What is the cost per resolved contact for AI vs a human agent?

The cost per contact for an AI agent is typically a fraction of a human agent's loaded cost, but the cost per resolved contact tells a different story. To compare honestly, divide total cost (vendor fees, integration upkeep, human escalation handling) by resolved contacts (not contacts handled). The AI is still usually cheaper on routine work; the gap is smaller than the vendor slide implies.

A vendor slide shows two numbers: a human agent costs $7 per contact, the AI costs $0.50 per resolution. The team multiplies by volume, prints a savings number, and presents it to finance. Twelve months later the actual savings are a quarter of the projection. The accounting was technically correct and misleading by omission.

What people in the field are saying

kdschemin's "The ROI isn't missing, the redesign is" argues that AI cost savings show up only when the operation is redesigned. The per-contact arithmetic is real but it does not arrive on the cost line on its own.

What does the vendor's number usually leave out?

Four things. The cost of the contacts the AI tried and failed (the human still picks those up at full cost). The integration upkeep: someone has to keep the knowledge base, the prompts, and the back-end connections current. The escalation tax: the residual human team gets the harder contacts, which take longer per case. The hidden volume: AI lowers friction, more customers contact, total cost can rise even when per-contact cost drops.

How do you compute the honest number?

Take total operating cost over a window (AI vendor fees plus engineering plus the residual human team) and divide by resolved contacts. Resolved means the customer did not come back about the same issue within a window. Compare against the same denominator pre-AI. The delta is the real cost-per-resolution change, which is usually smaller than the per-contact arithmetic suggests and still positive. The cost-per-resolution simulator runs this calculation across the main pricing models, so you can see the honest number for your own volume and resolution rate.

Where does the AI clearly win?

Pure-routine, high-volume contacts where escalation is rare and the answer comes from a well-maintained source: order status, FAQ, simple account changes. Cost per resolved contact drops sharply because both the numerator drops and resolution rate stays high.

Where does the AI not clearly win on cost?

Contacts that need judgement, where the AI tries, fails, escalates, and a human handles the case from a worse starting position than they would have without the AI. The AI's per-contact cost is small but the resolution-cost equation includes the failed attempts. On these contacts, the right move is often to skip the AI and route straight to a human.

What is the practical pattern?

Segment by contact type. For each type, measure cost per resolved contact under the old workflow and under AI. Route to whichever wins on resolution-cost, not on per-contact cost. The AI handles some types well below human cost; the AI handles other types more expensively when total costs are loaded honestly. The portfolio matters; the average alone does not.