Estate agency
Parallax Property — +82% viewing requests
Buyers were finding Parallax listings on the portals and never on their own site. When they did arrive, search took eleven seconds to return a result.
All case studies
The situation
Parallax Property is an independent estate agency competing against national portals for the attention of its own vendors' buyers. Their listings appeared on the portals within minutes of going live and on their own site within minutes too — but almost nobody used their site, and the agency was paying portal fees for leads on properties it was already marketing.
The measurable reason was speed. Listing pages were serving original camera-resolution photography, unresized, twelve to twenty images per property. The median listing page weighed 14MB and took just over eleven seconds to become usable on a typical mobile connection. Most people never saw it.
The second reason was search. Filters triggered a full page reload, lost their state on the back button, and could not be shared. Property search is inherently a back-and-forth activity done across several sessions and usually with a partner, and the tool made every one of those behaviors painful.
What the audit found
Fourteen megabytes per listing
Photography was uploaded straight from the camera and served at full resolution to every device. A buyer browsing six properties on a commute was downloading close to 90MB — often into a data cap, always into an eleven-second wait per page.
Search that forgot itself
Filters were stored in server-side session state. Pressing back returned an unfiltered list, so refining a search meant redoing it. Recordings showed people re-entering the same criteria four and five times in a session before giving up.
A form that asked the wrong question first
The viewing request opened with mortgage status and proof of funds. Those are reasonable questions for an offer and unreasonable ones for someone who wants to look at a house on Saturday.
Nothing worth sharing
Listing URLs were query strings with no preview data. Pasted into a message they appeared as a bare link, so buyers screenshotted portal listings to their partners instead — handing the relationship back to the portal.
What was broken
- Listing pages loaded 14MB of unresized photography on every visit.
- Filters reloaded the whole page and reset on the back button.
- The viewing request form asked for a mortgage status before a name.
- No listing had a shareable link that previewed correctly in messages.
What we changed
Images sized for phones
Responsive image sets and lazy loading cut the median listing page from 14MB to 900KB without any visible loss in quality.
Filters that hold their state
Search runs instantly against the URL, so results survive the back button and can be sent to a partner as a link.
Two fields to book a viewing
Name and phone. Everything else is asked on the call, once there is an actual conversation to have.
The technical detail
The specific decisions behind the numbers, in plain language.
An image pipeline sized per device
Uploads are processed into multiple widths in modern formats and served against the device's actual viewport, with the first image prioritized and the rest lazily loaded. Median listing weight fell from 14MB to 900KB with no perceptible quality difference on screen.
Search state lives in the URL
Every filter is encoded in the address. Results are shareable, bookmarkable, and survive the back button. Search runs against an index that returns in under 120ms, so filtering feels instant rather than transactional.
Two fields to book a viewing
Name and phone number, with the rest handled on the call. Qualification questions were moved to where a human can explain why they are being asked.
Rich link previews
Each listing generates its own preview image, price and headline, so a link sent to a partner arrives as a proper card. Shared links became a measurable acquisition channel within the first month.
Saved searches and alerts
Buyers can save a set of criteria and be notified when a match is listed, which puts the agency's own site ahead of the portal in the one moment that matters — the day a property goes live.
How it ran
Audit
Performance traces per template, search behavior analysis, portal comparison on identical listings.
Design
Search and listing templates, viewing flow, share cards.
Build
Image pipeline, search index, URL state, alerts and feed integration.
Measure
Channel-level reporting and two rounds of search refinement.
How we measured it
Tracked for ten weeks post-launch against the equivalent period before, with portal-referred traffic excluded so the comparison isolated the agency's own channel.
Viewing requests rose 82%. Median listing load time fell from 11.2s to 1.6s. Pages per session went up 2.4×, and the bounce rate on listing pages fell from 68% to 31%.
Direct and shared-link visits grew from 6% to 23% of total listing views over the same period. The agency's portal spend was subsequently reduced without a fall in total viewings booked — the effective return on the project, and the number the director actually quotes.
viewing requests
listing load, down from 11s
pages per session
“We were paying the portals to do the job our own site should have been doing.”
What we took from it
- In property, page weight is a commercial metric, not a technical one.
- If a search cannot be shared or reopened, it will be abandoned.
- Competing with a portal means being faster than it, not looking like it.