Let AI Answer the Inbox. You Set the Rules.

The Inbox Nobody Wants to Automate

A big share of any AR inbox is the same handful of requests. Send me a copy of that invoice. Send our statement. How do we pay you. Did our payment come through. Mixed in are replies that leave something out, like “we’ll pay next week” with no date or amount, or “we’re disputing this” with no evidence. Someone on the AR team answers each one by hand, because that’s the safe choice when you don’t know what an automated reply would say.

That’s the leap teams won’t take on faith. Will it quote something made up, like a balance that’s slightly off? Will it email an account you’re in the middle of negotiating with? Those questions come up again and again when we talk with AR and finance leaders, and until there’s a real answer, the careful move is to keep a person on every reply.

Here’s what we built to give you that answer. By the end, you’ll know exactly what an automated reply can say, what stays a draft until your team sends it, and what gets checked again right before anything goes out on its own.

What Could Go Wrong With an Automated Reply

The fear isn’t that an automated reply will sound robotic. It’s that it will get something wrong in a way that’s hard to take back once a customer has read it. A quoted balance that’s slightly off. A promise about a dispute nobody approved. An answer sent into a thread where you’re in the middle of a sensitive negotiation. Said in an email, any of those is difficult to walk back.

The cost of not automating is real too. Copy-and-paste replies to routine requests are, by internal count, one of the highest-volume workflows in an AR inbox. Every one of them is a few minutes of a specialist’s day that could go toward the accounts that actually need judgment. And the risk isn’t only about what gets sent wrong. Before the checks described below existed, an automated reply could still go out into a thread whose case had been frozen, or to an address that had already bounced, which is the kind of mistake that turns into a customer relations problem, not just a wasted email.

Draft or Send, One Reply Type at a Time

The first control is the simplest one: you decide how much runs on its own, and you decide it separately for each kind of reply. An invoice copy, a statement, payment instructions, a payment confirmation, and each of the missing-information asks (a promise to pay with no date, a dispute with no evidence, a wrong contact) can each be set to Off, Create draft, or Auto-send.

A draft waits in the thread for your team to approve before anything goes to the customer. An auto-send reply still waits a short moment first, and anyone on your team can cancel it before it goes. Think of that wait as an undo button for the email you’d otherwise be apologizing for. Turning this on for one reply type doesn’t turn it on for all of them. You can let invoice copies go out on their own while every dispute-related reply stays a draft.

Your Template Holds the Facts

The reply itself is built from a template your team wrote, filled in with that customer’s own invoices and a pay link where you have one. If you turn on AI wording, Daylit rewrites the greeting and the acknowledgement so the reply is a bit more humanized, but a checker sits between that rewrite and the send: anything that changes one of your template’s fields, or invents a date or an invoice number that wasn’t there, gets rejected, and the reply falls back to your template as written.

The missing-information asks work the same way, with one more layer of care. If a customer’s reply already gave you the amount but not the date, Daylit asks for only the date, and the reply acknowledges what they already sent instead of asking for everything again. Nobody gets asked twice for information they already gave you.

The Last Check Before It Sends

Every automatic reply is checked again right before it actually sends, because things can change between when a reply is drafted and when it goes out. If someone has frozen the case in the meantime, the reply doesn’t go. If the address has bounced, or your team has blocked it, or the customer has opted out, the reply doesn’t go. If your company has hit its daily sending limit, it waits.

Freezing a case is how you keep automation off a thread you’re actively negotiating on, even if a reply for it was already sitting in the queue to send. And blocking an address stops automated email specifically. A person on your team writing that same address manually gets a warning, not a hard stop, because a person’s judgment comes first.

One thing worth being precise about: your team records an opt-out, or it arrives as an unsubscribe or a spam complaint. Daylit doesn’t read an opt-out out of the words in a reply.

The Part That Stays Human

While AI handles most of it, the riskier replies stay held for review. Missing-information asks set to Auto-send don’t actually send on their own until Daylit approves unattended sending for your company specifically, an internal check that sits on top of your own settings.

A few other things worth knowing going in. It doesn’t send the invoice as an attached PDF, replies are body-only, with an invoice list and a pay link. It doesn’t answer any email a customer sends, only the four routine requests and three missing-information asks described above; everything else stays with your team. And it doesn’t quietly change its own behavior over time. Every reply follows the rules you set, until you change them yourself in Settings.

How It Works in Daylit, End to End

When a customer emails in, Daylit reads it and works out what they’re asking for. If it’s a routine request, it drafts a reply from your template. If something’s missing, it asks for just that piece. Either way, the reply sits as a draft or moves toward auto-send, exactly as you set that reply type up.

If it’s headed for auto-send, it waits a short, cancellable moment, and right before it actually goes, it’s checked one more time: is the case frozen, is the address bad, has the daily limit been hit. If everything clears, the reply lands in the same email thread the customer started, not a new one.

To see it, open Settings, then Outreach Policy, then the Auto-reply tab. That’s where each reply type gets its mode and its template.

See it against your own inbox

If these are the emails sitting in your inbox right now, book a demo and we’ll show you exactly what an automated reply would say.

The Takeaway

The question that stops AR teams from automating their inbox was never “can the AI write a decent email.” It’s “what will it actually say, and can I see it before it goes out.” Now there’s an answer. The reply comes from a template your team wrote. The facts in it can’t be changed by AI wording. Every automatic send gets checked again right before it happens, and the riskier replies stay held until Daylit signs off, not just you.

If you’re weighing whether to hand off any of your inbox, start with the reply types that carry the least risk, like invoice copies, and leave the rest as drafts. Watch what would have gone out for a week before you turn anything to auto-send.

Frequently Asked Questions

Will it email my customers without us approving it?

Only for the reply types you set to Auto-send. Everything else is a draft your team sends, or it’s off. Auto-send replies wait a short moment you can cancel, and the riskier missing-information replies are held for review until Daylit has approved unattended sending for your account.

What if the AI makes up a number?

The reply is your template. If you turn on AI wording, it only rewrites the greeting and acknowledgement, and a checker rejects any change to your template’s fields or any invented date or invoice number. If it’s rejected, your template goes out as written.

What if we’re in the middle of a negotiation with that customer?

Freeze outreach on the case. Automatic replies on that thread stop, even one that was already drafted and waiting to send.

Can I see what it would send before I turn it on?

Yes. Each reply type shows its template before you turn it on, and the missing-information replies show what triggers them, the wait, and the exact email they’d send.

Does it attach the invoice?

No. It sends the invoice list and a pay link in the body of the email, not an attached PDF.

Does it answer every email a customer sends?

No. It covers routine requests like invoice copies, statements, payment instructions, and payment confirmations, plus asking for what’s missing on a promise to pay, a dispute, or a wrong contact. Anything else stays with your team.

Your AR Inbox Isn't Slow. It's a Trust Problem.