biz59Support with two people: the response times customers expect from a small SaaS and how to meet them
Published benchmarks (62% of companies never answer an email; a one-hour first reply puts a SaaS in the top 15%), what Plausible, Transistor, Buttondown and Fathom promise, how two people cover a week without 63-hour Friday gaps, 2026 prices of shared inboxes, help desks and AI replies ($0.75 to $2.00 a resolution) with a worked example, where an AI answer angers customers, and six numbers to check every Monday.
In most small software companies support is two people who also do something else: a founder who answers in the evening and a developer or customer success person who answers between other jobs. The questions are always the same. How fast do customers expect a reply? Is a help desk worth paying for yet? Should an AI answer first? And what do you promise on the website so nobody is surprised?
Below are the published benchmarks, what small vendors actually promise, the arithmetic of covering a week with two people, the 2026 prices of shared inboxes, help desks and AI replies, and the six numbers worth looking at every Monday.
What customers expect, and what vendors deliver
Read the first block again: in a mystery-shopper test, most companies never answered at all. A two-person team that replies to every email within one business day is already better than most of the market, and one that replies within an hour during stated hours is in the top 15%. The bar is not instant replies around the clock. The bar is a reply, by the time you said, from someone who has read the earlier messages.
What small vendors actually promise
The small SaaS companies that publish a support promise keep it short and specific:
None of them promise 24/7 or a 15-minute response. What customers forgive in a small vendor is slower hours, if the hours are stated and kept. What they don't forgive is silence, being asked again for what they already wrote, and a bot with no way out.
The arithmetic of two people
Start from volume. Intercom's August 2026 analysis of 5,280 early-stage workspaces found a median of 33 conversations a month, and 9 a month for teams of one or two people. A SaaS with a few hundred paying customers typically sees more, but rarely more than a few a day. Take a product with 400 customers and 60 conversations a week, 12 minutes each on average including the look-up: that is 12 hours of work a week. Two people can carry that alongside their other jobs; the problem is not the hours, it is the gaps between them.
That table is why you state the hours in business time. "First reply within 4 business hours, Monday to Friday, 9:00-18:00 UK time" is a promise two people can keep. "We reply fast" is not, because the customer who writes at 6pm on Friday will measure you in calendar hours.
A useful split for two people: one is first responder on a given day, the other is backup and takes anything technical. First responder means reading every new message within the hour, replying fully to what can be answered in five minutes, and sending a short holding reply with a real time ("I've asked Sam, who built the export; you'll hear by 3pm") for the rest. Swap days. Nobody loses a whole day to the inbox, and nobody is the only one who knows how the product works.
Shared inbox or help desk: the 2026 prices
For two people the choice is less about price (all of these are under $1,400 a year for two seats) than about where customers write. If they email, start with a shared inbox: Help Scout's free plan or a Google Groups collaborative inbox is enough until you need timers. Move to a help desk when one of three things happens: you promise a response time and need a clock to show you're keeping it, you have more than one product or channel to route, or someone outside support (a developer, the founder) needs to see the history without being in the inbox.
Saved replies (macros) pay for themselves first. Look at the last 100 messages, group them, and write a reply for the ten most common questions: password reset, invoice copy, cancel and refund, plan change, the export everyone asks about. Each should be a draft a person edits, not a form letter; customers can tell. If a macro gets used more than ten times a month, the real fix is in the product or the docs.
Where an AI reply helps, and where it angers
Customers are less hostile to AI answers than they were two years ago, as long as a person is reachable. Gartner's July 2024 survey of 5,728 customers found 64% would prefer companies didn't use AI for service, and 53% would consider switching; the top fear was that reaching a human would get harder. Its August 2026 survey of 3,566 customers found 50% saying generative AI makes service easier, and 87% saying a path to a human agent is essential. Five9's March 2025 survey: 72% are open to AI if they can escalate to a person.
The failures are specific. In April 2025 Cursor's AI support agent, "Sam", told users that logging in on a second device was against a new policy; there was no such policy, users cancelled, and the company apologised and refunded. In February 2024 a Canadian tribunal ordered Air Canada to pay C$812.02 after its chatbot misstated the bereavement fare rules, and rejected the argument that the chatbot was "a separate legal entity". Anything an AI tells a customer about money, policy or their account is your promise.
Vendors quote high resolution rates (Intercom's site says Fin averages 76% across 12,000+ customers); measure your own over a month before believing any of them, and count a "resolution" only when the customer didn't write back about the same thing within a week. If you want the full cost model for a support bot against hiring, our support bot cost calculator runs it with your volumes.
Six numbers to look at every Monday
These need nothing more than tags and a spreadsheet. The last two are the ones that make support smaller: every recurring reason you remove from the product is a ticket you never answer again.
What to put on the website
When something does break, the rules change: updates every 30 to 60 minutes whether there is news or not, on a status page rather than in hundreds of replies. Our piece on what a small SaaS publishes after an outage has the timings and the wording.
Should an AI answer first? Put in your monthly tickets, the share the bot resolves, how many come back, its price per resolution and your hourly cost, and compare people only with bot first, per resolved ticket. Runs in your browser.
Open the support bot cost calculatorSources
Benchmarks: SuperOffice customer service benchmark (updated Mar 2023); Jitbit average support metrics (updated Apr 2026); Freshworks benchmark report 2023; Zendesk CX Trends 2026 (Nov 2025); Intercom on early-stage volumes (Aug 2026).
Small vendors: Plausible, Transistor, Buttondown, Fathom, Basecamp, all read 24 Oct 2026.
Prices (read 24 Oct 2026): Help Scout, Front, Missive, Zendesk, Freshdesk, Intercom, HubSpot Service Hub, Zoho Desk, Fin.
AI and customers: Gartner, July 2024; Gartner, August 2026; Five9, March 2025; AI Incident Database on Cursor's support bot (Apr 2025); CX Today on Moffatt v. Air Canada (Feb 2024).