The WorkWell Playbook

Executive summary

In November 2026, WorkWell stops being a 15-site pilot and becomes a national service. Every integrated care system in England will need a working answer to the same question the pilots spent eighteen months and 64 million pounds answering: how do you build a service that takes people with health conditions and gets them into, back into, or safely staying in work?

The good news is that the answer now exists, in data. Roughly 34,000 people started a WorkWell plan during the pilot. We have studied the national management information, the DWP's own evaluation, and, uniquely, the day-to-day operational records of four of the fifteen pilot systems: every referral, every caseload, every onward referral, every DWP outcome return. This paper distils that into seven decisions a new WorkWell lead has to get right, with real benchmarks for each.

Our point of view, compressed:

1. 1

   Primary care is your demand engine. Nationally, GPs account for 29% of starts. In systems where the front door sits inside primary care, we observe 68%. Your referral mix is a design choice, not a given.

2. 2

   Choose a front door plus embedded delivery. The best-performing shape we observed pairs a small triage function with coaches embedded in primary care neighbourhoods. Pure hub-and-spoke travels well in sparse geographies; all-hands intake does not survive scale.

3. 3

   Staff for mental health, size for flow. Half your caseload will present with a mental health barrier. A settled coach absorbs 10 to 13 new participants a month and carries 25 to 35 open cases. Do that arithmetic before you sign the staffing plan.

4. 4

   Be a backbone, not an empire. The pilots that work do not try to deliver everything. They hold the relationship and route: one anchor employment partner per area, a curated long tail of community services, and clean handovers.

5. 5

   Take the in-work half seriously. 43% of national starts are in work at first appointment. In our data the single biggest employer sector is health and social work. Your own system's workforce will be one of your largest client groups.

6. 6

   Make the DWP dataset a by-product of coaching. Where the DWP return lives inside the case record, outcome completeness runs at effectively 100%. Where measurement is a separate task, completion collapses to around 19%. This is the difference between commissioning evidence and commissioning anecdotes.

7. 7

   Mobilise in 90 days, data governance first. The single biggest launch delay in the pilot cohort was not recruitment or premises. It was data sharing agreements.

The rest of this paper takes each decision in turn: the evidence, the trade-offs, and what we would do.

1. You have less time than the pilots had

WorkWell was announced at Spring Budget 2023 and funded with 64 million pounds across 15 integrated care boards. Sites were expected to launch in October 2024. Most could not: the evaluation records tight mobilisation windows, procurement friction, and in one case a launch delayed by months over data governance alone. The sites that mobilised fastest had two things in common: pre-existing local partnerships and prior experience running work and health programmes.

Now the timeline inverts. The pilots had eighteen months of grace to find their operating model. The national rollout in November 2026 gives new systems a fraction of that, but with one enormous compensating advantage: the pilots already made the mistakes. The ramp curve is known. Nationally, starts grew from 330 in October 2024 to a peak of 2,800 a month in October 2025, then settled at around 2,300 a month. A single ICB-scale service should expect a similar shape locally: a slow first quarter, a steep second and third, then steady state.

2. Know the shape of demand before you design the service

The national picture, from DWP management information covering 34,210 starts to March 2026:

Our four-system operational dataset (7,500+ participant records) sharpens this. Mental health is not 47% but 53% once you combine depression and anxiety (44%) with other mental illness (9%). The modal WorkWell participant is referred by their GP, is as likely to be in work as out of it, and names depression or anxiety as the thing standing between them and work.

3. Decision 1: route your demand deliberately

Nationally, referral routes split GP or primary care 29%, self-referral 24%, Jobcentre Plus 22%, voluntary sector 9%. But that average hides a design choice. In the systems we observe, where the WorkWell front door is embedded in primary care and on the GP's screen at the point of the fit note conversation, GP referrals run at 68% and Jobcentre Plus at 5%.

Neither mix is wrong. They are different services. A GP-fed service skews in-work and early-intervention: people still attached to a job, often on or near sickness absence, where a fast, light-touch intervention protects the employment relationship. A Jobcentre-fed service skews long-term unemployed, which the evaluation warns pulls WorkWell beyond its light-touch design intent.

Three practical rules:

• Decide your mix, then build the pathway that produces it. If you want early intervention, invest in the GP integration: referral at the point of care, not a leaflet. If GPs must leave their clinical system, log into a portal, or fill a PDF, you will not see 68%.
• Formalise the Jobcentre pathway anyway. The evaluation's first recommendation. Volume will come; make sure it arrives triaged.
• Do not build year one around employer referrals. Employers account for 2% of starts everywhere. The employer channel matters (see Decision 5) but as a destination and a partner, not a referral source.

4. Decision 2: pick an operating model that fits your geography

Across the systems we observe, three viable shapes emerge, plus one anti-pattern.

Model A: neighbourhood-embedded teams. Coaching teams of 3 to 8 sit inside primary care localities, holding their own patch and its GP relationships. Strengths: the 68% GP referral engine, local partner knowledge, continuity. Costs: uneven demand between patches, duplicated management overhead. Best for: dense, well-bounded urban and mixed geographies.

Model B: hub-and-spoke with a roving pool. A central team (one system we observe runs a 13-person pool) deploys coaches across the footprint as demand dictates. Strengths: elastic capacity, single management line, fast to stand up. Costs: weaker local relationships, travel time in rural areas, coaches feel less ownership. Best for: sparse or polycentric geographies, and for launch phase anywhere.

Model C: single front door with specialist routing. A small triage function receives everything, resolves the light cases itself, and routes the rest to the right coach or partner. One area we observe runs this with a dedicated intake team; its neighbouring area lets every coach take intake. The front-door version produces cleaner caseloads and better data; the all-hands version survives only at small scale and degrades exactly when volume arrives.

Our view: hybridise A and C. A single front door for triage and data capture, embedded neighbourhood teams for delivery, and a small roving reserve you can point at surges. Start closer to B if you must launch fast, then devolve into neighbourhoods as volumes stabilise. Whatever you choose, choose the front door first: it is where your data discipline (Decision 6) is won or lost.

5. Decision 3: staffing and caseload arithmetic

The numbers that matter, from settled-state operations across the systems we observe:

The worked example: an ICB footprint expecting 300 referrals a month at steady state needs roughly 23 to 30 coaches in post before the ramp peaks, plus triage and management. Hire behind the curve and open caseloads blow through 35, durations stretch, and the light-touch promise quietly dies.

Two workforce warnings from the pilot evidence:

• Hire for psychological confidence. With half the caseload presenting mental health barriers, a coach whose toolkit is CVs and job boards will drown. The evaluation specifically flags training gaps around complex and neurodivergent needs. Budget real clinical supervision, not just line management.
• Fixed-term contracts are a retention trap. The evaluation records pilots losing staff to competing programmes offering permanent roles. The national rollout makes work and health coaching a durable labour market; act like it and recruit permanently where you can.

6. Decision 4: be a backbone, not an empire

The DWP evaluation lands on a phrase worth framing: the best pilots do not do everything, they "lubricate" the wider system, getting people to the right places with clear handovers. Our operational data shows what that looks like in practice. Across the systems we observe, WorkWell staff made over 1,400 onward referrals, and the pattern is consistent:

• One anchor employment partner per area. In every mature patch, a single specialist employment provider receives the plurality of onward referrals. Pick yours early, contract properly, and co-locate if you can.
• A curated long tail. Debt advice, housing, condition-specific charities, community groups: dozens of destinations, each receiving a handful of referrals. This tail is where holistic support actually happens. Curate 15 to 25 destinations, review quarterly, and prune what never gets used.
• Referrals back into social prescribing are a feature, not a failure. We observe WorkWell coaches routing participants into social prescribing when the barrier turns out to be loneliness or purpose rather than employment. That is the system working. Design the handover both ways.

The evaluation's caution: onward referrals sometimes did not meet participants' needs. The fix is not more destinations, it is feedback. Close the loop on every onward referral: did they attend, did it help. If your case system cannot tell you, that is a tooling problem worth solving (Decision 6).

7. Decision 5: take the in-work half seriously

Forty-three per cent of national starts are in work at first appointment, 13% on sickness absence. In our data the split is nearer half and half, and the single largest named employer sector is health and social work, ahead of retail and education. Read that again: the NHS and care workforce is WorkWell's biggest in-work customer.

This half of the service is the part the employment-support tradition understands least, and it is where WorkWell can be most distinctive. Out-of-work support has Jobcentres, providers, and forty years of programme design. In-work retention for someone six weeks into a stress-related absence has almost nothing between the GP's fit note and an occupational health referral their employer may not fund.

Practical implications:

• Build the fit note moment. The highest-value referral in the entire service is the GP consultation where a fit note is issued or renewed. That is your early-intervention window, and it is why the primary care pathway (Decision 1) matters more than any outreach campaign.
• Treat your own system as an account. If health and social work is the biggest in-work sector, your ICB, trusts, and care providers are your anchor employers. Set up a named route for them.
• Measure retention as an outcome, not just returns to work. In our data, of participants who completed their plan, 65% were in work at completion, and the largest single group had stayed in work full time. Staying in is a success story the sector undersells.

8. Decision 6: make the DWP dataset a by-product of coaching

Here is the starkest contrast in our entire dataset. Where the DWP national return is embedded in the case management workflow, completed by the coach inside the record they are already working, outcome completeness runs at effectively 100%: in one system, 4,039 outcome codes on 4,040 participant records. Where outcome measurement is a separate task in a separate tool, we observe wellbeing follow-up completion of roughly 19%.

Same coaches. Same participants. The difference is whether the data is a by-product of the work or an extra job after it.

This matters beyond compliance. The national rollout will be evidence-hungry: 34,000 pilot starts produced management information but, by DWP's own admission, no employment or health outcome statistics yet. The systems that can show, from their own records, who came in, through which route, with which barriers, and what happened next, will set the terms of their own commissioning conversations.

The SOP:

1. One system, no re-keying. The DWP return, the coaching record, and the onward referral log live in the same place. Every re-key is a completeness tax.
2. Capture at the moment of contact. The return is filled during or immediately after the session, not batched on Friday.
3. Watch two codes like a hawk. In our data, "other" outcomes run at 9% and exits with employment status unknown are the single largest closure category. Both are where evidential value leaks. Weekly data huddle, fifteen minutes, two questions: what closed as "other", and what closed as "unknown".
4. Follow-up is a diary discipline, not a survey. The 19% follow-up rate is not a motivation problem, it is a scheduling one. Book the follow-up contact when the plan is agreed.

9. Decision 7: the first 90 days

A mobilisation sequence, learned partly from what delayed the pilots:

10. What good looks like at month six

A scorecard to run monthly, with the ranges we observe in settled operations:

So what, now what

So what. WorkWell's pilot phase settled the existential question: there is real demand (34,000 starts against a cold start), a clear population (mentally health-led, half in work), and a working delivery grammar (front door, embedded coaching, backbone partnerships, data as by-product). The open question for every system that goes live in November 2026 is not whether this can work. It is whether they will adopt the operating disciplines the pilots paid to discover, or spend their first year rediscovering them.

Now what, if you are the lead:

1. This month: choose your operating model against your geography, start the data sharing agreements, and name your anchor employment partner.
2. Next quarter: stand up the front door and the GP pathway together, hire coaches for psychological confidence on permanent contracts, and configure the DWP return inside the case record before the first participant arrives.
3. By month six: run the scorecard above, publish your own numbers internally even when they are ugly, and close the feedback loop on every onward referral.

The pilots have already bought the map. The only unforced error left is refusing to read it.