as an individual skill deficit, leadership took organisational responsibility for building the infrastructure needed to support competent, values-aligned implementation at scale.

The PFA data in this study were collected across two sequential mentorship models, representing the organisation’s initial rollout of structured assessment training prior to its current train-the-trainer architecture (see Table 3 for a side-by-side comparison of both models). In both models, clinicians who completed PFA training also received concurrent training in skill-based treatment (SBT), a behavioural intervention informed by PFA findings that uses synthesised contingency arrangements to develop functional communication, tolerance of delays and denials and cooperation through contextually appropriate behaviour (Hanley et al., 2014).

In mentorship Model 1 (July 2022–February 2023), clinicians were trained in small-group cohorts recruited through three pathways: volunteer enrolment, team-based cohorts and newly hired clinicians. The 14-week program began with a 10-h external training course on PFA, followed by an individually supported analysis design meeting and one week of directly coached PFA implementation. The remaining weeks of the program focused on SBT implementation. All coaching was provided directly by the organisation’s central specialist team.

In mentorship Model 2 (April 2023–February 2024), the model was expanded to region-wide rollouts across a 20-week timeline. A locally embedded supervisory clinician, a resident in the organisation’s clinical leadership training program, assumed responsibility for coaching the tier-one behavioural support plan during the initial four weeks, while the specialist team retained direct oversight of PFA and SBT phases. Graduation criteria were expanded beyond PFA and SBT competency to include supervisory skill development, tier-one implementation fidelity and the full progression of SBT stages. This expansion reflected the organisation’s intent to develop clinicians not only as skilled implementers but as coaches capable of sustaining and disseminating evidence-based practices within their local teams, a key consideration for any organisation seeking long-term scalability.

Across both models, the coaching infrastructure was organised around three delivery modalities: synchronous group learning (approximately 8 h, covering clinical values, support plan development and analysis design); asynchronous self-directed content (approximately 12 h, via continuing education modules aligned to the current stage of treatment); and live consultation during active client sessions (20 or more hours, provided by specialist team members in real time). Live consultation was delivered through in-person observation, on-site co-facilitation and telehealth support, the latter accounting for 68.2% of all coaching contacts in this sample. This blended architecture was designed to balance the practical constraints of a large, geographically distributed workforce, including clinicians operating across clinic, home and community settings, with the need for individualised, competency-based coaching at the precise point of implementation. organisations seeking to replicate this approach should note that the infrastructure investment is substantial, but that the blended delivery model makes it achievable without requiring physical co-location of specialist staff.

Clinicians received one of two types of preparatory training prior to supported PFA implementation. In mentorship Model 1, the first 89 clinicians completed a 10-h foundational course from a third-party provider covering the values, procedures and clinical decision-making framework underlying the PFA. In mentorship Model 2, the remaining 162 clinicians completed a 2-h internally developed webinar that introduced the values underpinning the assessment, the structural components of the SR and EO conditions and the role of precursor behaviour within the response class hierarchy. Both pathways included orientation to the open-ended caregiver interview process and the PFA Design Workbook, a structured planning tool used to operationalise SR condition parameters, specify EO progressions, identify known precursor and high-intensity behaviour of concerns and confirm logistical requirements prior to implementation. Clinicians were expected to complete the caregiver interview and prepare a draft workbook before attending their design meeting. Technicians assigned to participating cases completed a separate 1-h overview webinar focused on the values and implementation components relevant to their direct service role.

Following preparatory training, all clinicians attended a 2-h design meeting with a Coach, conducted in small groups of two to three. These meetings were structured to accomplish a defined set of outputs: identification of SR condition components and delivery parameters, specification of the EO progression steps and known contextual triggers, environmental design planning for the assessment space, development of contextually appropriate behavioural targets per client and scheduling of the PFA before the meeting closed. The design meeting was the primary mechanism for individualising the standardised protocol to each client’s profile and setting and coaches used the session to identify and resolve design gaps that could compromise safety or functional differentiation during implementation.

Coaching during PFA implementation was provided by specialist clinicians from the organisation’s central clinical team, senior BCBAs with direct expertise in PFA and the broader foundations of care model. These individuals coached clinicians in real time and scored implementer independence using the structured rubric described in the dependent variables section. Coaching was delivered via telehealth (68.2%), in-person (24.8%) or through on-site observation from an adjacent space (7%). Coaching, clinician and implementer characteristics are summarised in Table 4. Telehealth sessions used HIPAA-compliant videoconferencing platforms with bug-in-ear headset communication, allowing coaches to provide real-time guidance without disrupting the session environment or alerting the client. This modality was a deliberate infrastructure choice enabling consistent coaching support across clinic, home and community settings without requiring coach co-location, a practical necessity given the geographic distribution of services across 12 states.

The coaches asked the implementer to see the area for the analysis (including but not limited to the SR context, materials and space needed for the EO context) and check to ensure all technology was working (including a bug-in-ear headset to ensure communication was provided directly to the implementer without the client hearing). The coach reviewed the steps to the procedure in general terms, reminded them of all necessary steps and components and said they would collect data using the IISCA + app.

The analysis used the same procedures outlined by Iovino et al. (2022) except for consideration of control, further explained here. Each assessment consisted of alternating SR and EO conditions. In the SR condition, all identified preferred items, activities and social interactions were freely available and no demands were placed on the client, approximating a calm, positive baseline state in which the client was HRE. In the EO condition, identified motivating conditions were introduced: preferred items were withdrawn or restricted and contextually relevant demands or expectations were presented based on caregiver interview data. The EO condition terminated immediately upon the occurrence of any member of the defined response class, including low-intensity precursor behaviours of concern, at which point all reinforcers were delivered simultaneously. The SR condition remained in effect until the client had maintained an HRE state for a minimum of 30 consecutive seconds before the next EO presentation was initiated. Functional control was considered achieved when a minimum of three replications produced consistent occurrences of behaviour of concern during EO conditions and a consistent return to zero rate during SR conditions. All steps that the implementer did not independently complete were coached to ensure accuracy.

Coaching sessions were conducted via telehealth (68.2%), in-person (24.8%) or on-site from another room (7%). The implementation of PFAs was primarily conducted by clinicians (70.7%), followed by Technicians (24.2%) and occasionally involved a combination of the coach, clinician and/or technician (5.1%). 22.0% of clinicians provided coaching to technicians remotely via telehealth.

The results of the 313 PFAs are presented in Table 5. Analyses were reviewed to determine if they met the criteria for control and safety, the duration of the analysis and the average time spent in each condition and rates of the two categories of behaviour of concern that occurred in each condition. Specifically, 309 (98.72%) of the PFAs were considered safe, while 301 also met the control definition (96.17%). A total of 12 (3.83%) PFAs did not establish control. Of those 12, behaviour of concern was not observed in three PFAs (0.96%), behaviour of concern was observed while remaining safe in five PFAs (1.6%) and four PFAs (1.28%) had behaviour of concern and were terminated due to being unsafe.

Across all conducted PFAs, the average total duration was 21.58 min, ranging from 5.4 min to 62.54 min. The average time spent in EO and SR conditions was 4.40 min (range 4 s to 40.59 min) and 16.99 min (range 4.27 min to 57.51 min), respectively. The average number of EOs was 5.6, with an average duration of 48 s per trial. The average time in SR before the first EO was 4.44 min (range 0 s to 21.49 min), while the remaining total time spent in the SR condition was 12.59 min. The average SR interval between EO conditions was 2.1 min (range 25 s to 8.04 min). An analysis of the duration and conditions across the PFAs is provided in Table 6.

There were 1627 R2 and 73 R1 behaviours across all PFAs, resulting in an average of 5.3 instances of R2 and 0.23 instances of R1 behaviours per PFA. An average of 4.94 R2 behaviours occurred during the EO condition (total of 1,545) and an average of 0.36 R2 behaviours occurred during the SR condition (total of 114) per PFA. For the R1 behaviours, an average of 0.05 instances occurred during the EO condition (total of 16), while an average of 0.19 instances occurred during the SR condition (total of 61) per PFA. Behaviours of concern were observed in the SR condition in 51 PFAs (16.29%). Within those 51 PFAs, 14 had R1 behaviour in the SR condition (4.15%), while 47 PFAs (15.02%) had R2 behaviours in the SR condition. An analysis of the behaviours of concern observed during the PFAs is provided in Table 7.

To assess social validity, a brief survey was sent to the clinicians; 90 responses were collected regarding the PFAs using RP and post-test items. The first of the survey items referred to the relevance and effectiveness of the training and resources, with an average RP score of 4.87 and a post-test score of 8.92. The second of the survey items referred to the comfort and willingness to conduct PFAs due to the training, which had an average RP score of 5 and an average post-test score of 8.52. The third of the items referred to the appropriateness of the support provided to conduct the PFA, which had an average RP score of 5.24 and an average post-test score of 8.26. The fourth item, which rated the safety of the experience of the PFA, had an average post-test score of 9.65. A summary of the clinician social validity surveys are provided in Figure 1.

Implementation independence was collected for 177 PFAs (56.55% of total PFAs). The average score for independent implementation was 86%.

The present study examined whether the PFA could be implemented safely and with clear functional control across a large, geographically distributed ABA service organisation. The findings are consistent with that goal. Across 313 assessments spanning 12 states, 96.17% demonstrated clear functional control and 98.72% met safety criteria, outcomes achieved in an average of 21.59 min across clinic, home and community settings. These rates align with or exceed differentiation rates and duration reported in prior reviews of FA research (92.3% in Melanson and Fahmie, 2023; 95.9% in Hanley et al., 2003) and were achieved under the considerably more variable conditions characteristic of large-scale clinical implementation. Together, these results suggest that the PFA’s core features: synthesised contingency conditions, precursor-contingent termination and an interview-driven EO construction, can be preserved and produce meaningful functional differentiation when delivered at scale by clinicians with varying levels of experience across diverse real-world settings.

A key question for any functional assessment methodology is whether it can be validly and safely administered in the conditions under which clinicians actually work. The results reported here speak directly to that question in the context of large-scale ABA service delivery. The practical constraints of coordinating implementation across multiple states, service settings, coaching modalities and clinicians with varying experience levels represent a considerably more demanding implementation context than the small-scale, closely supervised conditions typical of published PFA research. That high rates of functional differentiation and safety were achieved under these conditions is a meaningful finding for organisations considering whether to invest in building the infrastructure needed to deliver function-based assessment at scale.

The large-scale implementation of the PFA reported here demonstrates that a synthesised contingency approach can be feasibly and safely administered across diverse real-world service delivery contexts. This finding is relevant to ongoing discussions about the conditions under which functional assessment can be practically adopted at scale (Oliver et al., 2015; Roscoe et al., 2015). The synthesised EO and SR conditions that define the PFA were designed specifically to support implementation in applied settings by condensing the assessment into a single alternating design informed by open-ended interviews, reducing the need for extended session time and prioritising client safety through precursor-contingent termination. The present data suggest these design features translate into meaningful feasibility advantages when implemented at an organisational scale, even when accounting for the natural variability in clinician experience, setting and coaching support that characterises large organisations.

The scalability of any evidence-based assessment also depends on the infrastructure built to support it. In this study, a multi-tiered model involving technicians, clinicians and coaches was used to train and support implementation across sites. The average implementation independence score of 86%, reflecting the proportion of assessments completed without active coach guidance, suggests that clinicians were able to internalise and carry out the core PFA procedures following structured training and mentorship. This is particularly notable given that 68.2% of coaching was delivered via telehealth, indicating that remote support can be a viable mechanism for sustaining implementation fidelity in geographically distributed organisations. The average assessment duration of 21.59 min (Table 5) is consistent with prior work on the efficiency of synthesised contingency analyses (Jessel et al., 2016, 2018) and addresses one of the barriers to FA most frequently cited by practitioners: time (Oliver et al., 2015; Saini et al., 2020).

The safety outcomes reported here and trauma-informed approach strongly support the PFA. Of the 313 analyses conducted, only 4 (1.28%; Table 4) were terminated due to safety concerns. The expanded response class definition, which included precursor behaviours, allowed Clinicians to respond to early indicators rather than waiting for behaviours of concern to occur. This is reflected in the low rate of behaviour of concern observed during analyses, with an average of only 0.23 instances per analysis compared to 5.3 instances of precursor behaviour (Table 6). The emphasis on maintaining clients in an enriched synthesised reinforcement condition (average 16.99 min) versus limited exposure to EO conditions (average 4.41 min) aligns with trauma-informed principles by minimising potential re-traumatisation while still gathering necessary functional information (Table 5). It is also worth noting that R1 behaviour was recorded during the SR condition in 14 PFAs (4.15%), occurring across 61 total instances. This pattern likely reflects carryover from preceding EO presentations rather than a failure of the SR condition itself, that is, the client had not yet fully de-escalated at the time the SR condition was reinstated. This interpretation is consistent with the precursor-contingent termination logic of the performance-based PFA: because SR access is provided contingent on the first instance of any response class member, some residual behavioural momentum from the EO is expected before the client fully transitions to an HRE state. Future studies should examine the temporal relationship between EO termination and return to HRE more systematically, as this may have implications for understanding the clinical significance of brief SR-condition behaviour.

Taken together, the safety profile reported here reflects more than a methodological outcome. The precursor-contingent termination procedure operationalises a core assumption of compassionate assessment: that early signs of distress are sufficient reason to stop, regardless of whether escalation has occurred. The extended time clients spent in the SR condition, averaging 16.99 min compared to 4.41 min of test exposure (Table 5), means that the majority of each assessment was structured around access to preferred activities and positive social interaction. This is a dignity-preserving design choice. The training model that made these outcomes possible across 12 states and hundreds of clinicians can itself be understood as compassion infrastructure: a deliberate organisational investment in ensuring that safe, person-centred assessment is not a resource available only in research settings, but a routine standard of care.

The social validity data offer an important complement to the feasibility findings. In applied practice, an assessment’s value depends not only on whether it produces valid results but on whether clinicians find it acceptable and sustainable to use (Wolf, 1978; Slocum et al., 2014). The increase from retrospective pre-training ratings (averaging 4.87–5.24) to post-training ratings (averaging 8.26–8.92) indicates that clinicians perceived the PFA as substantially more accessible and manageable following structured training and supported practice. Clinicians’ ratings of their experience as safe (9.65 out of 10; Figure 1) are particularly meaningful given that safety concerns are among the most commonly cited barriers to conducting functional analyses in applied settings (Oliver et al., 2015). These ratings suggest that the trauma-informed design features of the PFA, including precursor-contingent termination, enriched SR conditions and brief EO exposure, were experienced by implementing clinicians as meaningfully safer than traditional FA formats.

Taken together, the social validity data suggest that the PFA model as implemented in this organisation, with its structured training, tiered coaching support and manualised procedures, was experienced as practically viable by the clinicians who delivered it. This matters for sustainability: assessments that clinicians find burdensome, unsafe or difficult to implement are unlikely to be used consistently, regardless of their scientific merit. The PFA’s alignment with practitioner priorities around safety, efficiency and immediate clinical applicability (Slocum et al., 2014) may be a meaningful factor in its adoption at scale and warrants further examination in future research examining the conditions that sustain implementation over time.

Several limitations should be considered when interpreting these findings. Firstly, interobserver agreement (IOA) data were not collected for any outcome measure, which limits confidence in the reliability of the obtained data. The absence of IOA is a meaningful constraint given the scale of implementation and the number of clinicians involved and future large-scale studies should prioritise systematic reliability checks even when logistically challenging. Secondly, procedural fidelity was not formally scored. While coaches used a structured implementation rubric and their expertise was integral to guiding clinicians, no aggregate fidelity score was computed for the team as a whole. The 86% independent implementation rate reported here reflects clinician independence from active coaching support, but should not be interpreted as a formal measure of procedural fidelity. Thirdly, referral for PFA was based on clinician judgment rather than a standardised screening instrument and the distribution of functional profiles across the sample was not systematically documented. While the referral criteria allowed for multiple contingency types, including automatically maintained functions, the degree to which the sample actually reflects this breadth cannot be confirmed from the data available. Future studies should implement structured pre-referral screening to document functional profile distribution and enable more precise generalisability claims. Fourth, no treatment outcome data were collected following the PFA, which prevents conclusions about the assessment’s treatment utility. While safety and functional control are important indicators of feasibility, the ultimate value of any functional assessment rests on whether it informs effective intervention, a question this study was not designed to answer. Fifth, social validity data were collected retrospectively from clinicians only. The perspectives of caregivers, clients and behaviour technicians were not obtained. This is a values-relevant limitation, not only a methodological one: a study framing its approach as person-centred and dignity-preserving should, in future iterations, centre the voices of the people receiving and directly supporting the assessment. Retrospective clinician self-report is also subject to recall and social desirability biases. Future work should prioritise prospective, participatory data collection with families and, where feasible, with clients themselves.

The findings from this study carry several implications for organisations seeking to implement function-based assessment at scale. The successful delivery of 313 PFAs across 12 states, three service settings and multiple coaching modalities suggests that large-scale dissemination of the PFA is achievable when supported by a structured training and mentorship infrastructure. The Horner and Kittelman (2022) framework for large-scale behavioural implementation identifies five essential conditions for sustained adoption: valued outcomes, use of behavioural science to define core practices, organisational systems that support fidelity, data-based decision-making and ongoing measurement and evaluation. The present implementation effort reflects each of these conditions to varying degrees, valued safety and functional control outcomes were established a priori, core PFA features were operationalised in a manualised rubric, a multi-tiered coaching structure was used to support clinicians across sites and data were collected systematically across assessments. These alignments suggest that the PFA is well-suited to this kind of organisational dissemination and that the Horner and Kittelman framework offers a useful organising structure for future implementation efforts. Importantly, clinicians responsible for applying evidence-based strategies in dynamic real-world environments require assessments that are safe, efficient and immediately actionable (Slocum et al., 2014), all features the PFA is designed to prioritise. The implementation model reported here was developed within a US insurance-funded service context and direct replication in other systems will require adaptation to local funding, credentialing and workforce structures. The transferable elements are the principles: values-aligned selection criteria, tiered coaching with blended delivery and a manualised assessment process that can flex across settings while preserving procedural integrity.

Future research should prioritise several directions to build on what this study accomplished. Most critically, subsequent work should collect treatment outcome data following PFA to evaluate whether assessment results lead to effective function-based interventions, a necessary step for establishing the clinical utility of the PFA at scale. Prospective IOA and formal procedural fidelity data collection should be incorporated from the outset of future large-scale implementations. Research examining whether clinicians continue to use the PFA independently following the conclusion of a formal implementation study, including its application with new clients, would provide a meaningful indicator of sustained adoption and training effectiveness (Horner and Kittelman, 2022). Studies comparing outcomes across training formats (e.g. 2-h vs 10-h) would help identify the minimum training dose needed to support independent, high-fidelity implementation. Examination of how implementation outcomes vary across coaching modalities, service settings and organisational contexts would further inform dissemination practice. Addressing these questions systematically will be essential for advancing the PFA from a promising dissemination example to an evidence-based model for organisational implementation.

The present study documents the feasibility of implementing the PFA across a large, multi-state ABA organisation serving clients in diverse settings. The findings indicate that the PFA can be embedded within routine clinical service delivery at organisational scale without compromising safety or the demonstration of functional control. The multi-tiered coaching model, delivered across in-person, telehealth and on-site observation formats, appears to have supported clinicians in achieving meaningful implementation independence (86%) and the high social validity ratings reported by clinicians suggest the process was experienced as acceptable and useful. Collectively, these results address a meaningful gap in the literature by demonstrating that function-based assessment need not be confined to research settings or small-scale clinical contexts.

What this study contributes is not a claim of efficacy, but a demonstration of reach. Training hundreds of clinicians across a geographically dispersed organisation to conduct structured functional assessments safely, efficiently and with high rates of functional differentiation represents a meaningful step towards closing the persistent gap between behavioural science and applied practice. The infrastructure developed here, including manualised procedures, a tiered coaching model and flexible delivery across settings and modalities, offers a practical starting point for other organisations seeking to integrate function-based assessment into routine care. The training model documented here demonstrates that progressive, values-driven assessment practice is not contingent on exceptional resources or specialist research infrastructure. With deliberate organisational design, it can become the standard. Continued research that adds treatment outcome data, prospective reliability measures and longer-term follow-up will be necessary to determine whether this implementation model supports not just assessment at scale, but effective intervention at scale. That is the next important question and the present findings establish a foundation from which to pursue it.

Hillary Laney is based at Centria Healthcare Autism Services, Farmington Hills, Michigan, USA.

Edward Sanabria is based at Centria Healthcare Autism Services, Farmington Hills, Michigan, USA, and E&E Behavior, LLC, Rocky Hill, Connecticut, USA.

Cindi Kennedy is based at Centria Healthcare Autism Services, Farmington Hills, Michigan, USA.

Cassi A. Breaux is based at University of West Florida, Pensacola, Florida, USA, and Centria Healthcare, Farmington Hills, Michigan, USA.

The authors extend our sincere gratitude to Liisa Podosek and Kalyn Riggs for their invaluable contributions as directors during the implementation of this system. Their exceptional leadership and dedication in coaching residents and clinicians across the organisation were instrumental to the success of this initiative. Their expertise in behavioural assessment and commitment to maintaining high clinical standards while supporting staff development helped facilitate this adoption of the PFA at scale. In addition, the authors thank Timothy Yeager for his advocacy, vision and support of this work.

The authors acknowledge that their affiliation with the organisation may present a potential conflict of interest; however, all authors affirm that the study was conducted with academic integrity and without external influence on the research design, data collection, analysis or interpretation of the findings. The authors declare no additional competing interests related to this work.

No funding was received to assist with the preparation of this manuscript.

This study was approved by the University of West Florida IRB (Approval Number: [2137871-1]). All research procedures involving human participants and/or animals were conducted in accordance with the ethical standards outlined by the committee, as well as the principles of the Declaration of Helsinki.