Friday afternoon. A senior engineer is recalculating the same quote for the third time. A 5 by 4 metre louvred pergola in anthracite, motorised screens, LED lighting.
The spreadsheet has been open since Tuesday. The price moved twice, because the steel supplier sent a revised rate sheet on Wednesday. The motorised louvre line is missing, because the catalogue page was photocopied from the 2024 edition.
The customer rings, polite but cooling. The engineer says it will go out before the weekend. It will not. By Monday a competitor with a configurator has closed the deal.
That is not a careless engineer. He has done this for fifteen years and he is good at it.
A customised pergola line can support 2,000 valid configurations, each with its own compatibility rules, pricing and parts list. Working memory holds about seven things. The arithmetic does not work, and what follows is not a personal failing.
We thought we had a discipline problem. We had a QA process. We had peer review. We had a senior engineer signing off every quote. The error rate sat at fifteen percent and would not move. The moment the rule library went into software, it fell under one percent.
The errors are in the structure, not the person#
The usual frame is that someone got something wrong. A mistyped measurement. Last quarter’s price. A missed component. All true, and none of it useful, because the cause sits a level down.
The rule library is implicit. Which post systems carry which spans, which finishes exist on which frames, which motors fit which louvre widths. None of it lives in one place. It is spread across an engineer’s notebook, a procurement spreadsheet, a salesperson’s memory of last summer, a CAD library, and three versions of a pricing sheet nobody has synced since March.
Every quote rebuilds a slice of that by hand, from scattered sources, under deadline. It works most of the time and fails predictably the rest.
This is why the usual fixes stall. QA steps, peer review, checklists, retraining. Each cuts maybe a third, because each reduces transcription slips. None moves the rule library, so the error rate drops to a new floor and stops. The floor is set by working memory, not effort.
The four failure modes#
Same four in pergola dealerships, garden-room builders, carport makers and fence companies. Different names, identical structure.
Outdated pricing. Steel moves, aluminium moves, supplier discounts get renegotiated. The sheet updates when someone remembers. Quotes go out at the old number, the order is signed, the build runs at the new cost. The customer never notices. Accounting finds the variance a quarter later.
Incompatible components. A 6 metre span offered with 80mm posts when the rule says 100mm above 5 metres. An undersized motor on a motorised louvre. A drainage channel sized for a 4 metre run specified on a 5. The installer arrives, the parts do not fit, and the next referral never comes.
Missing line items. The structure is quoted, the mounting hardware is not. The lighting is listed, the transformer is not. Caught in production it delays the build. Caught on site it costs a return visit. Missed entirely, you eat the parts to keep the relationship.
Wrong dimensions. The patio measures 4.2 by 3.1. It reaches the spreadsheet as 4.1 by 3.2. Parts get cut, the structure arrives, and it is the wrong size. Either a refit you pay for, or a conversation about whose measurement it was, which you also pay for.
- 10,835 hours saved Hrovat (Slovenia, modular houses); €133M in quotes generated after the rule library moved into the configurator. 415% more inquiries, 4x lead volume.
- 70% faster quoting Hausmart (outdoor living); the entire quote-to-cash process automated end to end.
- 166 hours saved over four months YourPergola (Hungary, louvred pergolas); 331 quotes generated across four months, no extra headcount.
- 1–10 Days to launch Premade visual sales system for a known category, with the rule library and 3D models already wired.
Public proof from the case studies. Numbers are sourced and unrounded.
All four get written up as discipline problems. All four are the same problem wearing different clothes: more rules than a person can hold, less time than looking them all up would take.
What changes when the rules live in software#
One library, in one place, version-controlled, queried live as the buyer designs.
An incompatible span and post pairing is refused the moment it is attempted. The oversized-louvre motor never appears in the menu, because the rule already filtered it out. Selecting integrated lighting pulls the transformer into the parts list on its own, because the configurable BOM resolves the full set every time.
Pricing comes from one source. Steel goes up Tuesday morning, every open quote shows the new number Tuesday afternoon, with nobody in the loop.
The same spec writes into the CRM, the PDF the buyer receives, and the BOM that feeds production. One source, three outputs, no retyping.
The numbers sit in the case studies. Hrovat recovered 10,835 engineering hours and generated €133M in quotes with the same team. Caribbean Blinds went from days to instant, with 40 percent of inquiries arriving configured. Spolding and Sons did £168K in 12 weeks. Same shape every time: volume up, errors down, hours back.
The call changes too. It stops being “what size were you thinking” and starts at the design the buyer already priced. That is the job the guided selling layer exists to do.
Where staying manual still wins#
Three cases genuinely break the pattern.
Truly bespoke work. If every order is a one-off with no reusable parts, there is no rule library to fire. Handmade joinery on a site-specific build is a carpentry project, not a product. Keep the engineer, the CAD and the hand-priced quote.
Products still being figured out. If the line changes weekly because nobody knows yet what sells, the rules move faster than the configurator can follow. A spreadsheet rewrites in five minutes. Build the configurator once the product settles.
Low value, low volume. A €1,500 awning sold once a month does not justify the build. The threshold sits around €5,000 average order value, close to where made-to-order manufacturing shifts from craft to system.
Everywhere else, the damage is real and invisible until the quarterly review. And the margin leak is the smallest part of it. Every error is also a referral that never happens.
Two Monday mornings#
Before. Twenty-two weekend form submissions, all some version of “interested in a pergola, please send pricing.” Eight calls, three answered. Notes into the CRM, email to engineering. The engineer recalculates, checks two compatibility rules from memory, applies the regional margin, emails back. The PDF goes out Thursday. By Friday two buyers signed elsewhere and four stopped replying.
After. The same twenty-two land on the configurator. Fourteen finish a design, see the price, and either book a call or filter themselves out. Monday holds eight configured leads with a render, a resolved parts list and a price the buyer has already accepted. The engineer spent zero minutes on any of them.
The line to take home#
Manual quoting errors get framed as quality control, which is a polite way of saying it was the engineer’s fault. It almost never is. The work is designed wrong.
So the conversation to have is not about training or sign-off. It is whether the rule library belongs in a person’s head or in software. Pick software and the build runs one to ten days for a known category, a quarter for a custom catalogue, and the error rate lands under one percent either way.
The buyer from Friday afternoon is on a competitor’s site by now. The fix is not calling them back faster. It is being the company they reach first.
People also ask.
What is a manual quoting error?
Any inaccuracy that enters a quote because a person built it from memory or a spreadsheet instead of a rule-based system. Usually called human error. Almost always a structural error that got handed to a human.
What are the four most common ones?
Outdated pricing, because the spreadsheet missed a supplier change. Incompatible components, because the rule sits in one engineer's notebook. Missing line items, rebuilt from a phone call instead of a parts list. And wrong dimensions, transcribed into the wrong cell.
Can I reduce them without buying software?
Partly. Version-control the pricing sheet with one named owner, run a checklist of the common compatibility traps, and quote in pairs above a threshold. That cuts maybe a third. The structure is unchanged, so the floor stays high.
How does a configurator reduce them?
It reads the rule library as the buyer chooses. An incompatible pair is refused before anyone sees it. A missing part cannot be selected. Pricing comes from one source, so a Tuesday change updates every open quote at once.
Will automating quotes cut sales headcount?
Not in the lines we have watched. The rep stops translating specs and starts closing buyers who arrive configured. Armat added 40 percent revenue in three months with no new hires. The bottleneck removed was the translation step, not labour.
