SOCIAL PROGRAMS · ACCESS TO BENEFITS
A French departmental family-benefits fund was preparing to deploy a unified digital platform, gradually replacing physical service counters and paper forms. Population concerned: 340 000 recipients and potential non-recipients.
The question was not whether users preferred digital services or the service counter. It was which frictions disappeared, which new frictions appeared, and for whom.
THE PROBLEM
The technical system was robust: optimized digital journey, integrated benefits simulator, pre-filling, proactive notification. The policy ambition was twofold: improve administrative efficiency and structurally reduce non-take-up by making the journey easier.
A platform removes a journey, speeds up an application and makes the service available at any time. It also adds other requirements: equipment, connection, credentials, digitized documents, reading, navigation and the ability to resolve an error. The issue is therefore not digital versus physical. It is: which frictions disappear, and which new frictions appear.
Several national studies documented the paradox of digital benefits-access platforms: they improved overall take-up while worsening access for the most vulnerable groups, widening the take-up gap between groups. Management did not want to reproduce this pattern and used our system to test 5 rollout configurations.
WHAT IS ESTABLISHED
WHAT DIFFERS FOR EACH GROUP
ACCESS TO THE SERVICE, ACCESS TO THE BENEFIT
Between an entitlement that is theoretically open and one that is actually received lie awareness of the entitlement, starting the application, identification, supporting documents, understanding the questions and final submission. Technical availability is only one step in this journey.
A MECHANISM: NON-TAKE-UP IS A TRAJECTORY
Conceptual diagram. It represents no measured value: it simply locates the points where an application can stop, from the moment an entitlement is unknown to the moment an error remains unresolved.
Non-take-up is not a stable category of people who simply do not apply. It can be produced by the journey itself, at different stages depending on the situation.
SELECTIVE FRICTION
The same additional step can be negligible for some users and blocking for others. This asymmetry produces the documented paradox: overall take-up rises while access for the most vulnerable groups deteriorates.
WHAT IS TESTED
The five configurations are not distinguished by their interface. They differ in what surrounds the platform: when support is deployed, which local support networks are mobilized, which non-digital channels are preserved, and how proactive the contact is.
3 800 synthetic recipients were interviewed across the 5 rollout configurations, with dynamic follow-ups on moments of non-completion: abandoning the digital application, not knowing about the entitlement, difficulty finding support. Trajectories were projected over 24 months, modeling benefits-access dynamics by profile and territory.
SIMULATED POPULATION
The reconstructed population covers 340 000 recipients and potential non-recipients in the department, calibrated using the fund's proprietary data — recipient typologies, interaction history, vulnerability indicators — and national studies on how people in France engage with digital administrative services.
It is structured into 14 profiles combining age, relationship with digital services, level of social vulnerability, residential geography and prior exposure to benefits-access programs.
AGE
engagement with digital administrative procedures is not distributed uniformly across generations, although it partly depends on them
ENGAGEMENT WITH DIGITAL ADMINISTRATION
using a phone every day and completing an administrative procedure online are not the same skill
LEVEL OF SOCIAL VULNERABILITY
documented vulnerabilities more often accumulate than occur in isolation
RESIDENTIAL GEOGRAPHY
6 areas of concentrated vulnerability were identified in the department: geography is a structural decision parameter
PREVIOUS EXPOSURE TO SUPPORT PROGRAMS
having previously received a benefit, or previously failed to obtain it, does not create the same willingness to try again
HISTORY OF INTERACTIONS WITH THE FUND
the population is calibrated to the fund's recipient typologies and proprietary vulnerability indicators
POTENTIAL NON-RECIPIENTS
the simulated population includes eligible people who appear in no service-usage data
ACCESS TO LOCAL SUPPORT NETWORKS
town halls, associations, social workers, digital mediators: support is not equally available everywhere
These configurations reveal part of the population's heterogeneity. The simulation operates on synthetic individuals, not a handful of persona types : these are not representative figures, but distinct situations, constraints and options.
REACTIONS
THE TAKE-UP GAP WIDENS DURING THE FIRST TWELVE MONTHS
During the first 12 months of rollout, take-up among digitally self-sufficient groups rises rapidly — +18 pts — while take-up among vulnerable groups stagnates or declines. The gap widens during this window, independently of the platform's intrinsic quality. Dedicated support deployed at the same time as launch is the only mechanism that prevents it.
WHEN SUPPORT IS DEPLOYED MATTERS MORE THAN WHETHER IT EXISTS
Support deployed after launch, in response to observed difficulties, closes 34 % of the widening take-up gap. The same support deployed 6 months before launch, with upstream training for vulnerable groups and identification of local support networks — town halls, associations, social workers, digital mediators — prevents 78 % of it. Sequencing is structurally more decisive than volume.
THE GEOGRAPHY OF VULNERABILITY IS A DECISION PARAMETER
Uniform closure of physical service counters across the department widens the take-up gap by 22 pts over 24 months. Differentiated closure, preserving counters in the 6 identified areas of concentrated vulnerability, limits the widening to 4 pts. The non-digital channel does not serve the same function everywhere.
THE PLATFORM'S TECHNICAL QUALITY IS NOT ENOUGH
The technical system was robust: optimized digital journey, integrated benefits simulator, pre-filling, proactive notification. Yet national studies documented the same paradox in other rollouts: rising overall take-up alongside deteriorating access for the most vulnerable groups. The support architecture, not the interface, separates the two outcomes.
+18 pts in take-up among self-sufficient groups while vulnerable groups stagnate; 34 % of the widening recovered reactively versus 78 % prevented upstream; 22 pts of widening with uniform closures versus 4 pts with differentiated closures. It is the same platform, the same population, the same department.
OPTIMIZE THE AVERAGE
The diagram reproduces the dynamic documented in the case: during the first 12 months, take-up among digitally self-sufficient groups rises by 18 points, while take-up among vulnerable groups stagnates or declines.
A rising average performance is therefore not an indicator of equitable access. The two movements can coexist and cancel each other out in an average.
WHEN SUPPORT IS DEPLOYED
Support deployed after launch recovers 34 % of the widening take-up gap. The same support deployed 6 months beforehand prevents 78 % of it.
Upstream sequencing requires reversing the conventional rollout schedule: train local support networks and identify the groups that need support before the platform is even launched.
RECOVERY AFTER FRICTION
The robustness of a system is revealed by what it does when the standard journey stops working. Many users do not need full support: they need intervention at the right moment, then they continue on their own.
WITHOUT A SAFETY NET
THE SAME CHAIN, REINFORCED
THE ABSENT ARE A POPULATION
A service observes its users. Non-take-up, however, largely happens outside its data. That is why the simulated population includes potential non-recipients.
THOSE WHO NEVER START
A person who does not know they are entitled, or who never starts the application, leaves no trace in the service's data.
THOSE WHO DROP OUT EARLY
Dropping out before identification can remain invisible: the system mainly observes those who stay in the journey.
THE RISK FOR THE DECISION
Optimizing the journey for users who complete it improves measured performance without reducing non-take-up.
WHAT THE SYNTHETIC POPULATION ADDS
The reconstructed population covers 340 000 recipients and potential non-recipients: it includes people who would appear in no usage data.
COMPARISON
Qualitative comparison derived from the simulated configurations. The levels reflect the documented differences: 78 % of the widening prevented by support deployed 6 months before launch, versus 34 % recovered reactively; 4 pts of widening with differentiated closures, versus 22 pts with uniform closure. No configuration is cost-free: the most robust is also the most demanding to orchestrate.
THE MOST ROBUST
Support 6 months before launch trained local support networks service counters preserved in areas of concentrated vulnerability proactive contact at first use
THE MOST FRAGILE
Platform only uniform closure of service counters, on the same schedule
DECISION
PROJECTION, THEN OBSERVATION
The fund selected the strategy combining deployment of the unified digital platform on the original schedule, dedicated support for vulnerable groups operational 6 months before launch — upstream training, identification and training of 240 local support contacts: social workers, community digital mediators, town-hall staff, coordinators of retirees' associations —, preservation of the 6 physical service counters located in areas of concentrated vulnerability, and proactive outreach to vulnerable groups for support at first use.
At 18 months after rollout, the overall benefit take-up rate is up 14 pts compared with the previous period. Take-up among vulnerable groups is up 8 pts, versus a sector projection of −4 pts under a standard rollout. The take-up gap between digitally self-sufficient groups and vulnerable groups narrowed by 6 pts instead of widening.
The methodology has been referenced by France's national family-benefits fund and is being rolled out in several other departmental funds. These figures are those documented by the client: they describe what was observed after the decision, without establishing exclusive causality between the system and each of these changes.
TAKEAWAY
Without dedicated support for vulnerable groups, the take-up gap structurally widens during the first 12 months, regardless of the platform's intrinsic quality. Digital technology does not mechanically produce non-take-up: a rollout architecture does when it leaves drop-offs without a response, and without a trace.
The second lesson concerns timing. Support deployed before launch prevents the widening more effectively than a system deployed in response to observed difficulties: 78 % versus 34 %. This principle requires reversing the conventional rollout timetable. It applies to family-benefits funds, but also to other digital administrative rollouts: taxes, pensions, healthcare, education and social housing.
POSSIBLE FUTURES
A
DIGITAL FIRST
Unified platform deployed on the original schedule, primarily self-service journey, uniform closure of service counters
B
REACTIVE SUPPORT
Same platform, support system deployed after launch in response to observed difficulties
C
PLATFORM AND HUMAN SAFETY NET
Platform on the original schedule, operational support 6 months beforehand, trained local support networks, counters preserved in areas of concentrated vulnerability, proactive contact at first use
METHOD
340 000 reconstructed recipients and potential non-recipients, 14 profiles combining age, relationship with digital services, level of social vulnerability, residential geography and prior exposure to benefits-access programs, 3 800 synthetic recipients interviewed across 5 rollout configurations, dynamic follow-ups at moments of non-completion, and a 24-month projection modeling benefits-access dynamics by profile and territory.
This case adds a capability to the library: simulating the architecture of a public service as an access system, where the same friction does not have the same effect in different situations, and where the populations that drop out earliest are precisely those that would never appear in the service's usage data.
This case comes from a simulation carried out for a French departmental family-benefits fund. The client is not named, the territory cannot be identified, no personal data entered the system, and the detailed results remain the client's property. The comparisons between configurations published here are qualitative; the quantitative figures cited are those documented by the client.
The simulation does not score any applicant, decide who should receive a benefit, identify any vulnerable person by name, or automate any individual administrative decision. It concerns the service architecture and its consequences at the scale of synthetic populations, upstream of the decision.
Your next decision
Describe your need. We can point you to the right level of support.
State your decision. See the future it produces.
