A Swiss study tracked Individual Placement and Support participants for six years. During the first two years, IPS participants had significantly higher employment rates than controls, 40% versus 28%. By year six, the difference had disappeared. Both groups hovered around 33-36%, statistically indistinguishable. 1
This finding is routinely cited as evidence that IPS “doesn’t work,” or at least that its effects “don’t last.” The implication is damning: what’s the point of an intervention whose benefits dissipate?
But consider what the finding actually shows.
Over six years, people’s circumstances changed. They faced new health episodes, new caring responsibilities, new redundancies, new barriers. They needed support again. It wasn’t there. The IPS group’s advantage disappeared not because the original intervention failed, but most likely because there was no ongoing infrastructure to maintain it.
We built infrastructure because health needs are permanent.
The “effects dissipate” critique assumes career support should produce permanent transformation from a time-limited intervention. It assumes we’re building unicorns.
Career support is infrastructure. Infrastructure requires different evaluation.
The unicorn assumption
There’s an assumption baked into how we evaluate employment programmes, and it goes something like this: intervention produces transformation; transformation is permanent; person never needs support again. Success means the individual is “fixed.” Return visits indicate failure, that either the intervention didn’t work, or the person is somehow deficient.
This theory is false for almost everything that matters.
Education doesn’t prevent people needing to learn new things. Physiotherapy doesn’t prevent future injuries. Debt advice doesn’t prevent future financial difficulty. Mental health treatment doesn’t prevent future episodes. We don’t evaluate these services by whether recipients never need help again. We evaluate whether they delivered the right support at the time.
Career support is the only service evaluated this way.
Get someone a job. Measure whether they’re still in it at 26 weeks. Judge the intervention by that outcome. If they return in two years needing help again, that’s a new episode to be funded separately, not evidence that the first intervention failed. Returning is a normal feature of any career.
The absurdity becomes clearer with scale. If we evaluated GPs by whether patients ever returned, every surgery in the country would be deemed failing. The metric would tell us nothing about quality of care. It would just confirm that health needs recur, something we already knew, and built a system around.
Career needs recur too. Jobs end. Industries decline. Health changes. Caring responsibilities shift. Technology transforms roles. The average person will have multiple significant career transitions across a working life.
A system that counts each return as failure is measuring its own design flaw, not the intervention’s effectiveness.
What we already know
The evidence base for effective career support isn’t thin or contested. Individual Placement and Support has been validated across 28 randomised controlled trials in multiple countries and populations. The effect is consistent: roughly double the employment rate of traditional approaches. 2
A fair challenge is that IPS was developed for people with severe mental illness. The evidence base is overwhelmingly from mental health contexts. Why assume it transfers to everyone navigating career transitions?
Because the principles that make IPS effective aren’t about mental health specifically. They’re about how support works when people face complex barriers.
Sustained relationship matters because trust takes time to build and circumstances take time to understand, and that’s true whether someone is managing schizophrenia or recovering from redundancy in a declining industry.
Integration with other services matters because employment barriers rarely exist in isolation, and that’s true whether the connected service is a community mental health team or a housing association or a benefits adviser.
Rapid employer contact matters because the labour market teaches more than preparation programmes, and that’s true for anyone.
Zero exclusion matters because the people screened out as “not ready” are precisely those who need sustained support most, true regardless of why they’re struggling.
Time-unlimited availability matters because progress isn’t linear and circumstances change, true for everyone.
The mental health evidence base exists because that’s where researchers tested the model rigorously.
IPS has since been adapted for other populations (young people not in employment, education or training, and people with substance use histories) with strong results for the former and promising but mixed results for the latter, depending on the substance involved.
Career support outside clinical settings doesn't need to replicate IPS exactly, but it should be designed around the same spine: sustained relationship, integration with relevant services, no exclusion, support for as long as it's needed. The services it integrates with may differ, whether housing, benefits, sometimes health, often just the practical realities of someone's life, but the architecture can be the same.
This is how medicine operates. Penicillin isn’t re-tested every time it’s prescribed. The evidence was established decades ago. What gets measured now is whether it’s prescribed appropriately. Clinical guidelines encode the evidence. Audit measures adherence to guidelines.
We don’t need more RCTs to know IPS works. We need to deliver it properly and consistently.
The IPS Fidelity Scale, adapted for UK use by the Centre for Mental Health measures whether services deliver the evidence-based model.3
Caseload size. Are workers supporting few enough people to build genuine relationships? 4
Integration with clinical teams. Is employment support embedded in mental health services, or operating in isolation?
Time to first employer contact. Do people meet employers within 30 days, or spend months on “job-readiness” preparation first?
Zero-exclusion policies. Is anyone turned away as “not job-ready”?
Duration. Does support continue as long as needed, or terminate after arbitrary periods?
Programmes with higher fidelity scores produce better employment outcomes, a finding replicated across multiple studies in different countries.5
This means we can measure what matters, whether the service is being delivered properly, rather than what we can’t control: whether the labour market, the economy, the individual’s health, and countless other factors aligned to produce employment.
Why outcome measurement fails
The dysfunction is predictable and well-documented.
It measures the wrong things. The primary DWP metric, “off-flow rate,” captures whether someone stopped claiming benefits. Not whether they found work. Not whether that work was good. Not whether their circumstances improved.
Someone sanctioned into destitution counts the same as someone who found a fulfilling job.
It incentivises gaming. Providers optimise for measured outcomes, not unmeasured quality. “Sustained employment at 26 weeks” incentivises placing people in any job that will last six months, regardless of suitability, progression prospects, or what the person actually wants.
Creaming (prioritising job-ready clients) and parking (neglecting complex cases) are rational responses to outcome-based payment. The Work Programme demonstrated this extensively as discussed in Part 3 of this series.
It punishes providers for things beyond their control. Employment outcomes depend on labour market conditions, employer behaviour, individual health, family circumstances, housing stability, transport access, discrimination, and luck. A provider operating in a buoyant labour market will outperform one in a depressed region, regardless of service quality. Attributing outcomes to the intervention treats correlation as causation.
Meanwhile, providers can control whether they deliver evidence-based practice i.e. caseload sizes, integration with other services, time to employer contact, exclusion policies. Outcome measurement ignores what providers can actually influence while punishing them for what they can’t.
It consumes resources that could deliver services. The measurement apparatus is expensive. Staff hours go to data entry, compliance reporting, audit preparation. Management attention goes to performance dashboards and provider ratings. Evaluators produce reports confirming what previous reports confirmed.
None of this helps anyone find work. It performs accountability without producing it.
What fidelity measurement would look like
Fidelity-based evaluation asks a different question. Not “did recipients get jobs?” but “did the service deliver evidence-based practice?”
For IPS-style career support, this means measuring things that are actually within a provider’s control.
Caseload. Are advisers supporting small enough numbers to build genuine relationships? IPS evidence sets this at 20-25 clients. Caseload size is measurable, within provider control, and directly linked to effectiveness. Advisers know people’s circumstances, their barriers, what’s changed since last time. You can’t do that with 100 people on your list.
Integration. Is career support connected to health services, housing support, benefits advice? Or does it operate in isolation, unable to address the actual barriers people face? Integration can be assessed through referral patterns, co-location, joint working arrangements.
Speed. How quickly do people meet employers? IPS emphasises rapid contact, within weeks, not months. Time to employer contact is measurable and predictive of outcomes.
Inclusion. Is anyone screened out as “too difficult” or “not ready”? Zero-exclusion is a core IPS principle. Services can be assessed on whether they turn people away and on what basis.
Duration. Does support continue as long as needed, or terminate after a fixed period regardless of circumstances? This matters enormously. I’ve seen people make real progress over months of engagement, only for the funding to run out just when things were starting to shift.
Practice. Are advisers trained in evidence-based approaches? Do they receive supervision? Is there a quality assurance process?
This approach has precedents. Healthcare regulates quality through clinical standards, professional registration, and practice audit, not by measuring whether patients recover, which depends on disease severity, patient behaviour, comorbidities, and luck. The Care Quality Commission doesn’t rate hospitals by whether patients ever get ill again. Ofsted doesn’t rate schools by whether students never need education again.
Schools are inspected on teaching quality, safeguarding, and curriculum delivery, not solely on student outcomes, which depend on intake, deprivation, parental engagement, and factors beyond school control.
Fidelity-based accountability for career support would apply the same logic: measure whether the service meets quality standards, not whether factors beyond its control produce desired outcomes.
The objections
Fidelity-based evaluation isn’t a perfect solution. There are legitimate concerns.
High fidelity doesn’t guarantee good outcomes. Some high-fidelity services still underperform, but fidelity measurement is more actionable. If a high-fidelity service underperforms, we learn something about the model itself. If outcome measurement shows failure, we can’t tell whether the model or the implementation was at fault.
Fidelity measurement can become box-ticking. Any measurement system can be gamed. Providers can document integration meetings that don’t produce genuine collaboration. They can report caseload figures that don’t reflect actual practice.
But fidelity gaming is more observable than outcome gaming. Auditors can visit services, observe practice, interview clients. Outcome data sits in spreadsheets, disconnected from reality.
Outcomes matter. No one wants high-fidelity services producing zero results, but there’s a difference between outcome monitoring at system level and outcome-based accountability at provider level.
We can track aggregate employment rates for populations receiving support, as a health check on whether the model still works, without attributing individual outcomes to specific interventions. Population health surveillance monitors disease patterns without blaming individual hospitals for patient deaths.
The medical analogy has limits. Penicillin has clear causal mechanisms. Career support operates through complex, contextual, relational processes. Social interventions may require more adaptation to local circumstances than clinical protocols allow.
But this argues for measuring “fidelity to principles” rather than “fidelity to rigid protocols.” The IPS evidence identifies principles that work: sustained relationship, integration, and rapid employer contact. Not scripts to follow mechanically.
What this means in practice
The shift from outcomes to fidelity isn’t abandoning accountability. It’s redirecting accountability toward what services can control.
Outcome measurement asks: “Did this person get a job?” It attributes the answer to the service, regardless of labour market conditions, employer behaviour, individual circumstances, and everything else that determines employment.
Fidelity measurement asks: “Did this service deliver evidence-based support?” It holds providers accountable for their actual practice, while recognising that outcomes depend on factors beyond their control.
Career adaptability theory offers a useful lens here. Savickas identifies four resources people draw on when managing career transitions: concern about their future, a sense of control over their choices, curiosity about possibilities, and confidence in their ability to act. These are things a career support service can directly build through sustained relationships and good practice. Whether employers are hiring is not.
Measuring adaptability captures what the service actually did. Not what the labour market happened to allow.
Intermediate outcomes like career adaptability, self-efficacy, and confidence fit this framework as diagnostic tools rather than success criteria. These are established measures in the career development literature, tracking whether support is building the foundations for employment rather than demanding immediate job entry as the only evidence of effect. 6
And when someone comes back for support a second time? Under fidelity measurement, that’s not failure. That’s the system working, with people accessing support when circumstances change, as they inevitably will across a working life.
The question isn’t whether someone was permanently transformed. It’s whether effective support was available when they needed it, delivered according to evidence-based principles, for as long as their circumstances required.
The system needs it this way
Outcome-based evaluation persists not because it works but because it serves functions unrelated to quality.
It provides political cover. Ministers can announce programmes with outcome targets, claim credit if targets are met, blame providers if they aren’t. The measurement apparatus creates the appearance of accountability regardless of whether anyone is actually helped.
It justifies programme churn. Poor outcomes become evidence that the current programme “isn’t working,” justifying replacement with a new programme, which will face the same outcome measurement, produce similar results, and eventually be replaced in turn. The measurement framework sustains the cycle.
It avoids difficult questions. If we measured fidelity, we’d have to confront why services don’t deliver evidence-based practice. The answers are uncomfortable: inadequate funding, impossible caseloads, sanctions contamination, political interference. Outcome measurement lets us blame providers for “poor performance” without examining the conditions that make good performance impossible.
Shifting to fidelity-based evaluation would require accepting some uncomfortable things.
That we already know what works. That the task is implementation, not discovery. That outcomes depend on factors beyond what any employment service can control. That we need to measure services on their actual practice, not on results they can’t determine. That people will need support more than once, and that’s not failure, it’s a feature of working life.
We know what effective career support looks like. The evidence is clear. The fidelity tools exist.
What we lack is the commitment to deliver it properly, and to measure whether we are, rather than whether forces beyond our control produce the outcomes we want.
* * *
Next week: what this means for AI, and how to tell whether new technology extends what works or undermines it.
Note: I have used Perplexity to help with researching this series.
Pichler et al., ‘Long-Term Effects of the Individual Placement and Support Intervention on Employment Status: 6-Year Follow-Up of a Randomized Controlled Trial’, Frontiers in Psychiatry, 2021. https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2021.709732/full
IPS Employment Center evidence database. https://ipsworks.org/index.php/evidence-for-ips/
Centre for Mental Health / IPS Grow, IPS Fidelity Scale for Connect to Work (CtW), January 2026. The 25-item scale, adapted for UK use from the Dartmouth IPS Fidelity Scale, measures caseload size, integration with support teams, speed to employer contact, zero-exclusion policies, and other key delivery elements. Used to assess whether Connect to Work providers are delivering evidence-based IPS. https://www.centreformentalhealth.org.uk/wp-content/uploads/2026/01/IPS-fidelity-scale-for-Connect-to-Work-CtW-1.pdf
Bond, G.R., Becker, D.R. & Drake, R.E. (2011), ‘Measurement of Fidelity of Implementation of Evidence-Based Practices: Case Example of the IPS Fidelity Scale’ (IPS caseload standard of 20-25 clients). https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1468-2850.2011.01244.x
NAO, Supporting People to Work Through Jobcentres, March 2025, para. 13: “DWP has funded jobcentres to operate with a caseload per work coach of around 100 cases in the Intensive Work Search category.” https://www.nao.org.uk/press-releases/work-coach-shortage-leads-dwp-to-reduce-support-for-universal-credit-claimants/
Drake, R.E., Bond, G.R., Becker, D.R. & Noel, V.A., ‘Individual Placement and Support: History, Current Status, and Future Directions’, World Psychiatry, vol. 22, no. 2, 2023, pp. 272–282. https://pmc.ncbi.nlm.nih.gov/articles/PMC11114326/
Bond, G.R., Peterson, A.E., Becker, D.R. & Drake, R.E., 'Validation of the Revised Individual Placement and Support Fidelity Scale (IPS-25)', Psychiatric Services, vol. 63, no. 8, 2012, pp. 758–763. https://pubmed.ncbi.nlm.nih.gov/22660842/;
Kim, S.J., Bond, G.R., Becker, D.R., Swanson, S.J. & Langfitt-Reese, S., 'Predictive Validity of the Individual Placement and Support Fidelity Scale (IPS-25): A Replication Study', Journal of Vocational Rehabilitation, vol. 43, no. 3, 2015, pp. 305–317. https://journals.sagepub.com/doi/full/10.3233/JVR-150770
Savickas, M.L. and Porfeli, E.J., 'Career Adapt-Abilities Scale: Construction, Reliability, and Measurement Equivalence across 13 Countries', Journal of Vocational Behavior, vol. 80, no. 3, 2012, pp. 661–673. https://www.sciencedirect.com/science/article/abs/pii/S0001879112000139. The scale measures four psychosocial resources for managing occupational transitions: concern, control, curiosity, and confidence. One of the most cited instruments in career development research.