Scaling is rarely defeated by bad intent; it is defeated by âreasonable variationâ that accumulates until the model no longer exists. When new sites interpret workflows differently, partners apply inconsistent thresholds, and supervision becomes uneven, outcomes start to move before leaders notice. This article sits within Scaling What Works and connects directly to payer and commissioner accountability in Integrated Funding Pilots, focusing on the governance and implementation controls that keep a model deliverable at system scale.
Why fidelity is the real scaling challenge
âFidelityâ is not about forcing identical practice everywhere. It is about protecting the handful of model elements that create valueâtimely risk identification, reliable follow-up, safe escalation, and accountable decision-making. Most scaled programs fail because leaders scale activity (more visits, more staff, more sites) without scaling the operating system (decision rules, supervision routines, audit trails, and corrective action).
Commissioners and oversight bodies will ask two questions that a scaled model must answer in plain operational terms: (1) How do you know the model is being delivered as designed in every location? (2) What do you do, specifically, when it is not?
System expectations leaders must meet
Expectation 1: Documented controls that demonstrate fidelity, not just intention
Funders expect providers to define ânon-negotiablesâ (critical elements) and show how they are measured. This is not a narrative requirement; it is an assurance requirement. If the model claims to reduce avoidable acute use through early escalation, leaders must show the workflow steps, the threshold rules, and the evidence that those steps occur consistently.
Expectation 2: A corrective-action pathway that is timely and auditable
Oversight is not satisfied by âwe provide training.â Scaled services must show how variance is detected, triaged, corrected, and re-checked. Commissioners want to see that corrective action is proportionate (supportive when possible, formal when needed) and that it produces measurable improvement.
Defining what must not drift
Before scaling, leaders should identify the smallest set of critical elements that explain outcomes. These become the âfidelity spineâ of the model: decision thresholds, response times, minimum contact cadence, documentation requirements for risk and restrictive practices, escalation routes, and supervision frequency. Everything else can adapt to local context, but the spine cannot.
A useful way to define the spine is to write it as a set of âif/thenâ rules that can be observed: if risk tier changes, then escalation occurs within a defined time; if a visit is missed, then a same-day safety check occurs; if a high-risk decision is made, then a supervisor sign-off is required.
Operational example 1: Fidelity checklists embedded into daily supervision
What happens in day-to-day delivery: Each team uses a short fidelity checklist during daily huddles and end-of-day supervisor review. The checklist focuses on a small number of critical elements (for example: risk tier assigned for all new intakes; escalation completed within threshold; follow-up contact completed after escalation; supervisor sign-off for high-risk decisions). Supervisors sample a defined number of cases each day, record findings in a simple log, and assign corrective actions (coaching, shadowing, refresher training, or workflow adjustment). Findings are aggregated weekly and reviewed at program level.
Why the practice exists (failure mode it addresses): When scaling, leaders often rely on periodic audits that are too slow. Drift can persist for weeks before being detected, making it harder to correct and more likely to become normalized.
What goes wrong if it is absent: Teams develop local âworkaroundsâ that feel efficient but bypass key controls. Variance becomes invisible until outcomes deteriorate, at which point it is unclear whether the problem is the model, the workforce, or local practice changes.
What observable outcome it produces: Drift is detected early and corrected quickly. Leaders can show an audit trail that critical elements are being delivered, and can demonstrate improvement after corrective action through reduced variance rates over time.
Operational example 2: A structured variance-to-correction workflow with escalation thresholds
What happens in day-to-day delivery: Variance is categorized into tiers (for example: Tier 1 documentation gaps; Tier 2 missed follow-up; Tier 3 repeated missed escalation; Tier 4 safety incidents or restrictive-practice concerns). Each tier has a required response: Tier 1 triggers coaching and re-check within a week; Tier 2 triggers immediate workflow review and supervisor observation; Tier 3 triggers formal improvement plan with leadership oversight; Tier 4 triggers incident review, safeguarding pathways where relevant, and system partner notification where required by contract. The variance log records the issue, the corrective action, the date completed, and the re-check result.
Why the practice exists (failure mode it addresses): Without a defined pathway, corrective action is inconsistent. Some teams are over-managed while others are ignored, and variance becomes a personnel problem rather than a system problem.
What goes wrong if it is absent: Problems repeat, staff lose confidence, and commissioners see unmanaged risk. The organization cannot prove that it responds proportionately to issues or that learning is applied across sites.
What observable outcome it produces: Faster resolution of recurring issues, fewer repeated variances, and clearer accountability. Commissioners can see a defensible response to risk, including evidence that corrective actions were completed and effective.
Operational example 3: Protecting partner interfaces through shared decision rules
What happens in day-to-day delivery: As scaling expands across partners (primary care, behavioral health, EMS, housing, or county services), leaders publish a shared set of interface rules: referral criteria, escalation thresholds, response times, and handoff documentation requirements. These are trained jointly, reinforced through quarterly partner huddles, and monitored through âhandoff auditsâ that check whether referrals include required information and whether partner responses meet threshold expectations. Where partner performance creates risk (for example, delayed acceptance of high-risk referrals), the service triggers an interface escalation route with named contacts and a documented resolution timeline.
Why the practice exists (failure mode it addresses): Even if the provider delivers perfectly, outcomes can collapse if partners apply inconsistent thresholds or if handoffs are unreliable. Scaling increases the number of handoffs and therefore the risk surface.
What goes wrong if it is absent: Referrals arrive incomplete, high-risk cases bounce between services, and escalation fails. The model is blamed for poor outcomes that actually arise from interface breakdowns.
What observable outcome it produces: More reliable handoffs, fewer âlost referrals,â improved timeliness for high-risk responses, and clearer shared accountability. Interface audit results provide evidence that scaling did not weaken cross-system reliability.
Making governance usable, not bureaucratic
Good governance is light enough to run every day and strong enough to withstand scrutiny. Leaders who scale successfully define a small fidelity spine, measure it frequently, and respond predictably to variance. That approach preserves outcomes while still allowing local adaptation where it does not compromise safety, rights, or system reliability.