Answer the help-desk line—even when every engineer is mid-ticket
For a managed service provider or an internal IT team, the phone is the channel people reach for when something is actually broken. Calls spike at the worst moments: a server goes down and every affected user dials in at once, each opening a duplicate ticket. After hours the trade-off gets sharper—either a senior technician is woken for a printer question, or a business-down call sits in voicemail until morning while the client starts evaluating other providers.
ThunderPhone gives an IT-services business an AI phone agent that answers every routed call, runs the provider's written severity checklist, creates structured tickets through connected systems, and escalates to on-call staff only under the conditions the provider defined.
Move from "someone left a voicemail" to a complete, triaged ticket
An IT-services agent can be configured to:
- identify the caller, their company, and a callback number;
- collect the affected system, the symptoms, any error message, and how many people are affected;
- classify severity against the provider's written definitions and route business-down calls to the on-call transfer path;
- create a structured ticket in a connected ticketing or PSA system and read the ticket number back;
- answer ticket-status questions from a connected lookup;
- state known outages and scheduled maintenance instead of collecting duplicate reports;
- take new-business inquiries with a different intake than support calls; and
- place approved outbound calls for outage notifications and maintenance reminders.
The agent should not improvise troubleshooting, adjudicate identity, or touch credentials. You define the questions, the severity rules, and the actions it may take.
Sample help-desk call flow
Agent: Thank you for calling Northgate IT support. Can I have your name and company?
Caller: Dana Reyes, at Harbor Freight Logistics. Nobody in our office can open the shared drive.
Agent: Thank you, Dana. Is anyone able to work right now, or is the whole team stopped?
Caller: Pretty much everyone is stopped. Email still works, but all our shipping documents are on that drive.
Agent: Understood—that meets the criteria for an urgent response. I have created ticket 48291 with those details and I am connecting you to the on-call engineer now. Please stay on the line.
Because the answers matched the provider's written business-down criteria, the agent warm-transferred instead of promising a callback. A routine caller—one person locked out, a slow laptop—would finish the structured intake, get the ticket number, and hear the configured response-time expectation. The severity definitions live in the prompt, so changing the triage policy is an edit, not a retraining project.
Where an AI phone agent fits in an IT-services operation
Severity triage by your written rules
Experienced dispatchers separate "the server room is dark" from "my second monitor flickers" in the first minute, and the agent can run the same checklist every time: what is affected, how many users, is anyone able to work, what does the client's agreement say about priority. Business-down calls route to the on-call path; everything else becomes a ticket at the assigned priority. Triage stops depending on who happened to answer.
After-hours on-call escalation with a paging fallback
Evening and weekend callers reach the same agent, not a voicemail box. Routine issues become complete tickets waiting in the morning queue. Calls that meet the after-hours criteria trigger the on-call chain: a warm transfer to the primary engineer, and if the engineer cannot take the call, the configured fallback—an outbound call to the backup or a webhook into the paging system—so a business-down report never dead-ends.
Ticket-status calls answered from a connected lookup
"Any update on my ticket?" calls interrupt engineers without moving the ticket. With a connected lookup tool, the agent can retrieve the current status and last note and relay what the provider allows it to share. Questions the systems cannot answer route to a person instead of a guess.
Outage-notification outbound calls
When an incident is confirmed, an outbound campaign can call affected client contacts with the approved notification: what is impacted, what the team is doing, when the next update will come. Inbound, the agent states the known issue to callers reporting the same symptoms and records a lightweight "me too" report instead of opening duplicates. The organization defines consent, identification, and calling-window rules for all outbound use.
Scheduled-maintenance reminder calls
Before a maintenance window, a campaign can work through the affected contact list, and deliver the approved reminder—what is happening, when, and what users should expect. Voicemail detection can leave the configured message, and the retry policy governs additional attempts, so notice does not depend on someone finding time to hand-dial the list.
Password-reset and access requests, routed under policy
Reset requests are a known social-engineering target, so the agent's role is deliberately narrow: it collects exactly the verification fields the client's written policy specifies, records the request and answers on a ticket, and routes it to the path the policy names—a technician, a self-service reset flow, or a queued task. The agent never performs verification beyond collecting the configured fields, never decides whether a caller is who they claim to be, and never resets or reads out credentials itself.
New-client inquiry intake
A prospective client is a different conversation from a support call, and the agent can switch intake accordingly: company size and industry, current IT arrangement, what prompted the call, and timeline. High-intent inquiries can warm-transfer to sales during business hours; after hours the agent captures the full picture so the callback starts informed, not from a one-line voicemail.
Vendor-call coordination
Chasing an ISP, hardware, or software vendor often means long holds and keypad menus before a human appears. On approved outbound workflows, the agent can place the call, navigate the phone menu, wait on hold, and deliver the scripted request—turning technician hold time into a structured note on the ticket.
Multilingual end-user support
An MSP's client roster can include warehouses, clinics, and offices whose end users prefer different languages. One number serves them all: the agent starts in the primary language, switches when a caller uses a configured additional language, and keeps the same triage, ticketing, and escalation logic in every language.
Features that matter to an IT-services operation
Structured intake without a rigid phone tree
The agent asks follow-up questions based on the caller's answers—asset details for hardware, error text for applications, scope questions when several users are affected. The ticket arrives with the fields the team needs, without forcing every caller through the same menu.
Tickets that land in your PSA or ticketing system
Connect a custom API for ticket creation, status lookup, and contact matching, or attach remote MCP servers for automatic tool discovery. HMAC-signed webhooks push call events—with structured outcomes like severity classification and collected fields—into whatever queue the team already watches. The agent can use only the operations those connections expose.
Per-client knowledge that stays current
Attach escalation matrices, supported-environment notes, known-issue lists, and maintenance calendars as text, PDF, or DOCX documents (up to 5 MB per text file and 50 MB per PDF or DOCX), or import pages from the company website by URL. The agent searches those sources during the call and routes questions the material does not answer.
Your numbers, inbound and outbound
Production numbers can be brought in through supported direct VoIP connections or manual SIP configuration for any SIP trunk, so the support line clients already know keeps working. Verified imported numbers support inbound answering and outbound campaigns, and the agent can hold or navigate keypad phone menus when a workflow requires it.
Transfers that preserve context
Cold transfer routes immediately. Warm transfer, available on eligible production VoIP or SIP numbers, lets the agent keep the caller on hold while it briefs the on-call engineer privately—company, symptoms, scope, ticket number—before connecting. If the engineer cannot take it, the agent returns with the configured next step.
A website channel on the same brain
An embeddable web widget lets users start a voice conversation from the support portal or company site. Widget sessions run on the same agent and produce the same tickets and transcripts as phone calls, so "get help" clicks and phone calls land in one pipeline.
Multilingual conversations
ThunderPhone supports 47 voice languages, with mid-conversation switching when a voice supports the selected set.
Reviewable calls and repeatable tests
Review transcripts, recordings when enabled, and graded call logs, then turn recurring failure cases into simulations: a business-down report that must trigger escalation, a routine request that must not, a reset request that must follow the verification policy, an unavailable on-call engineer. Regression suites with pass-rate gates catch prompt changes that break the triage script, and A/B experiments compare intake variations on live traffic before rollout.
Guardrails an IT provider should configure
On a support line, what the agent is prevented from doing matters as much as what it can do. A well-configured IT-services agent:
- never performs identity verification beyond collecting the fields the written policy specifies;
- never resets, creates, or reads out credentials, access codes, or MFA secrets;
- never gives troubleshooting instructions beyond the approved script steps;
- never promises response times or remedies other than the configured ones; and
- routes anything outside its scope—legal threats, security-incident reports, cancellation demands—to a person by the defined path.
These boundaries live in the prompt and tool permissions, and simulations can verify them continuously—a scripted attempt to extract a reset without the required fields should fail the grade every time.
What would this cost each month?
Suppose an IT-services provider handles 1,000 AI-assisted calls per month, averaging 4 minutes each:
1,000 calls × 4 minutes = 4,000 minutes per month
- Spark:
4,000 × $0.02 = $80/month - Bolt:
4,000 × $0.05 = $200/month - Storm:
4,000 × $0.09 = $360/month
That is base engine usage, not a quote. Premium voice or language paths can add up to 3¢ per minute, and phone-number or carrier charges depend on the telephony setup. Hold time bills a flat 2¢ per minute; voicemail and phone-menu minutes bill at the engine rate. Optional add-ons such as verbal acknowledgements, call supervision, and long prompts carry their own published surcharges. The agent builder shows the current all-in rate before calls are placed.
Compliance and client data
ThunderPhone is GDPR and HIPAA compliant. The provider remains responsible for configuring its own workflows appropriately—recordings and any required notices, consent for outbound calling, data retention, per-client verification policies, and what information the agent may collect and share through integrations.
Frequently asked questions
Can a ThunderPhone agent create tickets in our PSA or ticketing system?
Yes, when connected to a system that exposes the operation: custom APIs, remote MCP servers with automatic tool discovery, and HMAC-signed webhooks that deliver call events and structured results into your queue. If your platform has an API for tickets, contacts, or status, the agent can use those operations.
Can it tell a business-down emergency from a routine request?
It can follow your written severity definitions. The agent asks the scoping questions you configure and routes calls that meet your criteria to the escalation path, filing everything else at the assigned priority. It applies your rules; it does not invent its own judgment about what counts as an emergency.
Can it reset passwords or verify a caller's identity?
No. The agent collects the verification fields the client's written policy specifies and routes the request to the configured path. It never performs verification beyond collecting those fields, never decides a caller is legitimate, and never resets or reads out credentials itself.
What happens after hours if the on-call engineer does not answer?
The configured fallback runs: an outbound call to the backup contact, a webhook into your paging system, or a scheduled retry. The caller is told what to expect, and the ticket records every escalation attempt.
Can callers ask about the status of an existing ticket?
Yes, with a connected lookup tool. The agent retrieves the ticket and relays the status and notes you permit it to share; anything it cannot look up is handed to a person.
Can it call our clients during an outage or before maintenance?
Yes, through individual outbound calls or campaigns on an eligible production number, with voicemail detection and configurable retries. The organization is responsible for consent, identification, do-not-call requests, calling windows, and applicable outbound-calling laws.
Can it support end users in languages our technicians don't speak?
Yes. Configure any of the 47 supported voice languages as additional languages, and the agent switches mid-call when the caller does, keeping the same triage and escalation behavior. The selected voice must support the configured language set.
How do we test it before clients hear it?
Test in the browser first, then run AI-caller simulations of the scenarios that worry you—a business-down report, a social-engineering attempt on the reset flow, an unavailable on-call engineer. Simulated calls are graded against your criteria and re-run as a regression suite whenever the prompt, severity rules, or knowledge base change. Simulations are billable real calls at standard rates.