On a Tuesday morning in early March, the phone rings at a small tax practice. A client wants to know whether her T4 slip has arrived. The person who picks up is a CPA whose time bills at $170 an hour, and she spends four minutes on the call.
That call is not billed. It is capacity. Four minutes of the most expensive labor in the building, spent on a question a calendar reminder could have answered.
In tax season, the people answering the phone are the people who bill by the hour. Every routine call they take is an hour that does not go back onto a client's return.
The crush, and why it concentrates
The peak is not a guess. It is a calendar.
In the United States, individual returns are due April 15, 2026. In Canada, the filing deadline is April 30, 2026. For self-employed filers in the US, the deadline stretches to June 15. Then the extensions land, pushing a second, smaller surge toward October for returns that were set aside in April.
Between January and April, four things happen at once. Clients receive their slips and want to talk about them. New clients decide this is the year to get organized. Existing clients ask where their refund is. And every question a client could answer with a portal login gets asked by phone instead, because the phone is what people trust.
The work does not spread evenly across the year, and it does not spread evenly across the day. It lands in a compressed window, on a small team, while that same team is trying to do the filing work that actually pays the bills.
The evidence for how hard that window is comes from the busy-season hours themselves. In surveys of UK accounting staff, 78% reported working more than 50 hours a week during busy season. Peer-reviewed research has documented busy-season burnout as a recurring pattern in the profession. The people answering your phone in March are already past a normal week before the first call arrives.
The arithmetic that matters
Let's put a number on it, with the caveat up front: this is an illustration, not your practice. Substitute your own hours and your own rate.
The anchor is real. The AICPA PCPS 2025 MAP Survey puts the median net hourly billing rate at $170 per hour, up 6.9% year over year. That is an industry body survey, and it is the strongest accounting-specific figure available.
For Canadian firms, Sapere, a Canadian firm guide, reports billed rates of roughly $150 to $400 an hour at small and mid-sized practices. That source is weaker, so treat it as a range rather than a precise number.
Now the illustration.
Say a four-person practice spends four hours a week on routine, non-billable phone work during peak season: status checks, document questions, scheduling, "did you get my email." That is one CPA, one morning a week, gone.
| Example figure | |
|---|---|
| Routine phone hours per week | 4 |
| Billing rate (AICPA median) | $170/hr |
| Billable capacity consumed per week | $680 |
| Weeks from January to the end of April | 16 |
| Billable capacity consumed across the season | $10,880 |
Now the other side of the ledger. An OrdrNow AI receptionist is $199 per month plus a one-time $199 setup fee, including 300 minutes of calls. Over four months that is about $796, or $995 once you include the setup fee.
At $170 an hour, the monthly fee equals roughly 70 minutes of billable time. That is the whole comparison. If the assistant gives back more than about an hour and ten minutes of a biller's month, it has paid for itself. Everything above that is margin.
Run it with your own numbers. If your routine call load is two hours a week, halve the season figure. If you bill at $250, the same hours are worth more. If your team already has a full-time administrator on the phone, the question shifts to what else that administrator could be doing instead. The point is not the specific total. The point is that the phone has a cost, and most practices have never written it down.
Why the phone breaks in tax season
The clearest picture of what peak tax-season call demand looks like does not come from a firm. It comes from the tax authorities, whose phone systems absorb the same concentrated surge at national scale.
In the US, the IRS reported roughly 27.9 million calls to its automated lines during filing season, with only about 32% of calls answered by a customer service representative and an average hold of about 8 minutes.
In Canada, the Auditor General's June 2025 report found the CRA receives about 32 million calls a year, answered only 18% of them within its service standard, held callers for an average of 31 minutes, deflected 8.6 million calls, and in June 2025 met its standard on just 5% of calls.
Read those as a shape, not a benchmark. These are the tax authorities' phone systems, not yours. They describe what happens when tax-season demand hits a phone line all at once. Your practice is not the IRS, and your callers are not chasing a missing refund. But the pattern is the same one your front desk lives through every March: more calls than people, arriving faster than they can be handled, with the most valuable person on the team pulled into the queue.
When the queue wins, calls go unanswered. And an unanswered call in a trust-based business is rarely just a missed message.
What an AI receptionist can and cannot do here
The honest split matters more than the sales pitch.
What it handles
- Answering every call, 24/7, including the ones that arrive after hours and on weekends
- Capturing caller details, so a callback starts with structure instead of a voicemail
- Answering FAQ and policy questions from a knowledge base: hours, location, what to bring, how the engagement process works, where to send documents
- Booking and rescheduling consultations directly into your calendar
- Sending confirmation and reminder texts
- Triaging genuinely complex or high-value commercial enquiries and transferring them to a human
What it does not handle
- Tax advice
- Judgement calls about a client's position
- Anything that requires access to a client's files
- Anything that depends on integration with your practice-management or tax software. That integration is not built. Custom integrations are available on Enterprise only, and they are custom work, not a switch you flip.
If a caller asks a substantive question, the assistant's job is to capture the details and route it, not to answer it. That boundary is what keeps the tool useful instead of dangerous.
The limits, stated plainly
Any article about AI answering phones for accountants that skips this section is selling, not explaining.
Complex, multi-issue calls are where these systems are weakest. The best available benchmark is VAmoS Bench (arXiv, July 2026), which ran 3,300 simulated calls across 11 systems. Overall task completion was 63.1%. On simple tasks it was 65.8%; on complex tasks it fell to 52.2%. A client calling with a tangled question about a rental property, a foreign account, and a late slip is exactly the kind of call where you want a person, and the data agrees.
Confidentiality deserves a hard look. Tax clients share financial and personal information under an expectation of privacy. The assistant should collect only what scheduling and triage require, and sensitive details should route to a human rather than be explored on the call. Whatever you deploy, read how it handles call data and retention before you point a real number at it.
Callers may not want this. In a 2025 survey of 6,000 adults across the US, Canada, and the UK, 33% said they would hang up if connected to an AI. A separate Gartner study of 5,728 customers found 64% would prefer companies did not use AI in customer service. For a profession built on trust and long relationships, that resistance is real and cannot be waved away. The mitigation is not to hide the AI. It is to be transparent about it, to make reaching a human easy, and to keep the assistant on the calls where a person is not required.
The practical read: use the assistant for the routine calls, and make the handoff to a human obvious and fast.
A note on a statistic you will see everywhere
Search for this topic and you will find the claim that tax season call volume spikes 200% to 400%. That figure is vendor marketing. It has no traceable primary source, and this article does not rely on it. The same goes for the claim that "AICPA says practices miss one in four calls," which is attributed to AICPA without any traceable AICPA source and looks fabricated.
What the evidence does support is narrower and still useful. Tax-season demand is concentrated into a short window, and that window lands on a workforce that is already stretched. The finding that 78% of accounting staff work more than 50 hours a week in busy season is a survey result, not a law, but it describes a real condition. You do not need an inflated spike number to see the problem. You can look at your own March.
What to do next
You do not need to decide anything today. Two useful steps.
- Hear the thing work. Call a live demo at the demo hub and try to break it. Ask a scheduling question, then ask something only a CPA should answer, and listen to how it routes.
- Run your own arithmetic. The full cost breakdown, including what $199 actually buys, is in How Much Does an AI Receptionist Cost?.
If the numbers say the phone is cheap to leave alone, leave it alone. If they say otherwise, you will know what the alternative is worth.