Tiered AI Support System for a New York City Fixed Wireless & Fiber Business ISP

Client Profile

One of the largest and fastest-growing fixed wireless and fiber broadband providers in New York City, delivering enterprise-grade connectivity to thousands of commercial businesses. Established in 2002 to address the chronic under-connectivity of NYC’s commercial buildings — particularly in boroughs where traditional fiber deployment is prohibitively expensive — the company has grown into a major player in the metropolitan business broadband market.

The numbers tell the story of scale: over 800 lit buildings, more than 40 network hubs, and a near-net footprint exceeding 60,000 commercial buildings. The company is backed by a major infrastructure-focused private equity firm and serves a diverse customer base ranging from small businesses and startups to enterprise organisations and public institutions.

Their network combines fiber-optic and fixed wireless access technologies, enabling rapid deployment to buildings that would take traditional fiber providers months to reach. Speeds range from 10 Mbps to 10 Gbps, with SLA-backed service levels that compete directly with incumbent fiber providers — often at a lower price point and with dramatically faster installation timescales.

Industry: Telecommunications (B2B / Commercial Broadband) · Region: United States — New York City, New Jersey · Products Used: VoiceFlow AgentIQ (Multi-Persona) · CRM Integration · Network Serviceability API

The Challenge

The company’s success created a multi-front complexity problem that a single support operation was struggling to manage:

Three distinct customer types, one support team.

The company had expanded from its core commercial B2B business into two additional segments — residential service (addressing NYC’s digital desert problem in underserved boroughs) and wholesale partnerships (providing capacity to other service providers and channel partners). Each segment had fundamentally different support requirements:

• B2B commercial customers expected white-glove, SLA-driven support. When a business’s internet goes down, it directly impacts revenue. These customers need immediate acknowledgement, rapid diagnosis, and guaranteed resolution within contractual timeframes. They also expect proactive communication during outages and scheduled maintenance.
• Residential subscribers needed high-volume, cost-efficient support for typical consumer issues: setup help, WiFi troubleshooting, billing queries, and plan changes. The emotional register was different — less “our business is losing money” and more “my kids can’t do their homework.”
• Wholesale partners required a technical, peer-level interaction focused on circuit provisioning status, capacity planning, and backbone performance — a conversation between engineers, not between a customer and a support agent.

One support team trying to serve all three was inevitably optimising for one at the expense of the others. B2B customers were getting consumer-grade support. Residential customers were sitting in queues behind business-critical outage calls. Wholesale partners were explaining technical concepts to agents trained on consumer troubleshooting.

Sales was clogged with qualification calls.

New York City’s commercial real estate market moves fast. When a business decides to move offices, they need internet connectivity confirmed before they sign a lease. The most common inbound sales enquiry was a serviceability check: “Can you serve 123 Main Street?” or “What speeds are available in the Zipper Building?”

These queries were being handled by the sales team — skilled, expensive, commission-earning salespeople spending 10–15 minutes per call looking up buildings in an internal database, checking line-of-sight to the nearest hub, and quoting available speeds. For a company with 60,000+ near-net buildings, the volume of these queries was significant, and most didn’t convert to sales. The sales team was spending more time on qualification than on closing.

Outage communication was reactive and inconsistent.

When a network hub experienced an issue affecting multiple buildings, the support team would spend the first 30 minutes fielding calls from affected customers — each asking the same question (“Is there an outage in my area?”) — before a human had time to compose and send a proactive notification. By then, the reputational damage was done.

Our Approach

Rather than deploying a single AI agent and hoping it could handle the full range of interactions, we built a tiered system with three distinct AI personas — each optimised for its specific customer type, with appropriate tone, technical depth, and escalation protocols. All three agents share a unified knowledge base and customer database, ensuring consistent information regardless of which persona a customer interacts with.

What We Built

Persona 1: Enterprise Support Agent (B2B)

Designed for the commercial customer base — businesses, enterprises, and institutions with SLA-backed service.

• SLA-aware interaction model — the agent knows each customer’s service tier and contractual commitments. A Tier-1 enterprise customer with a 4-hour resolution SLA is handled differently from a small business on a standard plan.
• Proactive outage communication — when network monitoring detects an issue affecting a hub or circuit, the agent proactively contacts affected B2B customers via their preferred channel (call, email, SMS) with incident details and estimated resolution time. No more customers calling to ask — they’re informed before they notice.
• Circuit status and performance queries — the agent accesses real-time network performance data, answering questions about current throughput, latency, packet loss, and any recent incidents on specific circuits.
• Maintenance scheduling — coordinating planned maintenance windows with commercial customers who need advance notice and may require temporary arrangements.
• Escalation with full context — when human intervention is required, the ticket includes SLA status, customer tier, interaction history, and diagnostic results. The human agent never starts cold.

Persona 2: Residential Support Agent

Designed for the growing residential customer base — optimised for volume, empathy, and straightforward resolution.

• Connectivity troubleshooting — handling the standard residential support workflow: is it an outage? Is it your router? Is it your device? Let’s fix it.
• New subscriber onboarding — guiding residents through equipment setup, network connection, and account activation.
• Billing and account management — plan changes, payment processing, billing queries, and account updates.
• Tone calibration — warm, patient, and accessible. Residents in underserved communities may be connecting to high-speed broadband for the first time. The agent meets them where they are.

Persona 3: AI Serviceability Checker (Sales Support)

This is not a support agent — it’s a sales acceleration tool integrated into the inbound sales workflow.

• Instant building lookup — prospects call or chat with a serviceability query, and the agent cross-references the company’s building database in real-time. “Yes, we serve that building with up to 1 Gbps” or “That building is in our near-net footprint — we can have service available within two weeks.”
• Automated quoting — for standard service configurations, the agent generates a preliminary quote based on the building’s connectivity options, including speed tiers and indicative pricing.
• Qualification and handoff — if the prospect is interested, the agent captures contact details, service requirements, and any special considerations, then schedules a follow-up with a human sales representative. The sales rep receives a warm, pre-qualified lead rather than a cold enquiry.
• Broker and agent support — real estate brokers and tenant representatives frequently check serviceability on behalf of their clients. The agent handles these intermediary queries efficiently, providing information that helps the broker recommend the ISP to their clients.

Projected Impact

Metric	Target
B2B customer satisfaction	Improved through SLA-aligned, proactive support
Residential support cost per interaction	Significant reduction through AI-first resolution
Sales team time on qualification calls	Eliminated — redirected to closing and relationship management
Outage notification time	Proactive (minutes) vs. reactive (30+ minutes)
Serviceability query response time	Instant (from 10–15 minutes per manual lookup)
Sales conversion rate	Improved through faster qualification and professional first impression
Support team capacity	Reallocated from repetitive queries to complex, high-value interactions

Why This Matters

New York City is one of the most competitive broadband markets in the United States. This ISP competes against incumbents with orders of magnitude more resources. The AI deployment doesn’t just reduce costs — it creates a service experience that the incumbents can’t match.

When a potential customer calls to check if their building can be served and gets an instant, accurate answer with a preliminary quote — compared to the incumbents’ “we’ll get back to you in 3–5 business days” — the competitive advantage is visceral. When a B2B customer receives a proactive outage notification before they’ve even noticed the issue, the trust that builds is worth more than any SLA document.

The multi-persona architecture is the key design decision. A single agent trying to serve a C-suite executive whose trading floor is offline and a first-time broadband user in a public housing building would inevitably fail at both. By building distinct personas that share a common intelligence layer, the system delivers appropriate experiences for each audience without fragmenting the underlying data and knowledge.