Enquiries · case study

For two months,
the calculator was quietly wrong.

A browser-based quoting calculator kept giving instant answers the whole time it was drifting away from the actual cost model behind it — not broken, not offline, just wrong, on every job, for as long as nobody compared the two.

Start a conversation See the whole crew

You go home. It doesn't.Every promise gets a row and a due dateAnswered the same day, not the next working dayA drawing before the quoteUnverifiable is never a passNine desks, one address eachNothing is sent that a human cannot read backThe folder exists before you price itYou go home. It doesn't.Every promise gets a row and a due dateAnswered the same day, not the next working dayA drawing before the quoteUnverifiable is never a passNine desks, one address eachNothing is sent that a human cannot read backThe folder exists before you price it

The story

A tool nobody was checking against the model it was supposed to mirror

A calculator that always returns an answer never announces that the answer has quietly stopped being right.

A quoting calculator built for fast, self-serve, bespoke pricing had, over roughly two months, drifted away from the canonical cost model it was supposed to mirror. An overhead figure baked into the tool no longer matched the real, current number. A tax loading was being applied that didn't belong on this kind of business at all. The markups used didn't match the actual margin target the cost model was built around.

None of it produced an error message, because none of it was a bug in the ordinary sense — the calculator worked exactly as built. It simply hadn't been checked against the model it was meant to represent, and the two had quietly stopped agreeing with each other.

The fix that mattered wasn't correcting the numbers once. It was building a permanent, automated comparison between the calculator and the canonical cost model, run on a standing schedule, so the next drift is caught before it has two months to compound.

Two systems, quietly disagreeing for two months

Sourced from Hobfolk's own operational record, dated and on file — see the other four. Client identity is never named; the numbers and the root cause are not softened.

What changed

The permanent fix, not just the patch

One canonical cost model, everything else checked against it

A quoting tool is a presentation of the cost model, never a second source of truth — and it's checked against the model on a schedule, not trusted to stay aligned on its own.

Drift is treated as a standing risk, not a one-off bug

Any tool that presents pricing independently of the canonical model is assumed to be able to drift, and checked accordingly — not trusted because it worked correctly when it launched.

The comparison runs automatically, on a schedule

Not triggered by someone noticing a quote looks off — a standing check compares the tool's output against the model's on its own.

A working tool is not the same claim as a correct tool

The calculator never crashed and never looked broken. Working and being right are different properties, and only checking the second one catches this.

The specific drift is logged, not just corrected

Which figures moved and by how much is recorded, so the next audit has a baseline to compare against.

Every self-serve pricing surface gets the same standing check

Not just the one that was found wrong — the same drift-check discipline now applies to every tool that quotes a price without a human in the loop.

How it unfolded

The sequence, in order

A quoting calculator launches
Fast, self-serve, built to mirror the canonical cost model at the time.
The underlying cost model changes
Legitimately, elsewhere, for good reasons — an overhead figure, a margin target.
The calculator is not updated in step
Nothing forces the two to stay in sync; the calculator keeps answering instantly.
The gap compounds, invisibly
For roughly two months, every bespoke quote through the tool is quietly wrong.
A comparison is finally run
The calculator's numbers checked line by line against the canonical model.
A standing, automated check replaces the one-off fix
Wired into the regular schedule, so the next drift is caught in days, not months.

Said plainly

What this desk still won't do

Where this desk stops

It does not send a quotation on its own. Every price is checked and released by a person. Above your set threshold it needs three separate sign-offs, not one.
It does not discount, negotiate or agree a payment plan. Those are commercial decisions and they stay with you.
It does not invent a lead time. If the floor has not confirmed a date, the reply says so rather than guessing a date that will slip.

Questions

Asked before, answered here

How is this different from just testing the calculator once at launch?

A launch test only proves the tool was right on the day it launched. This is a standing, scheduled comparison specifically because the underlying model is expected to keep changing after launch, and nothing about the calculator would ever announce that it had fallen behind.

Why didn't anyone notice from the quotes themselves?

Because a wrong quote doesn't look wrong from the outside — it looks like a normal answer to a normal question. The only way to catch it is to compare it against something else, not to read it in isolation.

Does this mean every quoting tool needs this?

Any tool that gives a price without a person checking it against the current cost model in that moment is a candidate for exactly this kind of drift, and gets the same standing check applied.

What's actually logged when a drift is found?

The specific figures that moved and by how much, so the next check has a clear before and after, rather than just a pass/fail.

When was your own quoting tool last checked against your actual cost model?

Tell us how it's built, and we'll tell you plainly whether it could be drifting right now without anyone knowing.

Start a conversation Read the method

For two months,the calculator was quietly wrong.