Virtual Survey Moving: What Actually Changes When You Automate the First Look

title: Virtual Survey vs In-Home Estimate: What Changes When You Automate the First Look slug: virtual-survey-moving-automate-first-look category: Case study excerpt: How MoveAI replaces the initial home visit with a photo-based estimate — and where that tradeoff actually holds up. stack: Next.js, Supabase, Telegraf, OpenRouter (Claude Sonnet, multimodal), Vercel

Most moving companies still price a job the same way they did twenty years ago: someone walks through the apartment, eyeballs the furniture, and calls in a number. It works, but it doesn't scale past the number of estimators you can put in a car.

"Virtual survey" is the industry's answer to that — and the term gets used loosely. Sometimes it means a live video call with a human estimator. Sometimes it means a customer filling in a web form with room counts. Neither is really automated; the first still needs a trained person on the other end, the second produces data too coarse to trust without a follow-up call.

MoveAI sits closer to full automation than either: a customer photographs their furniture, and a multimodal model (Claude Sonnet via OpenRouter) reads the photos against a price table to produce a cost estimate directly — no call, no form, no human in the loop for the initial number.

Why photos, not video Video carries more information per interaction, and it's the obvious next step once you've committed to a virtual-survey approach — several competitors in this space (Yembo, Drifted.ai, HomeSurvey.ai) are built around exactly that. We didn't go that route, for a plain reason: photos are something a customer already knows how to produce well on a phone, and they're cheap to process. Video requires the customer to narrate or pan competently, and it requires a processing pipeline that photos don't. For the volume and price point MoveAI targets, that complexity wasn't buying enough accuracy to be worth it.

That's a real limitation, not a hidden one. A stack of photos won't catch what a video walkthrough might — a closet packed floor to ceiling that never gets photographed, for instance. The product doesn't try to be a full replacement for an in-home estimate on every job; it's built for the segment of moves — mostly apartment-sized, mostly straightforward — where photo-based estimation is accurate enough to book with confidence, and where the alternative (a scheduled home visit) is the actual friction costing the mover the lead.

Why this is a two-sided product, not just a calculator The estimate itself is only half of what MoveAI does. The other half is what happens after: the client-facing Telegram flow that collects the photos and produces the quote, and a separate carrier portal where a moving company's own staff sees incoming requests, confirms jobs, and manages them — plus an admin layer for oversight across both sides.

That structure exists because the estimate is not the product a moving company actually wants. It's the input to the product they want, which is fewer unqualified calls and less staff time spent on the front end of a job that might not book. A calculator that just spits out a number to the customer doesn't solve that; a flow that hands the moving company a pre-qualified, priced request does.

Where this fits for a small moving company This is the part that matters for anyone evaluating "moving software" as a category rather than a single vendor: the honest pitch isn't that AI replaces the estimator. It's that AI replaces the first call — the one that used to happen just to find out whether a lead was worth a real estimator's time at all. For a small moving company running a handful of trucks, that first call is disproportionately expensive relative to what it produces, because most of it is just gathering information a photo could have given you directly.

The tradeoff moving companies are actually making when they adopt something like this isn't "automation vs. no automation." It's deciding which slice of jobs is safe to price from photos alone, and routing everything else — the multi-floor move, the piano, the storage unit nobody photographed — back to a real in-home estimate. Get that routing wrong in either direction and you either lose accuracy on complex jobs or keep paying for home visits on easy ones.

That's the actual design problem behind a virtual survey tool. Not "can a model read a photo of a couch" — it can — but where the line sits between what a photo can safely price and what still needs a person to walk the space.