DeXtra Specialized Cleaning is a commercial cleaning company in Somerville, New Jersey, serving facilities across New Jersey, Pennsylvania, and New York. We built them a quote engine with three lanes: an instant quote from square footage, a photo quote that an AI vision model reviews, and a custom quote for complex jobs that need a person. What makes it work is that the form follows how customers actually describe a job: room by room.
Key takeaways
- DeXtra's website gives customers three ways to get a quote: Quickest (instant math), Most Accurate (photos reviewed by AI), and More Detailed (custom scope).
- Multi-room jobs are captured one room per page, and each room's name, service, size, floor, condition, and photos land in the sales record and a scope sheet PDF.
- The photo lane runs on GPT-4.1 as of September 10, 2026, for steadier reads of floor type and condition.
- Every quote states its confidence as High, Medium, or Low and explains what can move the price.
- Test submissions stay inside SimplySync, so DeXtra's team only hears about real leads.
What did DeXtra need?
For a cleaning company, speed to quote is speed to revenue. A property manager who waits days for an estimate books whoever answered first. DeXtra was quoting every job by hand: phone calls, site visits for jobs that did not need one, and estimates written up manually.
The goal was simple jobs that quote themselves, and complex jobs that reach a person fast with everything already written down.
What did we build?
The quote engine lives on dextraclean.com, a Next.js website, and feeds the SimplySync sales pipeline DeXtra's team works from. The build order:
- Three lanes on one quote page. Quickest uses square footage and published service rates for an instant range. Most Accurate asks for photos and has an AI vision model review them. More Detailed collects areas, cleaning frequency, and hours for a custom scope.
- Room-by-room capture. The photo lane asks how many rooms first, then shows one page per room with a required room name, service, size, floor type, condition, and that room's own photos.
- A square-footage helper. Customers can enter an exact number, length times width, or a range, because most people do not know their square footage.
- A structured sales record. Every submission creates a contact note with the room-by-room breakdown and photo links, an opportunity named with the room count, and, for multi-room or custom jobs, an internal scope sheet PDF.
- Team alerts that match the lane. New estimates and walkthrough requests both alert the team with the customer, the range, and the service.
- A test guard. Submissions marked as tests alert only SimplySync, never DeXtra's team or any real contact.
- A website content engine. A scheduled job publishes cleaning guides to DeXtra's blog on a rotation of topics.
What worked?
Letting customers pick their own speed. Some buyers want a number in a minute; others want accuracy. Labeling the lanes by outcome (Quickest, Most Accurate, More Detailed) instead of by method made the choice obvious.
Room-by-room data. On September 11, 2026, a production test with two rooms produced a per-room contact note, per-room photo links, a scope sheet PDF, an opportunity named "2 rooms," and five files attached to the quote. DeXtra's team no longer has to call back to ask what was in each room.
Clear confidence wording. Customer emails describe estimate confidence as High, Medium, or Low. Instant quotes are labeled as instant math, and every quote explains what can move the price: measured square footage, floor condition, and access.
A person for the hard jobs. Large or unusual jobs route to a free on-site walkthrough, so the AI never has to guess on the quotes that matter most.
How did we refine it?
After launch, we put the form in front of testers on their own phones and used what they told us to sharpen it. Each round made the engine easier to use and more useful to DeXtra's team.
1. Photos that stack up, shot by shot (September 11, 2026). Customers can now take photos one at a time with their phone camera, and each shot adds to that room's set until the lane's limit. We checked it by adding five single photos and watching the count go 1, 2, 3, 4, 5, then stop at the cap.
2. Easy photo editing on phones. The remove button now sits in plain view on small screens, so a customer can swap a blurry photo with one tap.
3. Rooms as their own pages. The photo lane moved from one long form to one page per room. That gave the team a scope sheet PDF and per-room photos instead of a single pile of pictures.
