Still answering “what time do you close?” at 9:40pm?

Customer email that answers itself.

The repeat questions get answered without you. OttoMail runs on what you already have — your Gmail and one Google Doc you control — so there’s no one to hire and no new system to run.

Nothing sends on its own until you’ve approved 50 replies of that type and switched it on yourself.
Every claim cited to your docYour never-list always gets a person
Review queueApprove 1 clear
Friday walk-ins & last seatingGraduated — sending itself

We do take walk-ins on Fridays — the bar and terrace are first-come, first-served.1 Last seating is 8:30pm.2 Hope to see you!

Cancel sendSends in 3 minutes unless you cancel
Nut allergy — birthday dinner on the 14thYour never-list

You marked this type human-only. It always gets a person, with a draft already waiting.

What it’s costing you.

Your most expensive people are answering the cheapest questions.

The owner answers the parking question between deliveries. The GM stops a line check to confirm what time the kitchen closes. Field studies of interrupted work put the recovery at about 25 minutes per interruption, and the questions come all day. You’re paying management wages for work that needs none of their judgment. The fixes you’ve already priced are heavy: ignore it, hire for it, or wedge a support-desk ticketing system into a restaurant.

A Tuesday, before and after
Before: every ping is a context switch
After: one short review, twice a day
The predictable mail answers itself or waits in one queue you clear twice a day. Nobody has to stand over the inbox anymore.
The email distraction
$37,000 / year

An hour a day writing replies, plus two more lost to refocus, out of the five hours that a dozen interruptions break at the measured 25-minute recovery. Three hours a day at a manager’s full employer cost, about $48 an hour with benefits, comes to $37,000 a year.

The human drift you already have
40% of productive time

Everyone asks whether AI will drift. Human distraction is drift too: task-switching can eat up to 40% of productive time, and a manager answering the corkage question mid-rush shortens, improvises, misremembers. Your doc doesn’t. The answer holds at 6pm on a Friday.

The old fix
$68,000 / year

Hire it away instead: the median U.S. administrative assistant earns $47,540 — about $68,000 once benefits are added, which government data puts at 30% of total compensation. A real salary, spent retyping the same twelve answers.

Counted conservatively, repeat questions cost about $37,000 a year of management attention, or $68,000 to hire away.An owner’s hour runs higher still.
Published figures: U.S. Bureau of Labor Statistics, May 2025 wages and March 2026 benefit costs · UC Irvine interrupted-work field study (CHI 2005) · American Psychological Association. The dollar lines are our arithmetic on those figures, using full employer cost (wages plus the 30% benefits share) on both sides of the comparison.

Your Gmail and one Google Doc. That’s everything it runs on.

Your team doesn’t learn ticket numbers or macros. Connect one inbox or five — info@, bookings@, events@ — and the day-to-day is still one simple screen: a review queue across every inbox, cleared with your coffee. Replies go out from each address as itself, and the answers live in a doc you already know how to edit.

Every inbox, one queueNothing for your team to learnSet up on one call, about an hour
Where every answer comes from.

It answers from your real, live playbook. Nothing else.

That playbook is one Google Doc you own and edit, and every claim in every reply carries a citation back to the line it came from. Change the doc and every future answer changes with it. Otto can’t wander from what you wrote, because your playbook is the only place it’s allowed to answer from.

Every claim, cited

“Yes — dogs are welcome on the terrace.1 We just ask that they stay leashed.2”

1–2  Your doc › House policies › “Dogs allowed on terrace, leashed”
A second, independent AI checks every draft against the doc before it moves.
Your mail is never used to train AI models — not by us, and not by our AI provider.
The halfway fix.

A drafted reply is still your reply.

A drafted reply will still always need you — your time, your attention, your hand on the Send button. Most tools stop at the draft: it sits in your inbox, and you open it, check it, and send it, on every email, in every inbox you run. That saves the typing, which was always the smaller line on the bill. Otto is built to finish the job: once a category of question has earned your trust, those replies send themselves.

How draft-only tools leave it
Drafts23
Re: do you have parking?
Re: table for 12 on Saturday?
Re: are dogs ok on the patio?
20 more…
Every one still needs opening, checking, and sending.
Autonomy, category by category
Hours & directionsGraduated — answering itself
62 decisions · 100% approved as written
Availability & bookingstill apprenticing · 38 of 50
98% approved as written
Dietary & allergensYour never-list
You marked it human-only. Drafts still come ready.
The apprenticeship.

It starts as an apprentice. It graduates when you say.

Otto sends nothing on its own while it learns, and every approval builds a track record you can see. A category graduates at 50 approvals with 98% sent as written, and only when you flip the switch. The first category usually starts answering on its own early, and the rest follow as they earn it.

Here’s the math.

100%
of replies drafted for you, from email one
every question shows up with a finished draft; approving takes seconds
25 min
to refocus after one interruption
the measured average from UC Irvine’s field study of interrupted work
50
clean approvals before a category graduates
after that, those replies send on their own, each with a delay so you can still cancel it
1
document you own, behind every answer
Otto learns from the doc, never from your mail, so its answers can’t wander from what you wrote

Watch your inbox learn to run itself.

Every question Otto can’t answer yet gets logged in your customer’s own words, ranked by how often it costs you a reply. Answer it once in the doc and it’s answered for good. The share of mail handling itself climbs week over week.

Answering itself
58%↑ from 41% last week
Illustrative. Your own curve starts building from day one.

Answer the closing-time question one last time.

Set up with you on a call, usually about an hour. The apprenticeship starts the same day.

If you have to worry about a solution, it’s not a solution.