The AI configurator features that earn their keep (and the ones that do not)

A pergola buyer in Bristol opened a configurator on a Tuesday evening last month. She tapped through the size, picked anthracite louvres, added integrated lighting, and forwarded the priced PDF to her husband with one sentence in the body of the email. This is the one. Can we book the install in June. He read it on his commute the next morning, replied yes, and the dealer rang at lunchtime to take the deposit. None of that conversation needed AI. The configurator did the job that closes the deal. AI helped inside the loop in three places she never saw, and one place she did.

The conversation about AI in outdoor configurators changed in the last twelve months. In 2024, the question was will AI replace the configurator. In 2025, it was what AI features will the next generation ship. In 2026, having watched these features land on real dealer floors, the honest question is the one customer success keeps coming back to. Which AI features do customers still use after the first 90 days, and which ones get switched off.

We turned on three AI features at launch. By month three we had one of them still on. The other two were generating buyer confusion faster than they were generating leads.

The pattern across the customers we work with is consistent. AI features cluster into two camps. The first earns its place because it does a narrow job better than a person can do it at the same speed. The second is impressive on a demo screen and quietly turned off in production once the sales team gets tired of cleaning up after it. The job of this post is to draw that line clearly.

The four AI features that earned their place#

Each is a narrow job that AI does faster than a person can do it without losing accuracy. Each is now standard kit on the configurator stack across the customers we onboard.

Photo-to-draft design at the top of the funnel. The buyer uploads a phone photo of their patio. A vision model identifies the empty space above the dining table and the rough scale of the surrounding architecture. A first-pass configuration drops onto the scene with sensible defaults. The buyer sees their own garden with a pergola in it before they have touched a single dropdown. The configurator then takes over and ends in a real priced PDF. The AI did not design the pergola. It started the conversation visually instead of with a contact form. Buyers who reach the priced PDF from a photo upload convert at a higher rate than the ones who start cold, because they have already committed to a draft they recognise as theirs.

A two-minute qualifying chat before the configurator opens. This one earned its place by halving the time the sales team used to spend on the wrong inquiries. The chat asks three or four questions in plain language. What are you trying to solve. How big is the space. When are you hoping to install. Are you the homeowner or the contractor. The answers prefill the configurator and attach a short context block to the lead. The sales rep opens Monday morning to a configured PDF with a one-line note. Homeowner, Bristol, 4x4m terrace, wants installed by July, selected anthracite louvres. The chat is not selling. It is qualifying. The moment it tries to sell, the buyer drops out.

A plain-language summary of the configured quote. The configurator emails the PDF. Underneath that, AI writes a three-sentence summary in the email body. You configured a 4 by 5 metre louvered pergola in anthracite with integrated LED lighting and side screens. The total is 14,820 euros including installation. The install is scheduled for the second week of June if you confirm by Friday. That summary is not for the buyer who built it. It is for the partner or decision-maker who never opened the configurator and is about to be shown the email on a phone screen. The PDF closes the deal with the buyer who designed it. The summary closes the deal with the second person whose approval the buyer needs.

Next-add-on suggestion based on the configured spec. Once the structure is built, the configurator surfaces one or two add-ons that the spec is missing rather than a generic upsell. A pergola without lighting gets a lighting suggestion. A configuration with screens but no heater gets a heater suggestion. A manual louvre with motorised accessories gets a motorisation prompt. The model reads the buyer’s own design and points out the gaps the buyer would notice on install day. Customers click on these suggestions roughly twice as often as on a generic “customers also bought” carousel.

These four share a shape. Each is a narrow task at one step of the funnel. Each runs in the background and either succeeds invisibly or stays out of the way. None tries to be the configurator. None tries to be the salesperson. The buyer never says what a clever AI feature. The buyer says that was easy.

If you want the longer field view of where AI shows up across pergola design specifically (inspiration imagery, garden-photo composition, intent scoring, multilingual dealer copy), the ai pergola design post covers it.

What gets pitched, sold, and quietly turned off#

Three features show up in almost every AI configurator pitch in 2026 and get switched off inside the first 90 days more often than they stay on. Naming them is the most useful part of this post.

Fully autonomous AI design. The promise is that the buyer types one sentence and gets a finished pergola. In practice, two things go wrong. The autonomous design picks combinations the factory cannot ship: a roof profile that does not exist in the catalogue, a screen colour the supplier discontinued, a finish with lead times the install crew cannot meet. And buyers do not trust a design they did not build. They spend twelve minutes adjusting an autonomous draft to look exactly like what it started as, just to feel like they made the choices. The shorter route is to let them build it with smart defaults, which is what the draft-from-photo feature above does. Autonomous design demos well. It generates returns and reconfigurations on a real dealer floor.

Voice-driven configuration for the homeowner. The technology is real, the speech recognition is accurate enough. The problem is the room the buyer is in. A homeowner evaluates a 15,000 euro pergola on the sofa with their partner, or at a kitchen counter while making dinner. None of those rooms is one where you talk out loud to a configurator about post finishes. Voice will land first inside the dealer office where the sales rep is alone at a screen. For the buyer at home in 2026, voice is a 2028 feature being marketed in 2026.

An AI chatbot that tries to close the sale. Distinct from the qualifying chat at the top. This one shows up as a popup after a few minutes inside the configurator and tries to do the seller’s job. It offers discounts. It quotes lead times. It promises install dates. The buyers who engage with it accept the answers as commitments. The sales team then spends the next week explaining why the install date the chatbot promised is not the date the crew can deliver. The lifetime of this feature in production is usually about two months.

The common pattern across these three is that they try to replace a person, not assist one. The four features above assist a person at a narrow task. The three here try to remove the person and end up making more work for them. That distinction is the cleanest test we know of for whether an AI configurator feature will earn its keep.

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

Named numbers and full stories live in the case studies.

These are not AI numbers. They are configurator numbers. AI helps inside that loop in the four places above, and the conversion rates move by a few percentage points each. The headline is still the basic four jobs the configurator does. Render the build, apply the rules, price the spec, send the PDF. AI is the polish. The configurator is the product.

The Caribbean Blinds case study is the cleanest version of this story. The team turned on the qualifying chat and the plain-language summary at launch. They left autonomous design off because the install crew refused to quote against AI-generated specs. The qualifying chat is doing more work in the funnel than any other single feature. It is two minutes long. It asks four questions. It is not impressive on a demo. It is the highest-impact AI feature in their stack.

What buyers ask about AI in practice#

Walk a buyer through their decision and the AI question never comes up the way the vendor pitches it. They do not ask which model generated the inspiration scene. They ask whether the pergola will fit between the side return and the boundary fence. They do not ask whether the intent score was computed by a transformer. They ask whether the dealer is going to ring back the same day.

The mistake the AI conversation is making in this category is treating the basic configurator jobs as boring infrastructure and the AI experiments as the headline. The buyer reads them in the opposite order. The infrastructure is the product. The AI is the polish.

AI features earn their place when they make an existing job faster. They do not earn their place when they invent a job the buyer did not know they wanted. The buyer wants to see their pergola, in their garden, in three seconds, on their phone, with a real price. Every AI feature is judged against whether it makes that core experience faster, calmer, and more accurate. The ones that do, stay. The ones that distract, go.

A short forecast for 2027#

Three forecasts, each grounded in technology that is real in 2026 and product surfaces that are not quite ready.

Photoreal compositing inside the configurator window. Today the buyer toggles between the parametric 3D model and a separately generated photoreal scene. By next year, the photoreal layer renders on top of the parametric model in the same view, in real time, on a phone. Same model underneath, magazine-grade render on top. That render is the artefact buyers will forward to partners more often than any other.

Multilingual configurator copy that updates without a rollout. A copy change in the UK pushes to 30 locales overnight, with the dealer in each country approving rather than translating. The configurator stops being an English-first product with translation overhead and starts being multilingual by default.

Pricing micro-models that learn the edge cases. A small model trained on the manufacturer’s own historical pricing surfaces edge-case suggestions to the sales rep. This dealer in this region with this product at this time of year usually closes at 6.8 percent above this margin band. The rep accepts, edits, or rejects. Pricing stays human-supervised. The base just gets smarter.

None of these is autonomous design. None removes the configurator, the BOM, or the seller. All three are quieter upgrades the buyer will experience as the configurator getting better, without ever noticing the AI underneath.

If you want the wider view across pergola, veranda, garden room, and outdoor kitchen, the live configurator gallery shows builds running today. The pergola configurator is the most common starting point for outdoor living teams new to this stack.

The buyer designs. The price lands. The PDF arrives. AI sits inside that loop, making the right steps a little smarter and the wrong steps a little less likely. The loop itself does not change. The features that earn their place are the ones that make the loop calmer for the buyer and quieter for the sales team. Everything else is a demo waiting for a different decade.