Quoting custom furniture projects has always been part art, part science, and part educated guessing. You look at the design, mentally walk through every cut, every joint, every finishing step, and then you put a number on it. Sometimes you’re right. Sometimes you’re not.

I’ve been doing this long enough that my gut estimates are usually within 10-15% of actual cost. But “usually” isn’t always. Last year I underquoted a set of built-in bookshelves by about $1,800 because I underestimated how much Victorian ash the client’s design would actually consume once I accounted for grain matching and defect avoidance. That’s a painful margin to absorb on a $7,000 job.

Over the past six months, I’ve been experimenting with AI-assisted tools for material estimation, and the results have been genuinely interesting.

The Problem With Traditional Quoting

The traditional approach is a spreadsheet. You list every component, estimate the board metres of timber, the sheet goods, the hardware count, and the finish litres. Then you add a 10-15% contingency buffer for offcuts, mistakes, and material that doesn’t pass inspection.

It works, but it has blind spots. The biggest is yield loss. Cutting components from a 2400mm x 1200mm plywood sheet, the theoretical versus actual material usage can differ by 20% or more depending on the cut layout. A good cabinetmaker knows this instinctively, but “instinct” isn’t something you can hand to an apprentice.

The other blind spot is timber variability. When I order 50 board metres of spotted gum, I might reject 15% for colour inconsistency or grain issues. That rejection rate varies by species, supplier, and season.

Where AI Actually Helps

The AI tools I’ve been testing don’t replace the quoting process. They make specific parts of it faster and more accurate.

Cut list optimisation is the most immediately practical application. You feed in your component dimensions and sheet sizes, and the AI generates optimised cutting patterns that minimise waste. This isn’t new as a concept, software like CutList Plus has been doing nesting and optimisation for years. But the newer AI-driven tools handle irregular shapes and grain-matching constraints much better than the older algorithmic approaches.

On a recent kitchen project with 47 individual plywood components, the AI-optimised cut layout saved me about 1.5 sheets compared to what I would have achieved manually. That’s around $120 in material at current 18mm birch plywood prices, which isn’t transformative for a single project but adds up across a year’s worth of work.

Historical cost analysis is where things get more interesting. I’ve been feeding my past project data into an AI system, including material lists, actual versus estimated usage, and final costs. After about 30 projects’ worth of data, the system started producing material estimates that are consistently tighter than my own. Not by a huge margin, maybe 5-8% more accurate, but enough to matter.

The system picks up patterns that I process unconsciously but can’t articulate easily. For instance, it identified that my spotted gum rejection rate is higher in summer deliveries than winter ones, probably because the timber dries more aggressively during transport. I knew that was true in a vague way, but the AI quantified it at 18% summer versus 11% winter. That’s the kind of data-driven insight that’s worth having.

The team at Team400 wrote a solid overview of how small businesses are applying AI to operational problems like this. It’s not furniture-specific, but the principles around training models on your own business data apply directly.

What Doesn’t Work Yet

AI material estimation falls down with unusual materials. I recently quoted a piece using reclaimed hardwood from a demolished woolshed. There’s no historical data for yield on 120-year-old ironbark with nail holes and weathering. Experience is still irreplaceable there.

The tools also struggle with complex joinery where waste is geometry-dependent. And the setup time isn’t trivial. I spent 15-20 hours on data preparation before I got useful results.

Is It Worth It for a Small Workshop?

For a solo maker or a two-person shop doing fewer than 50 projects a year, probably not yet. The accuracy improvement is real but modest, and the setup investment is significant relative to the time saved.

For a workshop doing 100+ projects annually with a team handling quoting? Yes. The consistency alone is valuable. When I’m not available to quote a job, my apprentice can produce a material estimate that’s within a few percent of what I’d calculate. That wasn’t possible with the old spreadsheet-plus-intuition approach.

The technology is moving fast. The tools I’m using today are meaningfully better than what was available twelve months ago, thanks in part to developments from Autodesk Research in manufacturing optimisation. I expect that within another two or three years, AI-assisted quoting will be standard practice for any professional custom furniture operation.

For now, I’m using it as a second opinion on every quote. My gut estimate first, then the AI estimate, then I compare. When they disagree by more than 10%, I dig into why. That process has already prevented two significant underquotes this year. That alone has paid for the software subscription many times over.