Customers

What one pilot team found.

Seven days on a live Salesforce queue at a global enterprise software provider. Here's what happened, and a calculator to try the numbers on your own team.

Case study
An L1 enterprise software support queue

The team ran a Claude agent against live Salesforce cases for a week. For each case it read the thread, the internal notes and past resolutions, then drafted a reply for a specialist to review. It never sent anything itself.

L1 Support Lead · Global enterprise software provider

4.4×

less specialist review time per case

89%

of the queue was diagnostic work the agent could help with

21

drafts prepared across 16 live cases

0

drafts reached a customer without human review

ROI calculator

Work out what it saves your team.

The diagnostic share (89%) and precedent share (39%) come from a 16-case pilot at one client. Minutes saved and the hourly rate are estimates, so swap in your own team's figures.

L1 analysts on the queueEach analyst gets the savings below on their own caseload.
4
Cases per analyst, per monthHow often each saving below is earned.
23
Diagnostic shareThe cases the agent can research and draft. Provisioning and access requests save nothing, so they're left out.
89%
Of those, share with usable precedentCases with a past resolution on record. These earn the extra saving below on top of the base.
39%
Minutes saved on a diagnostic caseReading the thread and notes, working out who owes the next reply, and writing the draft. An estimate, so adjust it.
34
Extra minutes when precedent existsFinding and reading past cases on the same question, the slowest part by hand. An estimate, so adjust it.
18
Fully-loaded hourly rateSalary plus overheads. Set it to what your team really costs.
£60
Time returned
56.0 h

saved per month

672 h

saved per year

90

working days a year, at 7.5 h

37 min

average across every case, including ones it barely touches

Money returned
£40,305

of specialist time a year

£3,359

of specialist time a month

Time saved is modelled, not observed end to end. The pilot covered 16 cases at one client, so your results will vary. It assumes every draft goes past a human reviewer.

Want to see it on your kind of case?

Bring the case you'd expect us to get wrong. We'll run it in our sandbox, live.

Book a demo
30 minutes

Live on our sandbox. Nothing to prepare, no access to your systems.