AI Customer Support for One Person: What It Really Costs

🆕 Rewritten August 2026 — this page was a fabricated case study, and we need to say so plainly. The previous version opened with "Meet Sarah (name changed for privacy)", gave her 340 customers and $9,800/month in revenue, reported her support metrics as analysed over three months, and quoted her four times as though we had interviewed her.

She never existed — so nothing was ever measured on the business the numbers were attached to: not the 340 customers, not the revenue, not the 200 requests a month, not the 80-hours-to-8-hours saving, and not the response time.

One clarification we owe you, because it cuts the other way. The 72% resolution rate is a real figure — we measured it ourselves over 30 days in our five-chatbot comparison, on our own site, against a 27-question FAQ. What was fabricated was re-attributing it to an invented business with invented volume, and then printing it as a selling point on an affiliate card for Tidio, a commercial partner. Real number, invented provenance. That is still the worst thing we've found on this site, and a made-up testimonial for a product we earn from is not a stale-price problem. What federal rules now say about fake testimonials is set out in AI testimonials and the FTC fake reviews rule.

Everything traceable to her is deleted, quotes included. What replaces it is built from the vendors' own pricing pages, a tribunal decision, two consumer surveys with published samples, and a field experiment on 5,179 support agents — and it reaches a less flattering conclusion about the tool we're paid to recommend.

If you run a business alone, customer support is the function that scales worst. Every new customer adds questions, and there is no version of you that answers them at 2am.

So the pitch for AI support lands hard on solopreneurs, and it usually arrives with a resolution percentage attached. This article is about why that percentage can't be compared between vendors, what the meter actually costs, and the legal precedent that says you own whatever your bot tells people.

The billable unit is the whole story

Start here, because everything else follows from it. Three vendors, three definitions — each taken from the vendor's own documentation.

Read the first row again. Tidio charges when its bot replies — not when it helps. A conversation where Lyro answers wrongly, the customer gets annoyed and escalates to you is a billable Lyro conversation. That is not a gotcha; it's Tidio's published definition, and it is a perfectly normal way to price a service.

But it means the resolution rates you see advertised are measuring different events:

• Tidio's own claim: "Lyro AI boosts your resolution rate to 67% on average."
• Fin's own claim: 76% average, with "many customers achieving over 85%."

Those two numbers cannot be compared, and neither can be compared to a third vendor's. Note Intercom's middle condition especially — "they don't ask for more help after Fin responds" counts a customer who simply gave up as a resolved outcome. Again: that's Intercom's own published billing definition, not our characterisation.

The practical rule: before believing any AI support percentage, ask what event it counts. If the vendor won't say, the number means nothing.

What Tidio actually costs when you use it

We recommended Tidio in our chatbot comparison, and we still think it's the right pick for most solopreneurs. But the price we published here was wrong in a way that matters, and it's wrong in the vendor's favour.

The old version of this page said "Tidio (Lyro AI) — $29/month." Lyro conversations are metered separately from the plan. What's included is 50 Lyro conversations, one-off — Tidio's own wording, a lifetime allowance rather than a monthly one.

Then we got the rate wrong too, and in the vendor's favour again. We published "$0.50 per conversation" and built a cost table on it. Tidio's Lyro page does say "Pay just $0.5 per conversation" — but read the FAQ on the same page and it says "A Lyro conversation starts from $0.5." That is a floor at volume, not a rate you can buy.

What Tidio's own pricing page actually sells, re-checked 4 September 2026, is an add-on:

"Starts at $32.50/mo. From 50 Lyro AI conversations"

$32.50 for 50 conversations is $0.65 each, not $0.50. And you buy a tier, not conversations — the ladder runs 50, 100, 150, 200, 300, 500, 750, 1,000. So a solopreneur handling 30 AI conversations a month doesn't pay for 30. They buy the 50-tier:

We previously published $44 for that same 30-conversation case. It was too low by about 40%, because it assumed a flat $0.50 meter that isn't how Lyro is sold.

What we can't tell you: the prices of the intermediate tiers. Tidio publishes the entry price and the ladder's shape but not each rung, and we don't have an account to read them from. Note also that the paid add-on's allowance resets 30 days from payment, unlike the free 50, which is lifetime. Check the meter in your own account before budgeting — that advice was right the first time, and now we know why it mattered.

The rest of the stack, corrected

The old version budgeted $25/month for Help Scout. Help Scout has a free plan: 5 users, 1 inbox, 1 Docs site. For one person with one support address, that is the plan. ($25 is the per-user Standard price; $21 billed annually.)

Corrected 10 September 2026: Make Core was in the monthly column at $9, its annual price. Billed monthly it's $10.59, so the monthly subtotal is $59.59, not $58.

Plus the Lyro meter on top — $15 at 30 conversations a month, $75 at 150.

And the genuinely free route, which deserves more than a footnote: Tawk.to gives unlimited conversations, agents and chat history at no cost, and HubSpot's free CRM includes a live chat widget if you already use it — see the best free CRMs. Neither does AI resolution. That's the entire trade, and at low volume it is often the right one. More of this thinking in the $0 solopreneur stack.

You are liable for what your bot says

This is the part no solopreneur guide mentions, and it is settled enough to plan around.

In Moffatt v. Air Canada, 2024 BCCRT 149 (British Columbia Civil Resolution Tribunal, February 2024), the airline's chatbot told a customer he could apply retroactively for a bereavement fare. He couldn't — a different page on Air Canada's own website said so. He applied, was refused, and took it to the tribunal.

Air Canada's defence was that "the chatbot is a separate legal entity that is responsible for its own actions."

The tribunal called that submission remarkable, and rejected it:

"While a chatbot has an interactive component, it is still just a part of Air Canada's website... it should be obvious to Air Canada that it is responsible for all the information on its website."

Damages were $650.88 for negligent misrepresentation, with the total order around $812 once interest and fees were added. Small money for an airline. The principle is what matters: your bot's answer is your answer, and "the AI said it" is not a defence.

Now apply that to a one-person business. You have no legal review, no compliance sign-off, and a knowledge base you wrote in an afternoon between client calls. If your FAQ says refunds are available for 30 days and your actual policy is 14, the bot will say 30 — confidently, to everyone, at scale, at 2am.

What follows practically:

• Your refund, cancellation, delivery and pricing policies are the highest-risk documents in the knowledge base. Get those exactly right before anything else
• Configure the bot to escalate rather than guess on anything touching money or entitlement
• Re-check the knowledge base whenever a policy changes — this is the automation that goes stale silently, the same failure mode we describe in automating client follow-ups

Customers can tell, and they don't like it

Two surveys with published samples, both worth knowing before you put a bot in front of your customers.

SurveyMonkey polled 2,017 US adults, 10-11 December 2025 (non-probability online panel, margin of error ±2.5 points):

• 79% strongly prefer interacting with a human over an AI agent
• 84% believe human agents are more accurate
• 81% believe AI is used primarily to save the company money, not to improve service
• 89% say companies should always offer the option to speak with a human

A separate SurveyMonkey study put Net Promoter Score for AI interactions at −66, against +6 for human ones. People cited humans understanding their needs better (61%), giving more thorough explanations (53%) and being less likely to frustrate them (52%).

Gartner surveyed 3,566 B2B and B2C customers in February and March 2026 and found 87% say it is essential that companies using generative AI in service provide a route to a human agent. Gartner's own framing is the useful one: customers are increasingly comfortable letting AI resolve issues — and push back hard when it becomes a barrier to reaching a person.

So the escalation path is the product, not the fallback. A visible, easy, obviously working route to you is what makes the bot acceptable. Hiding it is what generates the sentiment above.

Then why do it at all?

Because the productivity evidence is real, and it's specific about who benefits.

In Generative AI at Work — Brynjolfsson, Li and Raymond, published in the Quarterly Journal of Economics in May 2025 — the authors studied a generative-AI assistant rolled out across 5,179 customer support agents, measuring issues resolved per hour:

• +14% on average
• +34% for novice and low-skilled workers
• Minimal effect on experienced, highly skilled workers

Customer sentiment improved too. The mechanism the authors describe is that the AI spreads what the best agents already know to everyone else.

Which tells you exactly who this is for. If you are already fast and good at support, this buys you less than the marketing implies. If support is the thing you do badly because you resent it — replies going out three days late, the same question answered differently each time — you are the case the research is most encouraging about.

And there is a reason that applies only to a business of one, which no study measures: you are asleep for a third of the day, and a bot that gives a decent answer at 2am beats a perfect answer on Thursday.

The build, honestly

No invented timings. This is the order that makes sense given everything above.

1. Write the knowledge base first, and treat it as a legal document. Start with your 20 most-asked questions. Policies on money — refunds, cancellations, billing, delivery — get written last and checked twice, because those are the answers that cost you if the bot gets them wrong. Notion's free plan gives unlimited pages for a single-member workspace.

2. Spend the 50 free Lyro conversations deliberately. They're one-off, so don't burn them idly. Ask your own bot the fifteen questions you actually get, plus three you know aren't documented, and watch whether it escalates or invents. That behaviour on unknown questions is the thing you're evaluating.

3. Put the human route where nobody has to look for it. 87% of customers want it. Make it a visible option in the widget, not a maze.

4. Only then wire up automation. Bug reports and feature requests routed into Notion, billing keywords flagged as priority, a weekly digest — Make Core — $9 a month billed annually, $10.59 monthly — handles all of it on 10,000 credits.

5. Watch the meter for the first two months. You now know it's per conversation, not per resolution, and that you buy a tier rather than the conversations you actually use. Every time you cross a rung on the ladder, re-derive the cost rather than assuming the plan price still describes your bill — the habit we argue for throughout the real cost analysis.

[ADD YOUR OWN EXAMPLE HERE] — if you've run an AI support bot long enough to know what it gets wrong on your product, that belongs here and is worth more than any benchmark.

Is this worth it for you?

Probably yes if: support is repetitive and documentable; you're slow to reply today; your policies are stable enough that a knowledge base stays true; and your volume is low enough that the meter stays small.

Probably not if: every question is unique; your product changes faster than you can document it; your answers carry money or legal consequence in a way you can't safely delegate; or you get few enough questions that answering them yourself is genuinely fine.

And regardless: publish the human route, own what the bot says, and price the meter rather than the plan.

For the wider toolkit, the 2026 solopreneur AI stack has the buying order, tools under $20 a month covers the budget end, and can AI replace a virtual assistant? works through the same reliability question for admin work rather than support.

---

Sources: Tidio's, Intercom's and Help Scout's own pricing and product documentation (checked August 2026); Moffatt v. Air Canada, 2024 BCCRT 149; SurveyMonkey (n=2,017 US adults, December 2025, non-probability panel); Gartner (n=3,566 customers, February-March 2026); Brynjolfsson, Li & Raymond, "Generative AI at Work" (NBER w31161; Quarterly Journal of Economics, May 2025). An earlier version of this article was built on an invented customer, including invented quotations attributed to her and invented metrics used as selling points on an affiliate placement. All of it has been removed rather than reworded. Links marked "Paid link" earn us a commission if you sign up through them, at no extra cost to you; no other link here pays us — how we make money.