AI lead scoring for auction houses

An auction specialist's time is the most expensive resource in the building. Every minute they spend on a low-value enquiry is a minute they're not on a high-value one. The trouble is, you usually can't tell the difference until you've already spent the time. AI lead scoring changes that.

This is a guide to how lead scoring works for auction houses, what it's good at, where it falls down, and how to set it up so it actually helps rather than getting in the way.

What lead scoring is and isn't

Lead scoring assigns a number, typically 0 to 100, to every incoming enquiry. Higher numbers mean a more promising seller. The score is not a valuation. It's not predicting the hammer price of the item. It's predicting how likely this seller is to consign something that's worth your specialist's time today.

The score is based on signals from the submission itself, not from the seller's history. We don't need a CRM. We need the enquiry text, the photos, and the contact details.

What signals matter

The signals fall into roughly four groups, in order of how much they actually predict:

1. Description quality

The single strongest signal is how the seller describes their item. A long, specific description with named makers, dates, dimensions, and provenance suggests an engaged seller who has done their homework. A two-line description that says "old painting, want to sell" suggests someone who hasn't.

This isn't about literacy. It's about engagement. Sellers who write a paragraph are sellers who care about getting a good outcome, which usually correlates with caring about getting the right house. They'll consign when you reply well.

2. Photo evidence

Three or four clear, well-lit photos signal a serious seller. One blurry photo of a corner of an object signals someone who's not really committed. Modern AI vision models can also assess the photos themselves: is this a recognisable category, is the item in reasonable condition, are there hallmarks visible.

3. Category and value-indicating keywords

Some words in a description are statistically associated with higher-value items. Hallmarked. Signed. Named maker. Provenance. 18th century. Inherited. Estate. These don't guarantee value, but they correlate with it. Conversely, certain phrases ("found in attic," "might be worth something") correlate with hopeful but low-value submissions.

4. Contact completeness

A small but real signal: sellers who give you their phone number and postcode are more committed than those who give only an email. They're more likely to want a real conversation.

Why custom prompts matter

Generic lead scoring is fine. Custom lead scoring is much better. Every house has different priorities. A Mayfair jewellery specialist scores differently than a regional generalist. A house that focuses on contemporary art doesn't want to be told that an "original Victorian oil" is high-priority.

Modern AI lets you write your priorities in plain English. We let our customers specify a custom scoring prompt. Real examples from our customers:

We specialise in fine jewellery and watches over £1,000 estimate. Score these higher. Items under £500 estimate are low priority. Furniture is rarely worth our time unless it's by a named maker. Always score Asian works of art highly because we have a buyer pool.

Our sales are heavily weighted to early 20th century European decorative arts. Score Art Nouveau, Art Deco, and Bauhaus items higher. Mid-century modern is in our wheelhouse but only signed pieces. Generic Victorian furniture is low priority.

Both of these prompts dramatically improved scoring accuracy for those houses. The AI doesn't need a database of every Lalique vase ever made. It needs to know what your house is good at and adjust accordingly.

How to use the score in practice

A score on its own is useless. What matters is how it changes specialist behaviour. Three concrete uses:

Sort the queue

Specialists open the platform and see their enquiries sorted by score, highest first. They work top to bottom. Within the first hour of the day they've handled the high-quality leads that arrived overnight. The 50-point enquiries can wait until lunch.

SLA differentiation

Some houses use score to set internal response time targets. 80+ scores get a one-hour response window. 50-79 scores get four hours. Below 50, same-day. This isn't about giving low-scoring sellers worse service; it's about ensuring high-quality leads are never the ones that fall through.

Routing decisions

Above a certain threshold, you might route to your most senior specialist. Below it, to a junior. The senior's time is reserved for the leads where it matters most.

Where it fails

Lead scoring isn't perfect. We've seen it get things wrong in predictable ways.

Genuine sleepers score low. Sometimes a one-line enquiry with a single mediocre photo turns out to be a Cartier diamond bracelet. The seller didn't know what they had, so the description doesn't reflect it. The score will be low. This is unavoidable, and it's why you should never use score as the only signal. Even low-scoring enquiries deserve a human eyeball.

Specialist subjects can confuse the model. A description that mentions an obscure 17th-century Dutch silversmith might not register as "value-indicating keywords" because the model hasn't seen the name often enough. Custom prompts help, but they don't fix this entirely.

Bad sellers can game the score. If you publish exactly how scoring works, sophisticated sellers can write descriptions that game it. This is a small problem in practice (most sellers aren't writing four enquiries to four houses optimised for AI scoring) but it's real.

How to start

If you're considering lead scoring, do it in stages.

Week one: turn it on with default settings. Don't change behaviour yet. Just see what scores get assigned to your existing enquiries and whether the scores feel right intuitively to your specialists.

Week two: write your custom prompt. Sit with two or three specialists and ask them what makes a lead high or low priority for your house. Translate that into a paragraph the AI can read.

Week three: start using score to sort the queue. Don't change SLAs yet. Just see if the highest-scoring enquiries genuinely are the ones that turn into the best lots.

Month two: introduce SLA differentiation if the scoring is reliable.

Month three: review. Look at lots that hammered well over the last quarter and trace them back to their original enquiry score. Adjust the prompt if the model is consistently underrating a category that's actually valuable for you.

The honest version

Lead scoring is the kind of feature that sounds boring in a demo and changes how your house works in practice. It's not magic. It's a sorting tool. But sorting is most of the job. Specialists who used to spend the first hour of every day reading every enquiry now spend that hour replying to the five they actually need to.

That hour, multiplied by every specialist, multiplied by 250 working days, is real money. It's also less burnout, fewer missed lots, and a faster cycle from first contact to consigned lot. For most houses, it's the highest-leverage change you can make this year.