Outbound calling is a numbers game that most call centers are losing before the conversation even starts. For a leading Caribbean telecommunications company, Strata Analytics combined predictive analytics with AI voice agents inside the client’s own Amazon Connect environment to reach the right customer, in the right market’s voice, at the right moment.
The Challenge: Calls That Never Reach a Person
In outbound operations like collections, most call attempts never reach a person. In a collections call center for a major telecommunications company, Strata’s analysis found that out of every 100 attempts, 58 landed on voicemail, 26 never connected, and only 16 resulted in an actual conversation. On top of that, capacity constraints left entire accounts uncalled, human agents were tied up on repetitive reminder calls, and generic bot voices didn’t sound like the customers in each market.
What We Built: AI Voice Agents on Amazon Connect
Strata implemented AI voice agents inside each client’s own Amazon Connect instance, combining three capabilities:
- Analytics that decide who to call and when. Segmentation and propensity models prioritize accounts and the best contact windows.
- Voice with each market’s accent. The agent speaks with a local voice, validated by phone with listeners from each market, and always identifies itself as the company’s virtual assistant. When it reaches voicemail, it leaves a recorded message in that same voice.
- Measurement against a control group. We measure the incremental effect on payments and outcomes, not just on call activity.
How We Work
- Mixed model: the AI agent covers repetitive volume and the accounts that go uncalled today, while people focus on negotiation and complex cases.
- Bounded MVP: one use case, one segment, one script, with success metrics agreed before kickoff.
- Compliance by design: the script is built together with the client’s legal team (virtual-assistant disclosure, recording disclosure, identity validation), and Connect enforces calling-hour and frequency limits.
Ready-to-Implement Use Cases
- Early and preventive collections: validates identity, informs the customer of their balance, and sends a payment link or logs a callback request.
- Inbound care: replaces menu-driven IVR with an agent that understands why the customer is calling.
- Visit and appointment confirmation: confirms, reschedules, or cancels appointments, reducing no-shows.
What the Evidence Shows
- The algorithm for choosing who to call raised the payment rate from 43% to 53% while making 8% fewer calls.
- Speaking with the customer raised on-time payment by 34 points compared to a written notice.
- Customers did not hang up on the bot any faster than they would on a human agent.
For telecom operators across the Caribbean and Latin America, the combination of predictive analytics, market-accurate voice, and rigorous measurement turns AI voice agents from a novelty into a channel with a measurable return — one that frees human agents to focus on the conversations that actually need them.