AI pergola design helps draw the line. It does not close the buyer.

A buyer types AI pergola design into Google and taps through six tools before her coffee goes cold.

Two are image generators that make pretty renders with no price. One picks a “matching” pergola from a catalogue. One is a generic 3D configurator with no real product behind it. One asks her to book a call before she has seen anything.

The sixth shows a louvered build in 3D, lets her set the size and the post finish, and emails a priced PDF. That is the one she forwards to her partner.

We get asked about AI on every demo now. The buyer never asks. The buyer asks if the pergola will fit. That is still the only question that matters.

AI is doing real work in pergola design. Just not where the press releases point. It earns its keep in four corners of the buying flow and sits out the part that decides the sale.

The four places AI earns its keep#

Inspiration images. A buyer at the top of the funnel does not know yet whether they want aluminium or wood, louvered or fixed. A generative model like gpt-image-1 produces a dozen believable moods in the time it takes to write the brief. Good for the inspiration grid and ad creative. The render is the cover. The configurator is the book.

Garden-photo composition. This is the one that closes the gap between “I can picture it” and “I can show my partner.” The buyer uploads a photo from their back door, the configured model drops in at roughly the right scale and lighting, and they see their own garden with the pergola in it. Buyers forward those renders more than anything else in the funnel.

Lead intent scoring. Time on the design, meaningful changes, spec depth, whether they came back to a saved build. No single signal decides anything. Together they rank Monday’s inbox, so the dealer rings the top five before lunch instead of working twenty in arrival order.

Multilingual dealer copy. A manufacturer running one configurator across 30 dealers in six countries used to translate the UI, the product copy, the PDF lines and the email cadence by hand. What took six weeks of agency time per locale now takes a week of internal review.

Same shape in all four. AI is the assistant. It is faster than a person at a narrow task and useless at the wide one.

What AI does not do yet#

Five places where the marketing has run ahead of the technology.

Replace the parametric model. A generative model makes a beautiful picture. It cannot rebuild the geometry when the width goes from 3.2 to 4.7 metres and the post count has to go from four to six. One knows beam spans and bracket types. The other knows what pergolas tend to look like in photographs.

Replace real SKUs. Every post, beam, louvre and fixing is a part in the factory and a line in the BOM. AI can suggest combinations. It cannot invent them. A render containing a roof profile your supplier does not make is a deal you cannot fulfil.

Replace pricing rules. Margin bands, install modifiers by postcode, dealer overrides, tax, promotional cut-offs. Finance and sales have to agree on those, change them, and audit them. The day a model sets price autonomously is the day you lost control of margin.

Replace mobile performance. Three seconds on 4G. That budget has not moved. It is won by mesh decimation, material sharing, progressive loading and caching. None of it glamorous, all of it worth more than any AI feature.

Replace the seller’s call. Above roughly 5,000 euros, the close happens between two people. AI scores intent and drafts the follow-up. It does not handle the price objection or say “we will be on site Tuesday at eight.”

Same pattern throughout. AI takes the narrow, fast, repeatable work. The wide, slow, trust-bearing work stays with people and rules.

What buyers actually ask#

Nobody asks which model rendered the inspiration scene. They ask whether the pergola fits between the side return and the fence.

Nobody asks whether the intent score came from a transformer. They ask whether the dealer rings back today.

Nobody asks whether the translation was machine or human. They ask whether the PDF reads like a real document.

1. 508 Inquiries in 90 days Caribbean Blinds (UK pergolas). 40 percent of inquiries arrive pre-configured through the 3D configurator.
2. 8% Landing page conversion Spolding and Sons (UK garden rooms). 5,800 visitors, 475 configured leads in 8 weeks.
3. 30 Pergola sales in 90 days Nordin (Lithuania, louvered pergolas). 574 configured leads from a standing start.
4. 10× Close rate vs industry Industry pergola close rates sit at 5 to 8 percent. Sellers running a configured funnel typically run 10 to 20 percent.

Named numbers and full stories live in the case studies.

The 8 percent landing-page conversion at Spolding and Sons is the cleanest proof. 5,800 visitors, 475 designed and submitted. One in twelve, in a category where the floor is nearer one in fifty.

None of that came from AI gloss. The buyer saw the room, priced the room, and got a designed PDF.

The mistake in the current AI conversation is treating that as boring infrastructure and the experiments as the headline. Buyers read it the other way round.

The 3 second test#

One test survives every conversation about AI here. Open the configurator on a phone, tap to the pergola page, start a stopwatch. A believable rendered pergola inside three seconds.

Above three, mobile bounce climbs fast. Above five, half of phone visitors are gone. Above eight, nobody uses the thing. That number does not move because GPUs got smarter. It moves because someone did the unglamorous work.

What buys you time, in order of impact:

1. Smaller default meshes. First paint only needs what is visible from the default angle. Hidden faces and internal brackets can wait.
2. Shared textures. One anthracite material in memory, reused on posts, beams and rafters. AI compression helps. Reuse helps more.
3. Predictive preloading. Start loading mesh data while the buyer is still reading the hero. By the time they tap “design yours”, it is cached.
4. A server-rendered first frame. On the slowest phones, paint a flat CDN image of the default build and upgrade to 3D in the background.

None of those are AI features. They decide whether the buyer ever sees the AI features.

What 2027 probably brings#

A forecast, not a promise. Three things where the technology is real now and the product surface is not.

Photoreal instant variants. Today the configurator renders a parametric model and applies a material. Next comes a photoreal composited preview the buyer can swap in: set the angle, drop into photo mode, see a near-magazine render in their own garden at their own time of day. Not the spec drawing. The share-with-my-partner artefact.

Voice as a power-user shortcut. Not for the homeowner first. It lands in the dealer office, where a rep can talk a configuration through mid-call faster than clicking. Width 4.2, height 2.4, anthracite slats, add the lighting. The kitchen table comes later.

Pricing micro-models. A small model trained on your own pricing history learns the edge cases the rule library never caught, and suggests. The rep accepts, edits or rejects. Pricing stays supervised. It just gets better informed.

None of those is autonomous design, and none removes the configurator, the BOM or the seller. Buyers will experience all three as “the configurator got better” without noticing the AI. That is the right shape for it here: invisible where it works, named where it is still being sold.

Next reads: the pergola configurator guide for what a real one does today, 2D versus 3D for when 3D earns its keep, and the gallery for live builds.

The buyer designs, the price lands, the PDF arrives. AI makes each step a little sharper. The loop does not change.