The problem
Inbound requests for quotes arrived as email, most of them unstructured and many as attachments. An ops coordinator re-keyed each one into a spreadsheet, checked rate cards across two separate systems, phoned a depot to confirm capacity, then wrote the quote by hand.
Volume had tripled in two years. Headcount had not. The backlog meant a customer often waited three days for a number, by which point a competitor had frequently already answered — and in freight, the first credible quote wins a disproportionate share of the work.
What we measured first
Before building anything we sat with the coordinators for a week and timed the actual steps. The result was not what anyone expected: the pricing itself took about four minutes. Everything around it — reading the email, finding the lane, chasing capacity, formatting the reply — accounted for the remaining two and a half hours of handling time per quote.
That single finding reshaped the project. We were not automating pricing. We were automating everything surrounding it.
What we built
- An extraction agent that reads the inbound email and any attachment, pulling lane, weight, dates and special handling into a structured record.
- A pricing step that queries the existing rate cards and the capacity system directly, rather than through a person on the phone.
- A drafted quote surfaced inside their existing CRM, with its reasoning shown, for a human to approve or amend in one click.
- A confidence threshold: anything unusual is escalated to a coordinator rather than guessed at. This is the single thing that kept the error rate down.
- Monitoring, retries and alerting on every step, plus a runbook their own team owns.
How it runs now
An RFQ lands. Within about eleven minutes a drafted quote is waiting in the CRM with its workings attached. A coordinator reads it, adjusts if needed, and sends. Roughly one in six is escalated for a human to price properly — which is by design, not a shortfall.
Nobody was made redundant. The two coordinators now handle around triple the volume, and spend their time on the awkward quotes where judgement actually pays.
What we would do differently
We spent three weeks on attachment parsing before discovering that eighty per cent of the awkward cases came from just two customers — both of whom would happily have used a simple form instead. It was a cheaper fix, and we found it far too late. We now check the distribution of edge cases before building for them.
"We'd tried automation twice before and both attempts rotted within a month. Mindspace built the boring parts properly — logging, retries, alerts — so it's still running a year later."
Leif ElgethumFounder & CEO, Retrolux