Case Study: AI Cut Catalog Lookups From 7 Days to Seconds

A salesperson stands in front of a client who wants that piece — the one in the photo on their phone. And the honest answer the salesperson can give is: “Let me get back to you next week.”

That’s where this story starts. Not with AI. With a lost week, and the deals that quietly die inside it.

The why first, because it’s the whole point: speed isn’t a nice-to-have in sales — it’s the sale. Classic HBR research on lead response found that contacting a prospect within five minutes makes you vastly more likely to win them than waiting even thirty. When a client has to wait seven days to find out whether you make what they want, you haven’t given them a quote. You’ve given them time to call someone else.

What was the problem?

The product was beautiful. The process was duct tape.

This is a manufacturer of high-end outdoor furniture and cabanas — real craftsmanship, serious clients, a deep catalog. But finding the right SKU was a human scavenger hunt. A client sends a photo or a vague description. A salesperson forwards it around. Someone who “knows the catalog” digs through files. Inventory and order data lived in one system, customer history in another, prospect data in a third — three systems that stored everything and connected nothing.

The result: 7+ day turnarounds to answer a question that should take a minute. Catalog knowledge trapped in a few people’s heads. And a quiet, ongoing tax on every deal — the gap between “I want this” and “yes, we make it, here’s the price.”

The conventional read is “they need better software.” That wasn’t it. They had software. What they didn’t have was a way to ask their own business a question and get an answer back.

What did we build?

Two things, and the order mattered.

First, we connected AI to the systems of record — their ERP, CRM, and prospecting tools — read-first, then narrow approved writes. The principle here is one we hold to on every engagement: an AI that can read your business is useful; an AI that can quietly rewrite your system of record without a human in the loop is a liability. So we started with read access and earned our way to action.

Then we built the part that changed the room — call it visual product search. Using retrieval over their product catalog combined with image embeddings, we made it so a salesperson can upload a photo of the piece a client wants and get matched SKUs back in the same conversation. No scavenger hunt. No “let me check.” The catalog, finally, became answerable.

There’s a framework worth naming here, because it generalizes: store-versus-execute. Most business systems are excellent at storing data and useless at executing on it — the data sits in the ERP, and a human still does all the work of turning it into an answer. The leverage isn’t more storage. It’s putting an execution layer on top of the data you already have. That’s what this build was.

A caveat, because we’d rather you trust the next thing we say: this worked because the catalog was rich and the data, while siloed, was real. On thinner or messier data, the first phase is cleanup, not magic. Most of the time, the honest first deliverable is “let’s make your data answerable” — and that’s a feature, not a delay.

What were the results?

• 7+ day catalog turnarounds → seconds. The lookup stopped being a project and became a question.
• Major back-office efficiency gains across the connected workflows — the manual clicking, cross-referencing, and re-keying between systems largely went away.
• Catalog knowledge, unlocked. It no longer lived in two people’s heads. Anyone on the team could ask.
• More room to sell. When the answer is instant, the conversation stays alive — and conversations that stay alive convert.

We won’t pretend a single number tells the whole story. But the shape of it is unambiguous: the same team, the same catalog, the same clients — and a sales motion that stopped leaking a week at a time.

Could this work for your business?

Ask yourself three questions.

Do your systems store more than they execute — i.e., is your team the integration layer between tools that don’t talk? Is there a moment in your sales or ops process where a customer waits days for an answer that should take minutes? And is the knowledge that unblocks them trapped in a few people’s heads?

If you said yes to two of those, there’s leverage sitting in your business right now. It doesn’t require a year-long transformation or a data-science team. It requires connecting what you already have and putting an execution layer on top.

That’s the work. Not buying more software. Making the software you own finally answer the question in front of you.

Your catalog already knows the answer. The only thing worth fixing is how long it takes to ask.