Finance cannot approve “improve customer service with AI”. They can approve “resolve tier-one order-status tickets in the helpdesk we already run”. The difference is a named workflow, a baseline you already have, and a cost that does not depend on a vendor’s industry average.
We will not supply that average. Anyone who will is guessing with your letterhead on it.
Name the queue, not the category
“Customer service” is a department. “Tickets whose intent is order status, currently sitting with a coordinator who copies from the ERP into Zendesk” is a workflow. The second one has an owner, a system, a volume you can count, and a mistake you can describe.
Write the case as one paragraph the CFO can reread without you in the room. Which queue. Which system it already lives in. What a person does today, step by step. What the agent would be allowed to write, and what still waits for a person. What you will stop doing if it works.
If you cannot get that paragraph out of the process owner in a meeting, you do not have a case yet. You have a theme.
Collect the baseline yourself
The after-number is meaningless if nobody wrote the before-number down.
Count, for a normal month, on this queue only: items arriving, items finished by a person, items that waited more than your own SLA, and hours of coordinator or agent time on the copy-paste steps. Use your helpdesk or CRM reports, not a spreadsheet reconstructed from memory the week before the meeting.
Do it before anyone builds. We will ask for this on the assessment because we cannot invent it later without cooking the result.
Reroute rate, first human response, time-to-first-touch on a lead, exception age on a bill: pick the one that matches the leak. One metric, owned by the same person who owns the queue. A dashboard of twelve new charts is how the case dies in month two.
Cost lines finance will actually read
There are four, and they are not all on our invoice.
The go-live. For us that is $9,900 to put one supervised agent live on one workflow, in your systems. That number is the build, the map, the wiring, and the check on real items. It is not a six-month programme and it is not a platform licence.
The monthly run. $2,900 a month once it is live, month to month. That is a named owner from our employed team, the exception path, and the change work when your own systems move. Leave any month after go-live.
Your reviewer’s time. Someone on your side still signs the writes that can hurt you, and someone owns the exception types. If that person does not exist, the monthly number is incomplete and the case should say so.
Model calls and the integration surface. These sit inside the monthly for our package. If you build internally they do not, and they move when volume moves. Do not hide them in “innovation”.
What we will not put in the case: a percentage saving, a headcount reduction, or a payback period built on a blog post. If your own baseline supports a comparison, you can make it. We will not make it for you.
What “good” looks like to a finance review
A named workflow. A baseline collected from the live system. A write scope that names the dangerous actions. A cost that matches a package, not a rate card. An exit: you can leave the monthly after it is live.
The assessment is free. If we cannot see a workflow worth putting live, we will say so on the call, and you keep the finding. That is also a result a finance review can live with.