Nortinia AI Call Center — what it solves, and when it pays off

Six months ago the first phone call was answered by a voice agent on Nortinia infrastructure. Since then a handful of Hungarian SMEs have been running it in production, and we have just enough data to articulate who benefits, who shouldn't bother, and when the investment turns positive.

The volume threshold: 300 calls per month

The first question every prospect asks is "is this worth it for us?" The math is simpler than people expect. The fixed monthly fee of Nortinia AI Call Center, plus the per-minute cost, becomes cheaper than a human call-center resource at roughly 300+ calls per month on a single line.

Below that threshold, the fixed costs (SIP trunk, model licence, integration) eat up the savings. Above it, the return scales fast: at 800 calls per month you save half an FTE; at 2,000 you save an entire position.

The 300-call number isn't arbitrary. A human agent typically handles 600-800 calls per month, and 300 is the point where the Nortinia rollout has clearly positive ROI on a six-month horizon.

What it's good at: short, scripted, decision-tree-shaped

The voice agent excels in three call types:

1. Appointment scheduling and confirmation — "you have a booking tomorrow at 14:30, can you confirm?" Simple yes/no/reschedule tree, high resolution rate.
2. Tier-1 information requests — opening hours, invoice status, shipment tracking, common FAQs. Answers come from a database; the agent only reads them out.
3. Structured surveys — NPS, satisfaction surveys, event-driven feedback. 5-8 questions, fixed answer options, calm consistent delivery.

What it's not good at

Now the harder part. We explicitly do not recommend these call types for AI:

• Complex technical troubleshooting — if a customer says "the red light is blinking but it wasn't yesterday," a human needs to interpret the context. AI can derail here and hurt CSAT.
• Emotionally charged complaints — an angry customer should not talk to AI. It escalates frustration, and demonstrably damages brand perception.
• Multi-party scheduling — coordinating three calendars with alternatives is human work. AI either gets lost or has to be over-prompted.

One of our early design partners learned this the expensive way: they put complaints on the AI too, and within three weeks we measured a 12% NPS drop. We pulled complaints back to humans, NPS recovered. The lesson: not every call belongs on AI.

Hybrid is the goal, not full handover

The best results come from SMEs where AI handles 40-60% of calls and the rest go to a human queue automatically. 100% automation is not the goal. The goal is for human agents to focus on calls that actually need human thought.

The typical split for a live customer:

• 42% — AI resolves it alone
• 38% — AI starts it, human finishes (with context handoff)
• 20% — immediately routed to human (sensitive topic, emotional signal, repeated misunderstanding)

The payback curve

• Month 1 — Telnyx SIP integration, call-flow mapping, FAQ ingestion. Net savings: zero, sometimes slightly negative.
• Months 2-3 — go-live and tuning. First ROI signals appear; resolution rate climbs from 25% to 40%+.
• Months 4-6 — stable run. Full payback arrives, and from here it's purely positive.

Over six months a typical SME zeroes out the rollout cost in 4-6 months and is in pure profit by month 7.

Four questions to ask yourself before you start

1. Do we genuinely have 300+ calls per month on the line we want to automate?
2. Are most of those calls really scriptable, or is each one unique?
3. Do we have a human team to pick up the handed-off calls, or will AI be our only front line?
4. Are we prepared to actively tune the system in the first 6-8 weeks, not just deploy it and forget?

If the answer to all four is yes, this is probably a good move. If not, let's talk first.