Case Study: UK Flooring RFQs in 90 Seconds

The bottleneck in a UK quote-to-cash workflow is almost never the quote itself. It is the time it takes to get the inbound RFQ into the system in a form the team can actually quote against. When that step takes 20 minutes per email, and 10 emails arrive every day, the maths starts to bite quickly.

This case study walks through what happened when a UK commercial flooring business replaced their manual RFQ data entry with an AI-powered processing pipeline. Inbound emails, spreadsheets, PDFs, and photos go in. Structured projects in Airtable come out. The brief was to clear the quote backlog without adding office staff. The result was RFQ processing time cut from 20 minutes to under 90 seconds, and 12+ hours of weekly staff capacity unlocked.

What was the RFQ processing bottleneck?

Designs4You receives 10+ inbound Requests for Quotes (RFQs) every day from commercial clients. Each RFQ arrives via email, often with attachments in varying formats: Excel spreadsheets, PDFs, photos of site conditions, and embedded images. Some include structured data tables. Others are free-text emails with photos attached.

Previously, a member of the office team would open each email, read through the content, identify the customer, extract the key details (site address, contact information, work description, order references), and manually type all of this into their project management system. This took approximately 20 minutes per RFQ when accounting for the full process of reading, interpreting, cross-referencing the customer database, and entering the data.

With 10+ RFQs arriving daily, this manual processing consumed hours of staff time. At peak periods, the backlog meant quotes could take up to 2 weeks to send, simply because the team could not process the incoming requests fast enough. This was not a speed problem. It was a capacity problem that was throttling the company's growth.

Where the time went	Per inbound RFQ
Reading the email and attachments	5 minutes
Cross-referencing the customer database	5 minutes
Typing project details into the system	7-8 minutes
Quality checking the entry	2-3 minutes
Total	~20 minutes per RFQ

For wider context, the UK Office for National Statistics consistently reports that admin and data entry absorb a disproportionate share of small-business labour hours, and quote-to-cash workflows in trades and construction are particularly exposed because the inbound format is genuinely messy: every supplier, contractor, and client uses different templates.

What did we build to handle messy inbound RFQs?

We designed and deployed an AI-powered RFQ processing automation that monitors a dedicated email inbox, intelligently extracts data from any format, and creates structured projects in Airtable, all within 90 seconds of the email arriving.

How it works:

The automation monitors a dedicated RFQ inbox via Microsoft Outlook. When an email arrives, it splits out all attachments and routes each one through the right extraction pipeline based on file type: spreadsheets, PDFs, Word documents, and even photos or screenshots, which are processed through vision AI to extract readable text. All attachments are uploaded to the cloud for permanent access from within Airtable.

The combined email and attachment content is then analysed by AI to extract structured project data: customer name, order number, site details, work description, contacts, and more. Each field receives a confidence score, and the system automatically matches the RFQ to the correct customer account using email domain matching, fuzzy company name matching, and site customer matching.

From there, smart routing takes over:

• High-confidence RFQs are auto-created as projects in Airtable, with a confirmation notification sent to the team
• Lower-confidence RFQs are flagged for human review, with clear indicators showing which fields need attention

Every run is logged for audit and performance tracking. The whole pattern follows the same human-in-the-loop design we apply to every quote-to-cash automation: AI handles the extraction and routing, humans keep the final say on anything where confidence is not high.

What changed after the RFQ automation went live?

Metric	Before automation	After automation
Processing time per RFQ	Around 20 minutes	Under 90 seconds
Time saved per RFQ	None	Around 15 minutes (75% reduction)
Daily time saved	None	2.5+ hours across the team
Weekly time saved	None	12.5+ hours of staff capacity
Formats handled	Whatever the team could read	Emails, spreadsheets, PDFs, images, photos
Human oversight	Manual everywhere	Confidence-based routing keeps humans in control of edge cases
Where the saved hours went	Crowded out	Quoting, customer relationships, growth activities

The headline number is the per-RFQ drop from 20 minutes to under 90 seconds. The more important number is the 12.5+ hours weekly of office capacity that came back, and the quote turnaround that collapsed because the inbound queue stopped clogging up the system. McKinsey's research on AI for document-heavy operational workflows consistently finds that the biggest gains come from automating the boring middle of a process where humans are essentially copying data from one place to another, which is exactly what changed here. The Federation of Master Builders has documented the UK construction sector's exposure to admin-heavy quote workflows, which is why this pattern shows up so consistently across small and mid-sized trades.

How human-in-the-loop design protects the edge cases

This automation was designed with human oversight built in, not bolted on. The confidence scoring system means:

• High-confidence RFQs are processed automatically, with a notification email confirming what was created. Staff can review and adjust if needed.
• Lower-confidence RFQs are flagged for review, with clear indicators showing exactly which fields need human attention. The project is still created in Airtable, but marked with a review status.

The team always has full visibility and control. The automation handles the repetitive extraction and data entry. Humans retain strategic oversight and final decisions.

Client testimonial

"The amount of time we were spending just getting RFQs into the system was ridiculous. Now it happens automatically, and the data that comes through is actually more accurate and consistent than when we were doing it by hand. The team can focus on the actual work rather than typing things into a screen all day."

Ricky Stoltzman, CEO, Designs4You

Technical stack

For teams interested in how this works:

Built on n8n workflow automation (36-node production workflow), Claude AI via OpenRouter (intelligent data extraction and OCR), Microsoft Outlook integration (email trigger and notifications), Airtable API (project creation and customer matching), Cloudinary (attachment hosting and management), and custom confidence scoring and routing logic.

What comes next

RFQ processing is the first automation in Designs4You's roadmap. The reason the team could move this fast on Phase 1 is that the foundation was already in place: we modernised their legacy project management system first, replacing a closed JobTrackerPro instance with a structured Airtable workspace designed for automation from day one. With structured project data now flowing into Airtable automatically, the next phases will build on this foundation:

• Phase 2: Automated quoting from a standardised pricing engine
• Phase 3: End-to-end project management and sign-off workflows
• Phase 4: Automated invoicing triggered by project completion

Each phase multiplies the value of what came before. The RFQ data feeds directly into quoting, which feeds into project management, which triggers invoicing. It is a compounding return on the foundational investment.

If you want the broader playbook this case study sits inside, how to automate RFQ processing in a UK B2B company walks through the same pattern in general terms: the time-to-quote benchmark, the 5-step build, and where Human-in-the-Loop stays in the workflow.

Frequently asked questions

How do you extract structured data from messy RFQ emails?

Route each attachment through the right pipeline by file type (spreadsheets, PDFs, Word documents, photos), use vision AI for any image-based content, then send the combined email and attachment context through an extraction model with confidence scoring on every field. The model handles the variation in format and language. The confidence score decides whether the resulting project is created automatically or flagged for human review.

What is confidence-based routing in document automation?

Confidence-based routing is a pattern where the AI scores how sure it is about each extracted field, and the workflow decides what to do with the document based on that score. High-confidence extractions are auto-created. Lower-confidence extractions are flagged with the specific fields that need a human eye. This is how you get the speed of full automation without the risk of silently miscoding the awkward edge cases.

Can AI handle photo and PDF attachments in quote workflows?

Yes. Vision AI reads text from photos and screenshots, and document AI extracts structured data from PDFs and spreadsheets. For RFQ-style workflows that mix email, spreadsheet, PDF, and image inputs, the automation pipeline normalises everything into structured project data before any business logic runs. The team's job changes from typing data into a screen to reviewing what was extracted.

How does AI RFQ processing reduce quote backlogs?

Two ways. First, the per-RFQ time drops from around 20 minutes of typing and cross-referencing to under 90 seconds of automated extraction and routing. Second, the team is no longer the bottleneck, so the backlog stops compounding when volume spikes. In this case, quote turnaround that had crept to two weeks at peak collapsed back to normal because the inbound queue stopped clogging up the system.

What do you need in place to automate RFQ processing?

Three things: a clear picture of the inbound formats you receive (email, spreadsheet, PDF, photo), a reasonably clean customer or product database to match against, and a definition of which fields matter enough to gate on confidence. With those in place, a single-workflow build handles around 10 mixed-format RFQs a day and hands back the hours that were going on manual typing and cross-referencing. The cleaner the existing data, the faster the matching step is to tune.