AI Receptionist for an MSP: Cost and After-Hours Ticket Logging

An AI receptionist costs a small MSP from $0 to about $150 a month for after-hours calls, or up to $300 if it also takes daytime overflow. Yes, it can log them as tickets: some services create PSA tickets directly, and most PSAs accept an emailed summary. It should log and escalate, never fix, promise times or reset access.

The money is small beside one lost client, so the real questions are what the assistant captures, when it wakes an engineer, and what it must refuse to say. Get those wrong and the plan fee stops mattering: one ransomware call logged as a routine fault, or one password reset done for a stranger, outweighs any saving on the subscription.

Follow me on Instagram@sagnikteaches

Monthly cost bands for four services

Prices are USD list figures from the vendors' own pages on 1 November 2026. After-hours volume for a small MSP is low, so the entry tiers matter most.

Connect on LinkedInSagnik Bhattacharya

Smith.ai also sells human-staffed plans for IT firms, starting at $300 for 30 calls on its human-first plans, and describes a hybrid in which AI takes routine intake and live receptionists take urgent outages. Its IT page lists no prices for that hybrid, so ask for a quote if you want it.

Subscribe on YouTube@sagnikteaches

A six-engineer MSP's after-hours bill, worked through

Say a six-engineer MSP supporting about 45 client businesses gets 30 after-hours calls a month between 6pm and 8am and at weekends. Twelve are genuine incidents, ten are routine requests (a printer, a locked account, a new starter), five are wrong numbers or sales and three are people chasing an earlier ticket. Average length is four minutes, so about 120 minutes.

• Smith.ai Free: 25 calls are free and five more at $3 add $15, so $15 if you enable paid overage. If the real calls are 25 or fewer, it is $0. Known spam does not count.
• Dialzara: 120 minutes exceeds the $29 plan's 60 by 60 minutes, so $29 plus $28.80 is $57.80; Business Pro at $99 covers it with room.
• Goodcall Starter: $79 for up to 100 unique callers. With around 20 distinct numbers, you pay for headroom you will not use.
• Rosie Professional: $49 covers 120 minutes comfortably.
• Smith.ai human-first: $300 for 30 calls.

So an AI receptionist for this firm runs about $0 to $99 a month, rising to a few hundred once volume or staffed cover is added. The choice is not about the bill. Compare it with what a missed P1 costs: a client without email or a file server over a weekend can cancel a contract worth thousands a year. Smith.ai's Pro plan at $150 adds onboarding help and is worth it if nobody on your team wants to configure call flows.

The ticket fields that make an after-hours call actionable

A call is only as useful as the ticket it leaves behind. Smith.ai's IT page describes structured intake for issue type, severity, affected systems, number of users, error messages and urgency. Build the same in whichever tool you use. Here is a filled-in example, illustrative:

Client: [company name], caller [first name], callback 555-0142 (matches contact on file: yes)
Received: Saturday 21:14
Issue type: server / file access
Severity: P2 (work stops for some users, no workaround yet)
Affected: file server, about 8 of 25 staff
When it started: around 20:30
Error message as read by caller: "network path not found"
Already tried: restarted PC, same on second PC
Recent changes: none known
Action taken by assistant: logged, escalated to on-call by SMS
Told caller: an engineer will review and respond; no fix time promised

The "matches contact on file" line matters. The assistant should compare the caller's number and name with the client's authorised contact list where the integration allows, and log "no match" otherwise.

Define P1, P2 and P3 once, in your own words, and give the assistant those definitions. A common split, adaptable:

• P1: whole-site outage, no email or internet for the whole company, ransomware or suspected breach, phones down for a business that lives on them.
• P2: a server, application or department down but staff can work around it, or one VIP completely blocked.
• P3: one user or one device, no deadline, can wait for the next working day.

Getting the call into your PSA

There are three routes, in order of effort.

1. Native integration. Smith.ai's IT page says tickets can be created automatically in PSA platforms such as ConnectWise, Autotask and Kaseya. It does not say which plan includes this, and the AI pricing page describes its native integrations as "Clio, Zapier & more" without naming a PSA, while reserving unlimited custom integrations for the $500 Enterprise tier, so get the answer in writing before you budget on the free or Pro plan. Ask too which fields map and whether the ticket lands in the right queue and board.
2. Email to ticket. Most PSAs offer a mailbox that turns an incoming email into a ticket, and most receptionist services can email each summary. Use a fixed subject format such as "[AH] [Client] [P2] [issue]" so a mailbox rule can set priority and queue. Check your PSA's documentation to confirm it supports this and how it sets priority.
3. Automation tool. Zapier's ConnectWise Manage and Autotask apps both include a Create Ticket action. ConnectWise needs a summary, the client company and a record type; Autotask needs the company's ID. Either way the automation must match the caller to the right client before it can file anything, so keep a clean company list. All four receptionists in the table list Zapier connections. Expect another monthly fee for the connector, so add it to the cost table.

The email route is the one I would start with, because it uses nothing that you have not already got. Add one safeguard: send a copy of every summary to a monitored shared inbox, so a PSA outage at 2am does not drop the call.

Escalation rules: when the assistant must wake someone

Write these as plain rules, with numbers.

1. Any P1 keyword triggers an immediate text and call to the on-call engineer. Keywords: "ransomware", "encrypted", "can't open any files", "everything is down", "hacked", "breach".
2. If the on-call engineer does not acknowledge within 10 minutes, the assistant escalates to the second contact and then the service manager.
3. P2 calls go to the on-call engineer by text without a call, to be actioned within 30 minutes if the client has an after-hours agreement, otherwise queued for the morning.
4. P3 calls are logged only.
5. A caller who says it is urgent, or asks for a person, is transferred, whatever the severity the assistant assigned.

Ransomware deserves its own line. A caller reporting a ransom note should hear a calm script and instructions that your own incident process already gives, such as "disconnect the machine from the network if you can do so safely, do not switch it off, and do not click anything", with an immediate escalation. Have your security lead write those words and keep them short.

What it must never promise or do

• No fix times. "Someone will be with you shortly" is a promise. Say "an engineer will review this and respond", plus the contractual response time only if it is true for that client at that hour.
• No troubleshooting. No "try restarting the router". A wrong step can destroy evidence or make an outage worse.
• No access changes. No password resets, MFA changes or new accounts. Those need your verification procedure.
• No confirming contracts or pricing. It should not say whether a caller is covered. After-hours cover often depends on the agreement, and a wrong yes creates a dispute.
• No commentary on security incidents. It does not say what happened, whose fault it is or whether data was lost.
• No pretending to be human. Asked directly, it admits to being automated and offers a transfer.

Eight test calls to place before it goes live

Write a short script for each and record whether the ticket came out right.

1. A known contact reports one user's printer jam. Expect a P3 log.
2. A known contact says nobody in the office can reach email. Expect P1 or P2 and an escalation.
3. An unknown number asks to reset the managing director's password. Expect a refusal, a log marked unverified, and an alert to a technician.
4. A caller whispers that every file has a strange extension. Expect a P1 and a text within a minute.
5. A caller asks the assistant to fix it: "Just tell me what to type". Expect a polite refusal and a ticket.
6. A caller asks, "Am I covered after hours?". Expect a refusal to confirm and a note for the account manager.
7. A caller who is angry about an earlier ticket. Expect a log linked to that client and a note for the service manager.
8. A wrong number selling something. Expect a polite end and no ticket.

Smith.ai's free plan lists 25 test calls and 50 simulated calls a month, so this exercise can cost nothing there. Goodcall includes AI test runs by plan. Run each scenario at least twice, in different wording, before you trust it.

A realistic failure, illustrative: a tester said, "the internet's down", and the assistant logged it as P3 because the caller sounded relaxed. The fix was a rule that any mention of internet, email or server being down for more than one person is at least P2, regardless of tone, and a question added at the start: "Is it affecting just you, or other people too?"

Where the AI stops and a person should take over

Use a person, or a hybrid service, for: callers from clients who pay for a premium response tier, any P1 that has not been acknowledged by an engineer, and calls where the caller is distressed. Smith.ai describes routing urgent outages and frustrated clients to live receptionists, which is the model most MSPs end up with: AI for the routine ten out of thirty calls, people and engineers for the rest. If you want the broader view of help-desk tooling after the call is logged, the help-desk tool comparison and the ticket automation tutorial cover that side, and the after-hours phone agent set-up covers the general steps.

Reviewing the first month

After thirty days, pull every after-hours ticket the assistant created and mark each correct, partly right or wrong on severity, client, and issue type. A target worth aiming for is no wrong P1 or P2, and at most one in ten tickets needing a callback to clarify. Count how often an engineer was woken for something that could have waited; if it is more than twice a month, tighten the P2 rule. Count calls where the assistant said something on the never-promise list by listening to a sample of recordings, and fix the script for each.

Then decide on the plan. If you used 12 calls and 30 minutes, the free or $29 tier is plenty. If urgent calls were mishandled in the first month, spend money on a hybrid plan instead of a bigger AI plan: the gap is judgement, not volume. The PSA AI comparison is worth reading if your tickets then need triage inside the platform.

Telling clients, and what goes in the contract

Clients should hear that after-hours calls are answered by an automated assistant that logs and escalates. A short notice on your emergency number page and in the next service review is enough: say what it can do, that it records calls, and that a person decides response. Check that the recording and storage fit your client agreements and any confidentiality terms; a client's voice describing their network is sensitive. Update the after-hours section of your service agreement so it says tickets are logged by the assistant and response times start from the logged time. Keep the wording factual and have your lawyer review it.

A final practical point: put the assistant's own number on your out-of-hours voicemail message and in each client's contact card, not only your main line, so a client in trouble reaches the right place on the first try.

The renewal risk behind one unanswered P1 call

The plan fee deserves a comparison with the thing it protects. Take an illustrative client paying $1,800 a month for managed support. If a Saturday outage goes unanswered until Monday and the client cancels at the next renewal, the loss is $21,600 a year, plus the cost of replacing that client. Say that outcome is rare, a one in forty chance per unanswered P1. If the MSP expects four genuine P1 calls a year across all clients, the expected annual loss from missing them is about four times 1/40 times $21,600, or $2,160. A $99 plan costs $1,188 a year; the $29 plan costs $348. Even with these deliberately modest numbers, cover costs less than the risk it removes, and the hybrid plan starts to make sense only if your expected P1 count is higher. The figures are illustrative; replace the contract value, the P1 count and the cancellation odds with your own.

Different clients, different after-hours rules

A single set of rules rarely fits every client. A simple way to keep the assistant honest is to group clients by what you have agreed with them, and give the assistant a one-line rule per group.

• Premium after-hours cover. Any P1 or P2 pages the on-call engineer at once. The assistant may say that an engineer will call back within the agreed window, quoting the window from your contract.
• Standard cover. P1 pages the engineer; P2 and P3 are logged for the morning. The assistant tells callers that the issue will be picked up at the start of the next working day unless it is a full outage.
• No after-hours agreement. The assistant logs everything and tells the caller that the team will respond next working day. If the client wants emergency help, the call is passed to the account manager, who can agree a one-off rate.

The assistant needs the client list to do this. Most services let you upload a list of numbers and company names, or maintain it through a CRM or Zapier link. Keep it short and updated: a quarterly check against your PSA's contact list catches leavers and new starters.

Sample calls, with what the assistant should log

Three short illustrations show how the same assistant should behave across severities.

A routine request. A caller from a small client says a new starter begins Monday and needs an account. The assistant should log it as P3 under onboarding, capture the starter's name, start date and manager, and tell the caller the team will confirm the next working day. It should not create the account or give a password.

An outage with partial information. A caller says, "Nobody can print and the shared drive is slow." The assistant asks how many people are affected and since when. It then logs P2 if more than a few users are affected, and sends an alert to the on-call engineer without offering a cause or a fix.

A suspicious call. A caller who is not on the client list says they are a director and need the remote access settings changed urgently. The assistant refuses, logs the call as unverified, flags it for the security contact, and does not repeat any client names or system details back to the caller. This is the kind of call that attackers rely on a friendly front desk to answer.

Keeping recordings, summaries and client data under control

A call about a client's network is client data. Before you switch the service on, ask the vendor in writing where recordings and transcripts are stored, how long they are kept, whether they are used to train models, and how you delete them. Smith.ai's pricing page lists recording, transcription and summaries on every plan and Goodcall publishes call-history retention of 30, 90 or unlimited days by plan. Set the shortest retention that still lets you review quality. For recordings that contain security details, decide who can listen: ideally your service manager and one deputy, not the whole team. The privacy terms checklist shows what to check in the vendor's terms, and your own client contracts may add requirements.

A two-week rollout plan

1. Days 1 to 2. Write the P1, P2 and P3 definitions, the client groups and the escalation rules. This takes about three hours.
2. Days 3 to 4. Configure the assistant, build the client list, set the ticket route (native, email or automation) and run the eight test calls.
3. Days 5 to 7. Fix failures, rerun the tests, and send the first client notice.
4. Days 8 to 14. Live, but with a human shadow: every after-hours summary is read by the on-call engineer the next morning, who marks it right or wrong. Keep your old voicemail as a backup number for the first fortnight.

At the end of the fortnight, hold a 30-minute review with the on-call team, go through each ticket created, and agree the changes. A small MSP can do this entirely itself; if you would rather talk it through with someone who has seen several call set-ups, book an AI implementation consultation through the contact page.

Questions MSP owners ask about AI phone cover

Can an AI receptionist create tickets in my PSA?

Smith.ai's IT-services page says tickets can be created automatically in PSA platforms such as ConnectWise, Autotask and Kaseya, but not which plan includes it, so ask. For other tools, most PSAs have a mailbox that turns an email into a ticket, so a summary emailed to it often works. Confirm your own PSA's intake options before you buy.

Should the assistant try to fix problems on the call?

No. It should collect symptoms and impact, then log and escalate. Troubleshooting by a machine on a live call risks wrong advice and, worse, someone acting on it. Smith.ai's own page says its receptionists coordinate with your support tiers but do not give technical advice, which is a sensible rule to copy.

What if the caller isn't a known client contact?

Log the call as unverified, record what they said and a callback number, and do not act on requests such as password resets or access changes. Attackers often phone support, so identity checks belong to your technicians using your normal verification steps, never to the receptionist.

Is a human-staffed service better for urgent outages?

For P1 calls it can be, because a person can judge tone and urgency. A hybrid works well: the AI handles routine intake and common questions, and urgent calls go to a live person or on-call engineer. Smith.ai describes exactly that split for IT firms.

Further reads

• How MSPs Use AI to Write Quarterly Business Reviews — Uses the logged tickets in client review reports.
• AI IT Documentation for MSPs: Client Runbooks From Screen Recordings — Turns recurring incidents into runbooks.
• How to Measure an AI Receptionist's Return in the First 90 Days — How to judge payback in the first 90 days.
• MSP Automation Ideas: Five AI Workflows and Their Set-Up Time — Five other workflows to automate, with set-up times.
• What Is the Cheapest Way to Handle Enquiries Out of Hours? — Cheaper routes to out-of-hours cover for any business.
• How Small IT Support Businesses Use AI to Resolve Tickets Faster — Resolving tickets faster after they are logged.
• Automated Managed Service Provider: Hardware Refresh Plans From Asset Lists — How a small managed service provider groups a client's asset list into replace-now, next-year and later buckets, then builds the budget table.
• AI Answering Service for Home Care Agencies: After-Hours Call-Outs — How a home care agency can use an AI answering service for sickness call-outs and urgent calls: scripts, escalation timers, message log and monthly costs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai pricing and IT managed-services pages, Goodcall pricing, Rosie pricing, Dialzara pricing, Zapier's ConnectWise Manage and Autotask app pages (all checked 1 November 2026).