MQLFlow

AI Quoting Automation Case Study

AI that interprets complex drawings, saving hours every day

We built The Marble Group a self-improving AI tool that reads varied drawings and calculates worktop sizes in about a minute, cutting costs and getting their customers quotes faster.

Topics

Results

  • 150

    kitchen plans a day to measure for quotes

  • Up to

    22 hours1

    a day gained for the team

  • 26

    AI models tested for consistent accuracy across varied plans

Services we delivered

AI consulting

AI consulting that established how AI could measure the client's plans reliably, and designed the solution, before they committed to a build.

  • Tested on a sample of the client's own plans before a significant investment, so the build started from evidence
  • Leading AI models compared for accuracy on PDF and image versions of the same plans
  • A report assessing each model's performance and the viability of the process
  • Solution design: how the tool would work, specified and scoped before the build

Custom AI development

We built the tool through custom AI development: AI that reads drawings, calculates worktop measurements and improves with every correction.

  • Vision models that recognise and interpret drawings, turning them into usable data for quoting
  • Accuracy that keeps improving with use, through a self-improving RAG loop that learns from the team's corrections
  • A system the client owns, built to order
  • Not tied to one AI provider, with automatic fallback

The brief

The Marble Group is the UK's leading stone worktop fabricator, with a team of more than 100 people. Around 150 kitchen plans reach them every day, and before a customer can be quoted, someone has to extract the measurements from the plan and calculate the worktop sizes, which at that volume is time-consuming work.

It's skilled work, not simple data entry. Plans come in every format and style, some with 3D renders, and details often have to be inferred, so results could vary from one person to the next. Customers expected a fast reply, and the general AI tools they had tried couldn't do it reliably. They wanted a custom AI system of their own, accurate at that volume and able to connect to their other tools. Our proof of concept investigated what it would take for AI to do the job reliably on their own plans, and we then designed and built the system.

MQLFlow built a custom AI system for our business from scratch, and it's honestly game-changing. No other developer we've worked with has delivered work of this quality so efficiently. I'm excited to keep working with MQLFlow to expand AI across the business.
Theo ConstantiTheo ConstantiDirector, The Marble Group

Considerations

We designed the tool around what mattered most to The Marble Group.

  • Owned, not rented. They wanted a system of their own rather than a subscription to someone else's, and nothing on the market exactly matched their needs, so it was designed and built to order.
  • Every kind of plan. Plans range from big retailers' commercial design packages to customers' own hand drawings, as PDFs and images.
  • Expert knowledge to capture. Knowing where to measure from, and which runs are worktop, takes experience, so the system had to learn from the team's own guidelines and corrections.
  • The team stays in control. The AI tool measures, and the team reviews each result and sets the price.

Approach

We worked out how AI could do the job reliably before any build, then built the tool around the team's feedback on real plans.

  • Proof of concept first. Working with their team on the instructions, we tested several leading AI models on a sample of their plans, as PDFs and as images, to find what gave consistently accurate results across the whole sample, not just on one plan.
  • A report that set the direction. Our report set out how each model performed and what the process would need, and it shaped the design of the custom build.
  • Built with the team. The team used the tool on real plans as it was developed, and their feedback decided what was added next.

Solution

A custom AI tool that reads kitchen plans, extracts the measurements and calculates the worktop sizes, with a self-improving feedback loop.

  • Interprets each drawing. Vision models recognise what each drawing shows, whatever its style, and infer details that aren't marked.
  • Every piece sized. Worktop runs, upstands, splashbacks and cut-outs for sinks, hobs and taps, each with a confidence score.
  • Uses 3D renders as a guide. Where a plan comes with 3D renders, it uses them for details the plan doesn't show, such as waterfall ends, upstands and full-height units.
  • Learns from every correction. The team's guidelines and corrections go into a knowledge base that the AI draws on for each new plan, so it improves with use.
  • Splits worktops for cutting. Each worktop is divided into pieces that fit the chosen stone's slab size.

Result

Each plan now takes about a minute, down from around ten when done manually. As the tool rolls out to their 150 or so plans a day, the team gains back hours of skilled work, and customers receive their quotes faster.

  • Faster quotes for customers. The team checks figures instead of working them out, so quotes reach customers sooner.
  • Consistent results. Every plan is measured the same way, whoever checks it, and the team's know-how stays in the business.
  • Ready to connect. The tool runs as its own service with an API, so it can feed their quoting system and other tools.
  • Owned by the client. The tool and the data belong to The Marble Group, and the data is never used to train an AI model.

Notes

  1. 1.Estimate: around ten minutes a plan manually and about one with the AI tool, a saving of up to nine minutes on each of around 150 plans a day, or 22.5 hours. ↩

Is someone on your team checking images or documents by eye?

Book a call. Tell us what your team checks, and we'll show you how AI could take it on, tested on your own files first.

Book a call

More case studies

All case studies