AGL: experiments inside a listed energy company

A listed energy company does not get a standing ovation for a prototype. It gets a question from risk, a question from retail, and a question from whoever owns the bill the household will not understand. AGL Energy Ltd is an Australian listed public company involved in the generation and retailing of electricity and gas. The people in the room already knew they could not "move fast and break things." The useful question was: what is the thinnest experiment risk and retail can live with?

Listed energy businesses do not get to "move fast and break things." They do get to run thinner experiments. Lean Startup T&E with AGL focused on problem interviews, offer tests, and kill criteria that risk and retail could live with.

What the room was for

AGL's world is generation and retail, households and SMEs, a licence, a brand, and a media environment that will not be kind to a cute experiment that lands wrong on a hardship customer. Retail leads, risk leads, product and digital counterparts, and the people who actually take the calls: that is the mix that can run lean. A centre of excellence alone cannot.

We did not ask them to pretend they were a startup. We asked them to write a loop the Australian Energy Regulator's marketing and hardship physics would not laugh at. Name the section you are bound by. Name the customer. Name the thinnest version. Name the kill.

Problem interviews came first. Not a survey mural. Conversations about bills, offers, solar and battery confusion, hardship, SME time-poverty. The point of lean interviews inside a retailer is not empathy theatre. It is to stop building an offer nobody asked for and legal would not let you describe.

If risk and retail cannot live with the test, you do not have a test. You have a launch fantasy.

Offer tests, not platform stories

The experiments that belong in this building:

• A non-binding expression of interest on a new bundle, with legal on the claims.
• A script in a call-centre bay or a store, measured on handle time and repeat contact.
• A paper or PDF explanation of a bill or a hardship pathway, tested for comprehension.
• A concierge for 30 customers, human-run, while you watch cycle time and complaints.
• A kill switch a named retail-risk lead can pull.

The experiments that do not belong on day one: a new tariff in the billing stack, a new app, a vendor platform justified by a 10-year case. Those might be later. They are not a loop.

AGL teams (and every peer retailer in AU and NZ) already have more ideas than loops. The T&E job is to make the loop cheap and visible. Write a test licence: what we will do, who it touches, what we will not change in production, what data, who can stop it, when it expires, what we will report.

Kill criteria a listed company can publish internally

"We will stop if complaint rate in the bay exceeds baseline." "We will stop if we cannot get 200 target customers to complete the concierge in 30 days." "We will stop if the claim we need to make will not clear legal." "We will stop if unit economics require a subsidy the board will not name."

Boards relax when they can see the stop. Energy boards especially. Show the stop before the concept.

We will not invent AGL's results. We will not quote a living executive. We will say the method left the room in a form risk and retail could reuse: interviews, offer tests, kill criteria. That is the case.

For the next retail product committee

We will test [offer or conversation] with [N] customers in [channel] for [N] days. We will not change [billing / CIS]. We will use [data] under [existing permission]. [Name] in retail risk can stop it. We will kill it if [metric] is worse than [baseline]. We will report learning on [date] whether we scale or stop.

If the committee cannot live with that paragraph, do not book a lab. Reframe. The regulator did not ban the paragraph. Ambiguity did.

Peers: Origin-shaped corporates, NZ retailers, UK energy suppliers inside Ofgem physics. Same paragraph, different letterhead.

What risk and retail each owe the loop

Retail owns the customer sentence and the channel. Risk owns the stop and the cohort cap. Product or digital owns the artefact. None of the three owns the loop alone. AGL-shaped rooms fail when innovation owns the loop and the other two visit on Friday. Put all three on the test licence. Put all three in the 30-day review. If risk cannot attend, you do not have a test. If retail cannot attend, you have a lab hobby.

The AER-shaped rules on marketing and hardship are not a veto of learning. They are a specification for the sentence the public will see. Write that sentence in the interview guide and on the landing page. Then listen. Then kill or keep. That is lean inside a listed energy company. It is not glamorous. It is how you stay allowed to run the next loop.

The test licence, written like a risk artefact

A listed energy company does not get a standing ovation for a prototype. It gets a question from risk, a question from retail, and a question from whoever owns the bill the household will not understand. AGL Energy Ltd is an Australian listed public company involved in the generation and retailing of electricity and gas. Collective Campus delivered Lean Startup training and enablement in that physics. The residue that travels is not a slogan. It is a test licence a product committee can mark.

Write the licence on one page before anyone books a bay, a page, or a concierge. Risk can mark a page. Risk cannot mark a mural. Retail can live with a page. Retail cannot live with a "we will figure out the cohort later" speech.

Use this page. Fill the brackets. Do not paste AGL numbers we do not have. We will not invent theirs.

What we will do. One sentence. "We will run a rewritten hardship conversation in one inbound bay." "We will offer a non-binding expression of interest on a bundle, with legal on the claims." "We will test a paper or PDF explanation of a bill for comprehension." If you need a second sentence, you have two tests. Split the licence.

Who it touches. Number, segment, vulnerability flags. AGL-shaped retail already knows that a hardship customer is not a sandbox. Name the cap. Name whether anyone on a hardship programme, a life-support register, or a payment plan is in or out. Default: out, unless hardship is the design input (see below) and retail risk has signed the script.

What we will not change in production. Billing. CIS. Tariff tables. Disconnection path. The sentence that goes on a regulated notice. Write the negatives. This is how you keep a 30-day loop from becoming a release train. Origin Energy-shaped and EnergyAustralia-shaped peers need the same negative list. So do Mercury, Contact Energy, and Genesis Energy in New Zealand. So do Octopus Energy, British Gas, E.ON UK, and Ovo under Ofgem.

What data we will use, under what existing permission. Call recordings you already have. QA notes. Existing research panels. A landing page that does not create a new marketing claim. If you need a new data feed, you do not have a lean test. You have a project.

Who can stop it. A name in retail risk. Not "the programme." Not "legal will be consulted." A human who can pull the kill switch without a steering pack. Put their mobile next to the date.

When it expires. 30 days is a default. 90 if you need a billing cycle. Expiry is a control. AGL boards, like any listed energy board, relax when they can see the stop. Show the stop before the concept.

What we will report. One behaviour: complaints in the bay, handle time, repeat contact, completion of a concierge, comprehension on a bill explanation. Not NPS of the workshop. Not unique visitors to an unapproved page. Report whether you scale or stop, on a date, in the same channel you would have used for a win.

Named owners on the licence. Retail owns the customer sentence and the channel. Risk owns the stop and the cohort cap. Product or digital owns the artefact. None of the three owns the loop alone. AGL-shaped rooms fail when innovation owns the loop and the other two visit on Friday. Put all three on the page. Put all three in the 30-day review. If risk cannot attend, you do not have a test. If retail cannot attend, you have a lab hobby.

Collective Campus will keep using this licence as the residue of Lean Startup T&E at AGL. Position B: we delivered the work. We did not ship a tariff. We did not endorse a retail product. We taught a loop a listed retailer can live with.

Which experiments belong (and which are launch fantasies)

The experiments that belong in this building are thin, reversible, and boring to a conference. That is a compliment.

A non-binding expression of interest on a new bundle, with legal on the claims. The customer is not on the product. They are on a sentence. You measure whether the sentence earns a hand-raise you can follow with a human. If legal will not clear the sentence, you learned. Validated learning still counts when the learning is "we cannot say that, so we will test this."

A script in a call-centre bay or a store, measured on handle time and repeat contact. The operator is the builder. AGL retail already has people who take the calls. NAB-shaped and Lloyds-shaped rooms have the same physics in a branch or a care bay. Energy peers in New Zealand and the United Kingdom do too. Staff the experiment with the operator, not a centre of excellence that will visit on Friday.

A paper or PDF explanation of a bill or a hardship pathway, tested for comprehension. Sit with 20 customers or 20 operators. Ask them to tell the page back. If they cannot, you do not have an offer problem yet. You have a sentence problem. The Australian Energy Regulator's marketing rules already tell you what you cannot claim. They do not tell you that you cannot test whether a household understands the path.

A concierge for a capped cohort, human-run, while you watch cycle time and complaints. Thirty customers is a shape, not an AGL statistic. A human runs the process. Nothing in billing changes. You are allowed to be embarrassed by the operations. You are not allowed to be uncontrolled.

A kill switch a named retail-risk lead can pull. This is an experiment, not a decoration. If nobody can stop it, it is a launch.

The experiments that do not belong on day one, and should die in the same tone as an Inkeros-style pre-lunch funeral:

A new tariff in the billing stack. That is a regulated change. It can be a later build if a thin test earned it. It is not a loop.

A new app. Apps are where energy programmes go to wait. If the question is "will this conversation reduce repeat contact," a bay and a script will tell you. An app will tell you about stand-ups.

A vendor platform justified by a ten-year case. Platforms might be true later. They are not a 30-day test licence. If the vendor cannot live inside the licence above, they are selling a keynote.

A hardship campaign that treats a vulnerable household as a growth segment. See the hardship section. This is how you get the sequel you deserve.

AGL teams, and every peer retailer in Australia and New Zealand, already have more ideas than loops. The T&E job is to make the loop cheap and visible. Collective Campus ran that job. We will not invent which AGL ideas survived. We will say which shapes can survive contact with risk and retail. That is the useful half of the case.

Origin, New Zealand retailers, and Ofgem-shaped peers

The paragraph travels. The letterhead changes. Name the peers so a target executive has a reason to visit, then a licence they can steal.

Origin Energy and EnergyAustralia. Listed or large Australian retailers in the same AER-shaped world AGL lives in. Same marketing rules. Same hardship physics. Same temptation to start in the stack. Steal the test licence, not AGL's unpublished results (we do not have them, and we will not invent them). If Origin or EnergyAustralia cannot put retail, risk, and product on one page, they are not ready for lean. They are ready for another transformation poster.

Mercury, Contact Energy, Genesis Energy, Meridian. New Zealand generation and retail, different regulator, same household bill. The Electricity Authority and the Utilities Disputes path are not plot twists. They are the brief. A concierge, a script, a non-binding expression of interest, a comprehension test: those still belong. A new tariff in the first sprint still does not. Walk the last bill that confused a household in Auckland or Christchurch the way you would walk one in Melbourne.

Octopus Energy, British Gas, E.ON UK, Ovo, ScottishPower. United Kingdom suppliers inside Ofgem physics, including standards of conduct, vulnerability, and the marketing of tariffs. Consumer duty thinking from the FCA side of the street travels into how you explain a bill even when Ofgem is the energy regulator. The test licence still holds: cohort cap, expiry, named stopper, no uncontrolled change to a customer in production. Octopus-shaped product culture already knows how to ship. The steal is not "be more startup." The steal is "write the stop a conduct owner will sign." British Gas-shaped incumbents already know how to run a bay. The steal is "do not wait for a platform to test a sentence."

A Canadian or US utility retail arm, a Singapore or Hong Kong licensed utility. Different letterheads again. Hydro-shaped and provincial retailers in Canada, a US retail-choice market, SP Group-shaped utilities in Singapore: the licence is still the artefact. If your innovation lead says lean dies on contact with a licence, hand them the page. Regulation is not the opposite of lean. Ambiguity is.

What we will not claim: that any of these peers ran the AGL room, that Ofgem blessed a method, or that Origin "should" copy a number. We name them because the physics rhyme. Collective Campus delivered T&E with AGL. Peers steal the licence. That is the flywheel.

The product committee paragraph, filled as a template

The existing residue is a paragraph a product committee can sign. Here it is again, then filled as a working template. The filled versions are not AGL results. They are blanks completed with shapes, not with invented complaint rates, conversion rates, or revenue.

The blank:

We will test [offer or conversation] with [N] customers in [channel] for [N] days. We will not change [billing / CIS]. We will use [data] under [existing permission]. [Name] in retail risk can stop it. We will kill it if [metric] is worse than [baseline]. We will report learning on [date] whether we scale or stop.

Filled template A, hardship conversation (shapes only):

We will test a rewritten hardship conversation with [N] inbound customers in one retail bay for [N] working days. We will not change billing or the CIS. We will use existing call recordings and QA notes under existing permission. [Name] in retail risk can stop it. We will kill it if complaint rate in that bay is worse than that bay's own baseline, or if a hardship or vulnerability flag is handled outside the signed script. We will report learning on [date] whether we scale or stop.

Filled template B, offer shape (shapes only):

We will test a non-binding expression of interest on [bundle or conversation] with [N] target customers in [web / store / inbound] for [N] days. We will not change tariff tables or the CIS. We will use [landing page / script] copy that legal has marked, under existing marketing permission. [Name] in retail risk can stop it. We will kill it if we cannot get [N] completed expressions without a claim legal will not clear, or if complaint rate on the path exceeds the channel baseline. We will report learning on [date] whether we scale or stop.

Filled template C, comprehension (shapes only):

We will test a paper or PDF explanation of [bill line / hardship pathway / solar connection step] with [N] customers or operators in [store / bay / remote interview] for [N] days. We will not change production notices. We will use the existing explanation as the control. [Name] in retail risk can stop it. We will kill it if comprehension (customer restates the path) does not beat the control, or if the sentence we need will not clear legal. We will report learning on [date] whether we scale or stop.

If the committee cannot live with a filled paragraph, do not book a lab. Reframe. The regulator did not ban the paragraph. Ambiguity did. Take template A to an AGL-shaped or Origin-shaped forum. Take it to a Mercury-shaped or Contact-shaped forum in New Zealand. Take it to an Ofgem-shaped journey forum in the United Kingdom. Same paragraph, different letterhead.

A practical covering note for the committee pack:

Attached is a one-page test licence and a filled paragraph. There is no platform. There is no ten-year case. There is a named stopper, a cohort cap, an expiry, and a report date. If you want a mural instead, say so and we will save the catering.

Collective Campus would rather watch a committee mark that page than watch another energy innovation day produce a stack of concepts with no licence. We will not put a fake AGL baseline in the brackets to make the template look "real." Real is a name, a bay, and a date the company already owns.

Hardship as a design input, not a CSR slide

Energy retail is a conduct business wearing a commodity costume. Hardship is not a theme for a Friday keynote. It is a specification.

The Australian Energy Regulator and state retailers already live inside rules on marketing, hardship, and disconnection. Those rules tell you what you cannot claim and what you must offer as a path. They do not tell you that you cannot sit with customers and listen. They do tell you that a cute experiment that lands wrong on a hardship customer is not lean. It is a conduct event.

Treat hardship as a design input the way a bank treats vulnerability or an Allens-shaped firm treats conflicts. Write it on the brief on day one.

Who is out of the first cohort by default. Payment plan, hardship programme, life-support register, known vulnerability flag. Default out unless the test is the hardship path itself and retail risk has signed the script. AGL-shaped, Origin-shaped, and EnergyAustralia-shaped rooms already have these flags. New Zealand retailers have their own customer-care and medically dependent physics. Ofgem-shaped suppliers have vulnerability and priority-services lists. Use the list you already have. Do not invent a new segmentation project to start a 30-day test.

What the sentence on the public path is allowed to say. Legal marks it. Marketing does not "punch it up" after. Hardship language that over-promises a discount you cannot keep is a marketing breach wearing empathy. Hardship language that under-explains the path is how you earn a second call and a complaint. Write the sentence in the interview guide and on the landing page. Then listen.

What you interview for. The last bill that confused them. The last hardship call. The last solar or battery promise that did not match the connection. Record the behaviour, not the compliment. "I would love that" is not evidence. "I called twice last Tuesday" is. Compliance can sit in, or they can review the guide. They should not ban the conversation. If they try, you asked a promotional question or you used the wrong channel. Fix the guide. Do not skip the customer.

What you prototype. A script a bay can run. A PDF a household can restate. A concierge that does not change disconnection or billing. A kill switch on complaint rate and on any handling outside the signed path. You do not prototype a new hardship brand campaign. You do not prototype a generator joke. You do not put a vulnerable customer in a growth experiment because the lab needed a story.

What you kill on. Complaint rate worse than the bay baseline. A single conduct flag the named stopper will not absorb. A claim that will not clear. Unit economics that require a subsidy the board will not name, if you are testing an offer that sits next to hardship (a bundle, a payment option) rather than the path itself. Publish the kill internally the way you would publish a win. Energy boards especially. This is how you earn the next loop.

Hardship as a design input also tells you which experiments never belong. A tariff change that first touches hardship customers. An AI script in a bay with no human check on the hardship sentence. A page that implies a government-style relief you cannot deliver. Those are not bold. They are how listed retailers get the media environment they already know is unkind.

Peers should steal this without waiting for a new AER consultation, a new Electricity Authority paper, or a new Ofgem open letter. The physics are already on your licence. Collective Campus delivered Lean Startup T&E with AGL in that physics. Named because the client is real and the constraint is the story. Not a product endorsement. Not a fabricated hardship statistic. If your energy innovation programme cannot write hardship on the test licence, you do not have a programme. You have a slide.