Should you replace an outsourced (BPO) support team with an AI agent? The answer depends on what your customers ask. This post gives you the 2026 numbers and a small model to test your own volume.
Illustrative example, not production-ready. It assumes every conversation the AI doesn't resolve goes to the BPO, and it ignores setup, tooling and management costs on both sides.
def support_costs(
conversations=10_000, # per month
minutes_per_conv=8, # human handle time
bpo_hourly_rate=12.0, # USD
occupancy=0.75, # share of paid time agents spend on tickets
ai_resolution_rate=0.40,
ai_price_per_resolution=0.99,
):
def bpo_cost(convs):
hours = convs * minutes_per_conv / 60 / occupancy
return hours * bpo_hourly_rate
bpo_only = bpo_cost(conversations)
ai_resolved = conversations * ai_resolution_rate
hybrid = ai_resolved * ai_price_per_resolution + bpo_cost(conversations - ai_resolved)
return {
"bpo_only": round(bpo_only),
"hybrid": round(hybrid),
"saving": round(bpo_only - hybrid),
"saving_pct": round((bpo_only - hybrid) / bpo_only * 100),
"bpo_per_conv": round(bpo_only / conversations, 2),
"hybrid_per_conv": round(hybrid / conversations, 2),
}
print(support_costs())
With the defaults, a hybrid saves about $4,573 a month (21%).
Two things to adjust before trusting the output:
ai_resolution_rate with the rate from a real pilot. Count true resolution, not deflection: a customer who leaves the chat and calls later isn't resolved.
| Work type | Best fit |
|---|---|
| Order status, password resets, policy questions | AI alone |
| Refunds, plan changes, warranty claims | AI drafts, a person approves |
| Complaints, cancellations, complex technical cases | People |
Customers back this up. In Gartner's survey of 5,801 US customers (run in early 2025), 54% trusted human agents more than AI for recommendations, versus 32% who trusted AI more. A separate 2026 Gartner survey found 87% say a human must be available when companies use GenAI.
Klarna reported in February 2024 that its AI assistant handled 2.3 million conversations in its first month and cut resolution time from 11 minutes to under 2. In May 2025, its CEO told Bloomberg that a cost-first approach had lowered quality, and the company resumed hiring human agents. The takeaway: cut the simple volume, not the human option.
Originally published on the Neura Dynamics blog, which includes the interactive calculator and a quality comparison. For help working out your own numbers, book a free consultation.