Public sector design thinking without the theatre

Government innovation dies in two places: the business case that pretends certainty, and the pilot that never expires.

Give teams a kill date. Publish the learning even when the service does not scale. That is how you earn the next 90 days. Agencies do not need another lab with a pun name. They need permission, a metric, and a window.

This is Design Thinking for agencies, remounted without the theatre. NSW Government and Service NSW. Inland Revenue in New Zealand. The UK Government Digital Service pattern and the departments that still tender for it. Services Australia. Same physics: legislation, volume, identity, comms, industrial relations, and a public that will not applaud your mural.

Permission is the first artefact

In a bank, a product owner can often say yes to a concierge. In an agency, the yes is distributed: policy, legal, privacy, comms, a deputy secretary, sometimes a minister's office. Design thinking that ignores that distribution will produce a prototype nobody is allowed to show.

Write permission as a page:

• What we will test
• Who it touches (number, vulnerability)
• What we will not change in the system of record
• What we will say if asked
• Who can stop it
• When it expires

NSW cluster work that moves has this page, even if they do not call it that. IRD NZ work that moves treats the customer (a taxpayer, a small business, an intermediary) and the legislation as the same brief. Services Australia work that moves names privacy, identity, and payments on day one. GDS-style assessments already ask who the user is and what you have learned. Steal the questions. Put them on the permission page.

The business case is lying to you

It is lying because it is asked to. A business case that requires a 10-year NPV to start a discovery will invent a 10-year NPV. Design thinking cannot survive a fiction that large.

GDS in the UK spent years making the opposite normal: start small, show the thing, assess against a standard, do not pretend. Departments still drift back to fiction when the spend threshold hits. The steal for NSW, for IRD, for Services Australia, for a UK department that just tendered for design-thinking T&E: split the ask. Discovery and a 90-day test get a small envelope and a kill date. Scale gets a different envelope after evidence. If your procurement cannot split the ask, the theatre is in the panel, not in the workshop.

Give teams a kill date. Publish the learning even when the service does not scale.

What Design Thinking looks like in an agency room

Day one. Frame the problem in the words of a citizen or a frontline officer. Frame the constraint in the words of the act, the privacy team, or the payments owner. Interview both. NSW Service NSW-style "make it easier" work dies when only the citizen is in the brief. IRD work dies when only the act is in the brief. You need both sentences.

Day two. Prototype the service slice and the decision pack. A paper script for a claims conversation at Services Australia. A redesigned letter at IRD. A referral concierge in a NSW health or transport adjacent service. A GDS-style transaction path in a UK department. Plus the one-page risk and comms note.

Residue. A framed problem, a tested concept, a decision owner who sat in the room, a kill date. If the tender asked for "capability uplift," make the uplift this residue, not a cohort photo.

IRD, NSW, GDS, Services Australia: steal without copying

IRD NZ. Tax and entitlements are high-shame, high-complexity journeys. The steal is designing for intermediaries (accountants, bookkeepers) as users, not only for the individual, and testing letters before systems. A letter is a prototype.

NSW. Service NSW proved that a front door can be simpler than the cluster behind it. The steal is a 90-day window on a single seam (licence, fine, appointment) with an operational owner. The failure is a lab that cannot touch a cluster system.

UK GDS. The service standard is a stealable sceptical friend. Show the thing. Write in English. Name what you will stop. Departments that tender for design-thinking training are often asking for this muscle, not for sticky notes. Give them the muscle.

Services Australia. Volume and vulnerability. The steal is thinner tests (scripts, concierge, comprehension checks) that privacy can live with. The failure is a platform story that cannot change a payment this year.

Don'ts for public-sector sponsors

Don't stand up a lab to avoid giving permission to a line team. The lab will need the same permission and will have less operational knowledge.

Don't run a hackathon for citizens as a substitute for a service test. Energy is not evidence.

Don't hide killed pilots. Internal publication is how you stop the next team repeating them. GDS-style blogs did this in public. You can do it on an intranet if you must.

Don't measure success as number of officers trained. Measure services changed or honestly stopped.

Don't let the steering pack reset every 90 days without a kill. That is how pilots become furniture.

A 90-day card

Problem (citizen or officer words). Constraint (act, privacy, payments, comms). Thinnest test. N users. Owner. Metric. Kill date. What we will publish if we stop. What we will ask for if we continue.

NSW, IRD, a GDS-assessed UK team, and Services Australia can all fill that card. If they cannot, they are not ready to tender for a lab. They are ready to reframe. Reframing is the work.

A 90-day card, filled for a NSW cluster seam

The flagship piece gave you blank headings. Here is a card filled for a Service NSW-adjacent seam a cluster owner would recognise. We will not pretend this is a live programme we ran last week. We will not invent a minister quote. We will show what "permission, a metric, and a window" looks like when someone has to sign it.

Problem, citizen words. "I paid the fine and I still do not know if I can drive, because the letter and the portal disagree."

Problem, officer words. "We tell people to wait five days and then they queue at the service centre on day two, because we never gave them a status they trust."

Constraint. We will not rewrite the act this quarter. We will not change the system of record. Privacy and payments stay as they are. Comms and the cluster deputy must see any sentence a citizen will read. Industrial relations: no new rostering, no implied change to role descriptions, frontline staff in the test are volunteers from one site.

Thinnest test. A human concierge at one service centre plus a plain-language status SMS at two existing milestones, for a capped cohort of people with this fine type. Paper script behind the counter. No new app.

N users. A number the site manager can staff. Vulnerability flags named (language, digital exclusion). Anyone outside the cap is served the old way.

Owner. Named service owner in the cluster, plus the site manager. Not "the digital lab." Not "Service NSW as a brand."

Metric. Repeat contacts about "did my payment work" at that site, plus abandoned counter queues for this enquiry type. Baseline from last month's counts the site already has. If you do not have a count, the first 21 days are for counting.

Kill date. 90 days from the permission page, or earlier if complaints rise or comms pulls the sentence.

What we will publish if we stop. A one-page learning note on the intranet: what we tried, what we heard, why we stopped, what the next team should not repeat. GDS-style honesty, even if it never leaves the building.

What we will ask for if we continue. A second site, or a change to the existing letter, still without a new spine system. A different envelope. Scale is not automatic.

NSW work that moves looks like this card even when nobody calls it design thinking. NSW work that dies looks like a lab mural and a business case that pretends the act will be rewritten by Christmas. Put the card in the tender if you are buying T&E. Put the card in the deputy's inbox if you are the service owner.

A 90-day card, filled for IRD NZ

Inland Revenue journeys are high-shame and high-complexity. Intermediaries (accountants, bookkeepers) are users. Letters are prototypes. Fill the card that way.

Problem, customer words. "The letter tells me I owe something I cannot match to the return my accountant filed."

Problem, intermediary words. "I spend a day translating your letter for a small-business client, then I call, and the officer cannot see what I see."

Constraint. Legislation is the brief, not a rumour of the act. We will not change core assessment this window. We will test a letter and a call script. Privacy: no new data matching. Comms: any public sentence is marked. The test is not a promotional campaign.

Thinnest test. A redesigned letter on one notice type, plus a script for the officers who take the resulting calls, plus a short guide for intermediaries, sent through an existing channel. Paper first. Comprehension check with a small set of customers and a small set of intermediaries.

N users. A capped cohort on one notice run. Named segment. Anyone else gets the old letter.

Owner. Named service owner for that notice type, plus a comms counterpart, plus an operations lead for the call queue.

Metric. Call handle time and repeat contact on that notice type, plus intermediary callback rate if you already log it. Comprehension errors spotted in the check. No invented NPS of the workshop.

Kill date. 90 days or one notice cycle, whichever the operations lead can live with. Expiry is a control.

What we will publish if we stop. The old letter was clearer, or the constraint blocked the sentence we needed. Write it down. The next team should see it.

What we will ask for if we continue. A second notice type, still letters and scripts, still no core rewrite.

IRD work that moves treats the customer and the act as the same brief. IRD work that dies treats only one of those sentences as real. Services Australia rooms can steal this card by swapping in privacy, identity, and payments on the constraint line. A UK GDS-assessed team can steal it by swapping in the service standard questions.

A practical rule: if the card needs a 20-page appendix to be understood, you are not ready to test. You are ready to reframe.

Comms permission is a design artefact

In a bank, a product owner can often say yes to a concierge. In an agency, comms is a user whose job is "do not surprise the minister, the journalist, or the public." Design thinking that treats comms as a late clearance will produce a prototype nobody is allowed to show. Write permission as a page the comms lead can mark in 20 minutes.

What we will test, in citizen words a journalist would not have to invent.

Who it touches, including vulnerability. NSW and Services Australia already know why this line exists. NHS-shaped rooms know it as patient and family risk.

What we will not change in the system of record, so a leak cannot be spun as a secret platform launch.

The sentence we will use if asked. Prototype it. Two sentences, not a holding statement novel. "We are testing a clearer status message with a small group at one site. Nothing about eligibility has changed." IRD-shaped rooms need a sentence that does not sound like a new tax rule.

Who can stop public visibility. Named comms lead, named deputy or SES. Phone.

Expiry. If the test dies, the sentence dies with it. No leftover campaign language.

FOI or OIA posture. What we will release if asked: the permission page and the learning note. GDS-style services earned the right to iterate by making learning cheaper than theatre. You can do that on an intranet if you must. Hiding killed pilots is how the next team repeats them.

UK departments lose months here. NSW clusters lose months here. IRD and Services Australia lose months here. Industrial relations can sit next to comms on the same page if the test touches a frontline role. Do not imply a role change in a script and then act shocked when a union asks. Write "no change to role descriptions, volunteer site only."

Interview comms the way you interview a customer. Their job to be done is "do not be surprised." A paper prototype of the sentence is more useful to them than a clickable journey they cannot approve. If they say no, ask what test they would sign this month. Write that test down. That is still design thinking.

How a 90-day window actually runs, week by week

Theatre loves a lab. Operators need a calendar.

Days 1 to 7. Permission page signed. Comms sentence marked. Frontline volunteers briefed. Baseline counts pulled from numbers the site already believes. If a signature is missing, the window has not started. Do not pretend.

Days 8 to 21. Thinnest test live. Daily 10-minute huddle at the site, not a programme pack. Blockers written in citizen or officer words. If the test cannot start, write "blocked" as a result and spend the rest of the window on the constraint. A blocked test is cheaper than a fake live one.

Days 22 to 60. Look-again at day 30. Keep, kill, or rewrite the script. Publish an internal note even if nothing scaled. Invite the stopper to the look-again. Surprise is how public-sector lean dies.

Days 61 to 90. Either a funeral with a learning note, or a narrow ask for the next envelope. No automatic scale. No sequel workshop as a substitute for a decision. If the steering pack wants more analysis, give them the card and the note, not a new discovery.

Services Australia can hang this on a programme board. NSW can hang it on a cluster owner. IRD can hang it on a service meeting. A GDS-shaped UK team can hang it on an assessment window. If aftercare has no home, you will get a sequel course. Sequels are how theatre bills the taxpayer twice.

A practical comms-and-permission pack:

One permission page, one comms sentence, one 90-day card, one named owner, one kill date. If a UK department, an NHS trust, a NSW cluster, IRD, or Services Australia cannot fill those four pages, they are not ready to tender for a lab. They are ready to reframe. Reframing is the work. The mural was never the work.