AI-powered landing page generator to test ideas fast

On the workshop table: battery packs, not sticky notes. Infinitev is an Australia and New Zealand-based EV and hybrid battery specialist. The company tests, grades, and remanufactures EV and HEV battery packs. The commercial question in the room was older than generative models: can we learn whether an offer has a pulse before we spend a quarter and an agency cycle? The new move was an AI Design Sprint residue: throwaway landing pages, drafted at the speed of the conversation, true enough to test.

Infinitev tests, grades, and remanufactures EV/HEV battery packs. Speed-to-learning on new offers mattered more than another campaign cycle. The sprint produced an AI-assisted landing page generator so the team could test ideas in days, not quarters.

Why this is an AI Design Sprint, not a hackathon

A hackathon would have produced a demo and a photo. Infinitev needed a decision: which offers are worth a week of real traffic or a week of sales conversation, and which die on the page.

Vanilla design-sprint discipline still applied. Frame the customer (fleet, workshop, consumer, insurer, depending on the offer), frame the constraint (claims you can stand behind on a remanufactured pack, channel, brand), pick a thinnest test. The AI layer compressed option generation and first-draft pages. Humans made the sentences true. That is the method we named in the flagship: models in the loop to accelerate artefacts, humans on the decision.

The people in the room were commercial, operations, and whoever owns the customer promise. Not a lab. A specialist business in AU and NZ does not have a lab to hide in. Good.

The residue was not "we have a generator." It was "we can test an offer this week."

What we actually built toward

A repeatable way to:

1. Draft a landing page (or a page variant) for an offer hypothesis.
2. Keep claims inside a rule someone operational can sign.
3. Put it in a test channel (paid, email, partner, sales-enablement) without waiting on a quarter-long campaign.
4. Read a behaviour: enquiry, call, qualified lead, bounce. Not "the team liked the headline."
5. Kill or iterate on a date.

The generator is a means. If the team fell in love with the tool and stopped reading results, we would have failed. Tools are how people avoid deciding. We kept the Friday (or the last afternoon) as a decision slot.

Research synthesis used the messy inputs a specialist already has: workshop conversations, repeat objections, the difference between AU and NZ channel reality, the questions people ask about safety and warranty. AI helped cluster. It did not invent the customer.

Constraints you write before the prompt

EV battery remanufacturing sits in a trust market. Over-claim and you will earn a regulator or a lawyer. Under-claim and you will earn silence. Write the claims rule on a page. Put it next to the generator. Same as a bank writing a guardrail before a prompt.

Channel economics matter. A page that cannot be followed by a human process (test, grade, remanufacture, return) is a lie. Operations in the room is not optional.

We will not invent Infinitev's conversion rates. We will not quote a living founder as if they sat for a profile. We will say the format left a residue the commercial team could run.

If you are a specialist in AU or NZ

If you sell a technical thing (batteries, equipment, a B2B service) and you wait on an agency cycle to learn:

• Book two days.
• Bring commercial, ops, and the claims owner.
• Use AI to draft ten offer angles and three throwaway pages.
• Pick one test channel.
• Decide on day two.
• Read results in 14 days.
• Kill or keep.

Atlassian-scale product teams already think this way. GovTech-scale services have a standard instead of a page. Capital One-scale orgs have design weeks. Infinitev is the specialist rhyme: small enough that a quarter matters, serious enough that a claim matters.

AU and NZ as one commercial problem, two channels

Infinitev works both sides of the Tasman. Offers that read true in an Australian workshop network can land wrong in a New Zealand one, and the reverse. The generator is useful only if the team can fork a page, change the claim and the channel, and test again in days. That is the residue of an AI Design Sprint in a two-country specialist: not one hero campaign, a habit of cheap tests.

We kept humans on the sentence a customer would take to a technician. Models draft. Operations signs. Commercial decides. If any of those three is missing, you have a content toy. Infinitev did not need a toy. They needed to learn whether an offer had a pulse before the quarter closed.

The claims rule before anyone opens a prompt

Infinitev is an Australia and New Zealand-based EV and hybrid battery specialist. The company tests, grades, and remanufactures EV and HEV battery packs. Collective Campus ran an AI Design Sprint with them. The useful residue was not a model zoo. It was a way to test whether an offer had a pulse in days, not a quarter. The first artefact in that room was not a prompt pack. It was a claims rule.

Write the claims rule on one page. Put it next to the generator. Nobody opens a model until the page exists. Generation is cheap. Untrue claims are not. EV battery remanufacturing sits in a trust market. Over-claim and you will earn a regulator or a lawyer. Under-claim and you will earn silence.

Steal the page. Fill it for your specialist category.

What we may say about the pack. Tested, graded, remanufactured: those are Infinitev verbs. They are not synonyms for new. They are not a warranty you have not written. They are not a safety certificate you do not hold. Write the verbs you can stand behind in front of a technician, an insurer, a workshop, or a fleet manager. If a sentence needs a lawyer to decode, it is not a landing-page sentence.

What we will not say. Range guarantees you have not measured on that grade. "As new" if you mean remanufactured. A comparison to a competitor pack you have not opened. A New Zealand claim written for an Australian workshop network, or the reverse, without a fork (see below). ACCC-shaped Australian Consumer Law and NZ Commerce Commission physics already sit on this list even if nobody from a regulator is in the room. Put the physics on the page so a model cannot charm them away.

Who signs the rule. Operations, because they live with the pack. Commercial, because they live with the offer. The claims owner (brand, legal, or the founder-shaped person who will take the call when a page is wrong). Three signatures. If any is missing, you have a content toy. Infinitev did not need a toy.

How the rule enters the prompt. Paste it. Do not summarise it from memory after lunch. The model will happily draft ten offer angles that convert in a vacuum. Your job is to make the first draft true. Humans on the sentence a customer would take to a technician. Models draft. Operations signs. Commercial decides.

What happens when a draft violates the rule. It dies in the room. No "we will legal it later." Later is how cheap generation becomes an expensive complaint. Kill the angle. Keep the habit.

We will not invent Infinitev's conversion rates. We will not quote a living founder as if they sat for a profile. We will say the claims rule was the constraint that made the sprint a decision machine instead of a content mill. Position B: Collective Campus delivered the sprint. We did not endorse a battery product. We taught a room to write the rule before the prompt.

A bank writing a guardrail before a prompt is the same move in a different building. A GovTech Singapore team putting a service standard in the room on day one is the same move. A Capital One-style or HK/SG desk putting conduct on the page is the same move. Infinitev is the specialist rhyme: small enough that a quarter matters, serious enough that a claim matters.

The Australia and New Zealand channel fork

Infinitev works both sides of the Tasman. Offers that read true in an Australian workshop network can land wrong in a New Zealand one, and the reverse. The generator is useful only if the team can fork a page, change the claim and the channel, and test again in days. That is the residue of an AI Design Sprint in a two-country specialist: not one hero campaign, a habit of cheap tests.

Fork means four things, written before you generate the second page.

The claim. Warranty language, returns, freight, what a technician in Auckland can actually do versus a workshop in Melbourne or Brisbane. If the Australian page implies a pickup you do not run in New Zealand, you did not localise. You copied. Copying is how specialists earn a quiet complaint in the smaller market and then wonder why the channel "does not work."

The channel. Paid, email, partner, sales-enablement: these are not the same on both sides of the Tasman. A partner workshop in Australia is not automatically a partner workshop in New Zealand. A fleet conversation in one country is not a consumer conversation in the other. Pick one test channel per page. Do not "also run it in NZ" as a Friday thought.

The follow-through. A page that cannot be followed by a human process (test, grade, remanufacture, return) is a lie. Operations in the room is not optional. If New Zealand follow-through is slower, more freight-bound, or more partner-bound, the page must say a clock you can keep. Speed-to-learning does not mean speed-to-over-promise.

The kill. Each fork has its own date and its own owner. A page that dies in New Zealand does not automatically kill the Australian test, and the reverse. Do not let a blended dashboard hide a dead fork. Blended dashboards are how campaign teams keep budgets.

Collective Campus kept humans on both forks. AI clustered the messy inputs a specialist already has: workshop conversations, repeat objections, the questions people ask about safety and warranty, the difference between AU and NZ channel reality. It did not invent a New Zealand customer from an Australian average. If your generator cannot fork, you bought a brochure engine.

Peers should steal the fork even when they do not sell batteries. A Redarc-shaped or industrial-equipment specialist selling into Australia and New Zealand already knows the channel split. A Singapore and Hong Kong specialist selling the same SKU into two regulatory voices already knows it. A Canadian and US specialist (Inkeros-shaped engagement is a different category, but the cross-border fork is the same physics) already knows it. Write the four lines. Then generate. Not the other way around.

The decision slot, or you scheduled a lab

The last afternoon is a decision slot. If you skip it, you scheduled a lab. Infinitev needed a decision: which offers are worth a week of real traffic or a week of sales conversation, and which die on the page.

Book the decision-maker for that slot before you book the venue. Commercial owns the offer. If they send a delegate who cannot kill, you will get a showcase of pages and a polite "let us take this offline." Offline is where generators go to become toys.

The slot has a script. Steal it.

Fifteen minutes: the claims rule, restated. If a page drifted, it is not in the decision set. It is in the bin.

Thirty minutes: each surviving page, one behaviour you will read. Enquiry, call, qualified lead, bounce. Not "the team liked the headline." Name the channel. Name the 14-day read. Name the owner who will look at the behaviour. If you cannot name those three, the page is a draft, not a test.

Thirty minutes: go, iterate, or kill. Go means the page ships into the named channel this week. Iterate means one change you can make before the same slot tomorrow or on a short Monday. Kill means a sentence of learning on the wall. Two or three goes is plenty. Ten pages with no kill is a content calendar.

Fifteen minutes: the calendar. 14-day read. Named owner. A hold. If aftercare has no home, you built a generator. Infinitev did not need a generator as a pet. They needed to learn whether an offer had a pulse before the quarter closed.

Hackathon energy will show up when the pages look good. They will look good. Models are fluent. Fluency is not evidence. Keep the slot boring. Atlassian-scale product teams already think this way about spikes. GovTech-scale services have an assessment window instead of a page. Capital One-scale orgs have design weeks. A specialist like Infinitev runs the slot instead of a quarter-long campaign cycle. Same noun: decision. Different artefact.

We will not invent what Infinitev chose. We will say the slot existed, and that without it the sprint would have been a demo. Collective Campus will keep ending AI Design Sprints on a decision. If your vendor ends on a repo or a prompt library, make them say so.

When the generator becomes a toy

A generator becomes a toy the moment the team loves the tool more than the read. Tools are how people avoid deciding. We kept saying that in the Infinitev room. Here is the diagnostic, so a peer can catch the slide.

You have more pages than tests. Ten angles and three throwaway pages can be a sprint. Forty pages and a backlog of "we should try this headline" is a studio. Studios are agencies. You were trying to stop waiting on an agency cycle. Do not become one.

Nobody is reading a behaviour on a date. Enquiry, call, qualified lead, bounce. If the only metric is "we can now generate a page in 20 minutes," you measured the toy. Speed of draft is an input. Speed-to-learning is the residue.

Claims drift because generation is cheap. The first page respected the rule. The ninth page said "as new" or promised a range the grade does not support. Cheap drafts are how trust markets get sloppy. Put the rule next to the keyboard. Kill the ninth page in public.

Operations left the room. Commercial kept prompting. The pages still look like offers. They no longer look like a process (test, grade, remanufacture, return) a workshop can run. A page without a process is a lie. The toy is now a risk.

The AU/NZ fork collapsed into one hero. Someone picked the prettier page and ran it in both countries. You are now in campaign logic. Campaign logic is what you came to avoid.

The decision-maker stopped attending. The generator has users. The sprint has no owner. This is the lab pattern in a smaller building. Infinitev-scale specialists do not have a lab to hide in. Good. Do not build one out of a prompt.

You started productionising the generator. A sprint artefact is allowed to be embarrassing. A production CMS with workflow, roles, and a brand book is a product. If you need that product, staff it like one, with a backlog and a kill date. Do not pretend a two-day residue is a platform.

When you see two or more of those, shut the generator for a week. Run the 14-day read on whatever already shipped. Kill or keep. Then, if you still need a page, write the claims rule again and generate three, not thirty.

Collective Campus would rather watch a specialist run two ugly tests than watch a beautiful internal tool that never faces a customer. We will not invent Infinitev's kill count. We will say the failure mode is general, and that we designed the sprint to resist it. Position B: we ran the room. We did not sell them a generator as a product. If a peer buys a generator without a claims rule, a fork, a decision slot, and a read, they bought a toy. We will keep saying that.

A two-day if you are another specialist

If you sell a technical thing (batteries, equipment, a B2B service, a medical-adjacent device, an industrial consumable) and you wait on an agency cycle to learn whether an offer has a pulse, steal the Infinitev shape. Do not steal a fabricated performance number. Do not steal a battery claim. Steal the two days.

Day 0, ninety minutes. Claims rule. Channel. Decision-maker. AU/NZ fork or your equivalent (SG/HK, CA/US, UK/IE). If any of the four is missing, do not book the room. A specialist in Singapore or Hong Kong writes the MAS-shaped or customs-and-claims rule their category actually has. A Canadian or US specialist writes the advertising and product-safety rule they already live with. A UK specialist writes ASA-shaped and sector-licence rules. Same page, different letterhead.

Day 1 morning. Dump the messy inputs you already have: workshop objections, insurer questions, technician FAQs, the last three offers that died in silence. Synthesise with a model in the loop. Humans argue. Claims owner in the chair. Generate ten offer angles. Kill the ones that violate the rule before lunch. That funeral is the morning.

Day 1 afternoon. Draft three throwaway pages, not a brand system. Fork at least one for the second country or the second channel if you sell across a border. Operations signs the sentence a customer would take to a technician. If operations will not sign, the page is not a test.

Day 2 morning. Expose each page to three humans who were not in the build: a workshop lead, a fleet manager, a partner, a sales engineer. Repair. Tighten the behaviour you will read. If they need the facilitator to explain the offer, rewrite the offer.

Day 2 afternoon. Decision slot. Go, iterate, or kill. One test channel per go. Owner. 14-day read. Circulate a page before people stand up. Delete the campaign readout that would have restated it next month.

Day 16 (the 14-day read plus a day). Kill or keep. Write the sentence. If you keep, you may iterate one variable. If you kill, you may generate again. You may not "leave it up" because the page was cheap.

Atlassian-scale product teams already run spikes inside a cadence. GovTech Singapore already assesses a service slice. Capital One-style orgs already run design weeks. Infinitev is the specialist rhyme Collective Campus will keep naming: Australia and New Zealand, packs on the table, a generator as a means, a decision as the residue. A peer in any of our geos can run the two days without buying a lab and without waiting on a quarter.

A practical booking note:

We will write the claims rule before any prompt. We will fork Australia and New Zealand (or our two channels) on purpose. We will use AI to draft ten angles and three throwaway pages. We will decide on day two. We will read a behaviour in 14 days. We will kill or keep. We will not productionise a generator in the sprint. If we fall in love with the tool, we will shut it and read the tests we already have.

That is an AI Design Sprint for a specialist. The rest is a toy with better autocomplete.