AI email review for Microsoft 365

Know when an email your team sends could become fraud, a data leak or a lawsuit.

Aftercomm’s AI judges each email against the policy you set, by what it means, not which keywords it contains. You hear within minutes, with what’s wrong and why, while there’s still time to act.

What Aftercomm catches

Hacked billing mailbox → customer

High

“We’ve changed banks. Please send this month’s $48,000 payment to our new account below, today if possible.”

Keyword filter

No match · never flagged

Aftercomm

Flagged · Payment fraud

Alert sent in minutes

Why: New bank details and a push to pay a large sum today, around the normal process: the pattern of business-email-compromise fraud.

  • Your policy, not keyword lists
  • Real verdicts on your own sent mail the day you connect
  • Aftercomm never keeps a copy of your email

7-day free trial · No credit card · 15-minute setup · Any Microsoft 365 plan

Microsoft 365 email — live todayGoogle Workspace, Slack & Teams next

The math nobody does

20,000 outbound emails a month.Nobody is reading them.

1025501002505001k

20,000

outbound emails every month

167 hrs

to skim each one for 30 seconds

1

full-time person doing nothing else — every month, forever

Assumes a conservative 20 sent emails per person per business day, 20 business days a month, and 30 seconds to skim each one. Your HR, legal and finance teams already have full-time jobs.

And a keyword filter can’t read them either.

Rules match words. The message that turns into a complaint, a lawsuit, or a headline usually doesn’t contain the word — the risk is in what it means. Aftercomm’s AI reviews each message the way your CFO, legal counsel, compliance or HR would, against your policy, and tells you why it matters.

The compliance problem isn’t the risky message. It’s finding out too late.

Most incidents aren’t malicious. They’re one tired employee, one bad message, and weeks or months before anyone with authority to respond even knows it went out. By then it’s a legal matter, not a coaching conversation.

What Aftercomm flags on day one, by who needs to know

Finance · CFO, controller

Money paid to the wrong account

A hacked company mailbox sends customers new bank details. Someone updates a supplier’s account on an emailed request. Non-public forecasts go to someone with no reason to have them.

Payment fraudFinancial disclosure

Compliance · CCO

Business moved off the record

“Text me on my personal phone so there’s no record.” A guaranteed return promised in writing. A hint to trade on news that isn’t public yet.

Evading oversightOff-the-record commitments

HR · people team

A personnel decision, in writing

A candidate passed over because of her family plans. Action against someone who complained. A colleague’s medical details shared around. A hostile or harassing message.

Discrimination and retaliationEmployee privacyConduct

Legal · general counsel

An admission or a side deal

A known safety defect admitted to a customer. A promise kept out of the contract. Prices coordinated with a competitor. Counsel’s advice forwarded outside.

Legal exposureOff-the-record commitments

IT & security

Data and passwords walking out

A departing employee mails the client list to a personal account. A password or API key pasted into an email. Customer records sent outside in the body of a message.

Data leaving the companyCredentialsPersonal data

Your own rules

What only your company knows

A deal codename, a discount limit, a client you can’t name. Add a line in your own words, test it on any email, and it applies from the next review.

Written by your team

All eleven areas are in the policy every workspace starts with, written to stay quiet on routine mail: quotes, contracts, disclaimers, scheduling, and an employee’s report to a regulator are left alone. How the policy works ↓

Your policy, not a keyword list

Write your own rules. The AI understands them.

A keyword tool lets you pick words, then flags every email that contains them. Aftercomm lets you write the policy itself and judges each email on what it means. Change it any time, or add rules only your company knows.

01 · What gets missed

Risky emails rarely use risky words.

Hacked billing mailbox → customer

We’ve changed banks. Please send this month’s $48,000 payment to our new account below, today if possible.

Your policyFlag changed bank details or pressure to pay a new account: the pattern of invoice fraud.

Keyword rule

No match

No alert

Aftercomm

Flagged

Payment fraud

02 · False alarms

Safe emails use risky words all the time.

Account manager → client

Does Thursday at 2 still work for the kickoff?CONFIDENTIAL: This email and any attachments are confidential…

Your policyDon’t flag routine email: updates, questions, scheduling.

Keyword rule

Flagged

Matched “confidential”

Aftercomm

Not flagged

Routine scheduling

03 · What only you know

Your sensitive topics aren’t on anyone’s list.

Marketing → ad agency

Can we push the spring campaign kickoff to April? Leadership wants to hold off until Harbor closes in March.

Added by your teamFlag any mention of Project Harbor to anyone outside the deal team.

Keyword rule

No match

No alert

Aftercomm

Flagged

Project Harbor

Teams add lines like “Flag any discount over 15% offered to a customer” or switch one off with “Don’t flag casual swearing.” Test a change on any email before you save it; every version is kept. Compared with Microsoft Purview →

How AI email compliance works

From an outbound message to a documented decision.

Setup takes about fifteen minutes in four steps, with no mail server changes, and ends with a first review of mail your team has already sent, so you see real verdicts on your own email the same day. After that, a high-severity flag reaches your reviewers within fifteen minutes, as fast as five; everything else arrives in a daily digest, or as it happens if you prefer.

Setup, once

01

Connect Microsoft 365

A Microsoft 365 admin approves two read-only permissions. No change to how your email is delivered, and nothing to install.

02

Choose the AI account

Aftercomm provides the AI, so there’s nothing to set up, and your email is never used to train AI models. Usage is billed at list price, with no markup. Or run it under your company’s own AI contract.

03

Choose reviewers and the policy

Who gets alerts: compliance, HR, legal. The policy is ready on day one; change any line at any time, and every change is versioned.

04

Pick the mailboxes

Start with a pilot group, or monitor every mailbox, kept in step with your directory. You pay only for mailboxes monitored.

Then, continuously

Aftercomm reviews the message. It doesn’t keep it.

So who actually sees your company’s email? Microsoft, where it already lives, and the AI model, for the few seconds of the review. Aftercomm’s software holds each message in memory only while it is reviewed, then discards it: Aftercomm never stores message content, and no one at Aftercomm reads it. The AI provider never trains on it, and Aftercomm has no access to any copy the provider keeps for its own safety checks (how the provider handles data).

  1. 1Your Microsoft 365

    Your team sends an email

    Mail flows exactly as it does today. Aftercomm never sits inline, so nothing is delayed, blocked, or altered.

    Aftercomm keeps Nothing — it stays in your Microsoft 365

  2. 2Aftercomm

    Aftercomm checks your chosen mailboxes

    Sent mail only, with read-only access granted by Microsoft. Checked every 15 minutes by default, as often as every 5.

    Aftercomm keeps Nothing yet

  3. 3AI review

    The judgement call

    The AI weighs the message against your policy, for that one call. It never trains on it.

    Aftercomm keeps Nothing by Aftercomm. The AI provider keeps only what its safety checks need, never shared with Aftercomm

  4. 4Aftercomm

    A verdict comes back

    Flagged or clear, with severity, categories, and reasoning. The body is dropped the moment it arrives.

    Aftercomm keeps The verdict, plus sender, recipients, subject and timestamp

  5. 5Aftercomm

    Reviewers are alerted, and act

    Your reviewers get the verdict, never the message, and triage in the portal. Every view is logged.

    Aftercomm keeps The audit trail: who reviewed it, what they decided, and when

Runs inside accounts you ownAftercomm

Step 3 runs under Aftercomm’s business agreement with its AI provider, which bars training on your email, or under your company’s own agreement if you choose. Full data-handling detail →

Admin console

Compliance sees the signal. Not the noise.

Reviewers get a clean feed of what tripped the rules and why: severity, category, and the reason in a sentence. Nothing to install. Nothing to train.

  • ✓Per-mailbox onboarding — start with a pilot group, expand later.
  • ✓Mark each flag acknowledged, actioned or a false positive, with notes, so the record shows what you did.
  • ✓Severity-graded verdicts (low / medium / high) so triage is instant.
  • ✓Alerts carry the verdict, never the message — no copy of your email in anyone’s inbox.

Aftercomm · Dashboard

Reviewed

1,204

Flagged

12

Flag rate

1.0%

Recent flagged

  • Re: Q3 headcount planning

    HIGH

    cfo@acme.com → external@...

    Discloses non-public financial figures to a wide external distribution.

  • Re: promotion slate

    MEDIUM

    director@acme.com → team@acme.com

    Cites a team member’s age as a reason against promotion.

  • files you asked for

    MEDIUM

    engineering@acme.com → personal webmail

    Customer export sent to a personal address with no business reason given.

Security

Designed to pass an enterprise security review.

The right stance for a tool that reads outbound communications is to hold as little of them as possible. Aftercomm holds none — no message bodies, no attachments, no thread history. What it keeps is the verdict, and what it asks for is two read-only permissions.

Two read-only permissions

Mail.Read and User.ReadBasic.All. No send capability, no write access, nothing else. Revocable in one click from your Microsoft admin center.

Alerts carry no message content

Reviewers are told what was flagged and why — severity, categories, sender, and the AI’s reasoning. Never the body, subject line, or recipient list. There is no copy of your email in anyone’s inbox.

Turn it off in one click

A switch in Settings stops all access immediately — reading mail, alerts, and opening messages — with a direct link to revoke Aftercomm in your own Microsoft admin center. No support ticket, no notice period, no waiting on anyone.

Opening a message is logged

A message is fetched from Microsoft 365 only when a reviewer opens it, and only in a session confirmed with a recent authenticator code. It is never stored. Every view is logged: who, when, which message, allowed or refused. You can switch viewing off entirely.

No AI password to steal

Aftercomm’s servers connect to the AI provider with short-lived credentials that expire on their own, not a stored key. If you connect your own AI account, its key is encrypted, and the means to unlock it is kept outside the database. Each customer’s data is walled off from every other’s.

Sensitive changes need a recent code

Adding an admin, redirecting alerts, editing the policy or pausing a mailbox needs a recent authenticator code, good for a working session of up to an hour, so a session cookie taken later is refused. Five wrong codes sign the account out everywhere.

Pricing

$5 per mailbox. The AI at list price, nothing else.

Aftercomm’s fee covers the platform, the admin console, and the compliance workflow. The AI review is passed through at list price with no markup, as its own line on the same invoice: $0.20 to $0.50 per 1,000 typical emails. Or connect your company’s own Anthropic account and pay for usage directly.

Aftercomm

Self-serve

$5/mailbox/month

Any number of mailboxes · start today · no minimum

Each mailbox: $5 a month, plus about 8 to 20 cents of AI for a typical 400 emails

  • 7-day free trial, no credit card
  • AI review of sent email from every monitored mailbox
  • Prorated billing — only for mailboxes actually monitored
  • Alerts to your compliance, HR, and legal reviewers
  • Configurable severity threshold
  • AI usage at list price, no markup ($0.20 to $0.50 per 1,000 emails)
  • Email support
Start free trial

Enterprise

Custom

For procurement, security review or hosting requirements

  • Everything above
  • Custom pricing and terms
  • Self-hosted in your own VPC or on-prem
  • Priority support with SLA
  • Custom DPA + security review
  • Regional deployment (US today, EU planned)
Contact sales

Prorated to the day — a mailbox monitored for 15 days in a 30-day month bills at half a mailbox-month. AI usage is what your reviews actually cost to run, itemized by mailbox in Billing. All prices in USD.

Want to see what a flagged-message alert looks like? Preview a sample notification →

Questions

Answered in full, on the page. Nothing here is behind a demo call.

Setup, platforms, and permissions

Can we decide what it flags?
Yes, entirely. Your policy is written in your own words, several pages long if you need it. It can cover conduct, discrimination, payment fraud, side deals, financial disclosure, legal exposure, credentials, and data leaving the company, plus rules only your company knows (a deal codename, a discount limit, a client you can’t name). Switch off what’s normal for you, and change any line at any time. Test any email against a change before you save it. Every version is kept, and each flag records the version that judged it.
What does it flag out of the box?
Every workspace starts with a written policy covering eleven risk areas: payment fraud and changed bank details, side deals and guaranteed returns, non-public financial information, legal exposure (admissions, bribes, collusion), evading oversight, credentials, personal data, data leaving the company, discrimination and retaliation, employee privacy, and conduct. It is as specific about what to leave alone: routine mail, quotes and contracts, disclaimers and footers, and an employee’s report to a regulator or their own lawyer.
Does it work in other languages?
Yes. Emails and the policy can be in other languages, and they can be mixed: a rule written in English applies to an email in Japanese, and the other way round. Aftercomm has been tested in Spanish, French, German and Japanese. The explanation of each flag is written in English whatever language the email is in, and the console and alert emails are in English for now; your browser and Outlook can translate them in one click.
Does this prevent people from sending risky messages?
No — by design. Aftercomm is post-send review: it checks monitored mailboxes on an interval you set, every 5 minutes at the fastest and 15 by default, so a high-severity flag reaches your reviewers within that window. Lower-severity flags are batched into a daily digest by default, or sent as they happen if you prefer. Sitting inline as a gateway would add delivery risk, and a fast heads-up is what compliance teams actually need in order to remediate.
Does it read incoming mail, or attachments?
Neither. Aftercomm reviews what your team sends, from Sent Items: the message text and the names of any attachments, never what is inside them. So a customer list sent as a file is judged by its name and the email around it. Incoming phishing and malware stay your email security gateway’s job; Aftercomm works alongside it, not in place of it.
Which platforms do you support?
Microsoft 365 (Exchange Online) today. Google Workspace is next, followed by Slack, Microsoft Teams, and other enterprise messaging systems. The review pipeline is platform-agnostic — each integration is a read-only connector, subject to the same rule that message content is never stored.
What do we need to give you to get started?
A Microsoft 365 Administrator to approve two read-only Microsoft permissions. That’s it: Aftercomm provides the AI, so there is no AI vendor to sign up with, and your email is never used to train AI models (enterprises can run it under their own AI contract instead). Not required: MX or mail-routing changes, a gateway inline with delivery, endpoint software, firewall changes, mailbox delegation, or a copy of your mail archive.
See the full requirements list →
Exactly which Microsoft permissions do you ask for?
Two, both read-only.
  • Mail.ReadReads message content in the mailboxes you select, so your policy can be evaluated against it. It cannot send, modify, or delete anything.
  • User.ReadBasic.AllReturns display name and email address only, so you can pick mailboxes from a list instead of typing addresses by hand.
Aftercomm can’t send, change, or delete anything in your Microsoft 365, and you can revoke everything in one click from your Microsoft admin center.
See the full permissions breakdown →
Do you offer self-hosted deployment?
Yes, on the Enterprise tier. The full application ships as a Docker Compose bundle you run inside your own infrastructure — VPC or on-prem — so message processing and the database stay within your compliance boundary. It is not air-gapped, and Aftercomm won’t pretend otherwise: the service must reach Microsoft Graph to read mail and the AI model’s API to get a verdict.

The AI, and what it costs

Which AI model does Aftercomm use?
Aftercomm’s review engine is the part you configure and audit: your policy, the instructions that apply it the same way to every message, the checks every verdict must pass before it is recorded, and the records it leaves. The language model underneath is Claude, from Anthropic, run on Aftercomm’s account under a business agreement: Anthropic never trains on the mail, and Aftercomm has no access to any copy Anthropic keeps for its own safety checks. Your email is read by Microsoft, where it already lives, and by the model for the moment of the review. Your own reviewers can open the messages it flags. No one at Aftercomm reads your email. If your security team wants the model to run under your own contract, you can connect your company’s Anthropic account instead; your agreement then covers every review, and Anthropic bills you directly.
How Anthropic handles data →
What does the AI cost?
What each review costs to run, at the model’s list price, with no markup, as a separate line on your monthly invoice: $0.20 to $0.50 per 1,000 emails of typical business mail, or about 8 to 20 cents a month for a mailbox sending 400 emails, however many mailboxes you monitor. Long messages and long threads cost more, short replies less, and Billing shows the cost for every mailbox, with the rates applied. The free trial includes it. On your own AI account, the provider bills that usage to you directly and Aftercomm charges nothing for it.
Can the review run through Bedrock or Vertex AI instead?
It’s on the roadmap rather than shipped. Today Aftercomm calls the model provider directly, on Aftercomm’s account or on your own. Running the same model through Amazon Bedrock or Google Vertex AI, so the call executes inside an AWS or Google contract you already hold, is planned. If it’s a requirement for you, say so: priority follows who asks.

Data, privacy, and control

Do you store the messages you review?
No. The body is processed in memory and dropped after the verdict returns — which also means it is in none of Aftercomm’s backups, because a backup can only contain what was written down. Attachments are never downloaded, and their filenames are never kept. Aftercomm retains only what an audit trail requires: sender, recipients, subject, timestamp, attachment count, severity, category, and reasoning. The AI provider never trains on it, and Aftercomm has no access to any copy the provider keeps for its own safety checks.
How the AI provider handles data →
How long do you keep what you do store, and can we erase it?
Message content has no retention period, because it is never retained. Flagged messages and their verdicts are kept for the life of your workspace, as your compliance record. For mail your policy clears, the subject, recipients, Outlook link and the AI’s note are erased after 30 days, leaving only a receipt that it was reviewed. A workspace owner can erase all of it in one action from Settings, confirming by typing the workspace name; the response hands back a count of exactly what was destroyed.
Retention and deletion in detail →
How do reviewers get notified about flagged messages?
Your reviewers get an email from Aftercomm saying what was flagged, how serious it is, and why — severity, risk categories, the sending employee, and the AI’s explanation. It deliberately does not contain the message body, subject line, or recipient list. A reviewer who needs to read the message opens it in Aftercomm, where it is fetched from Microsoft 365 at that moment, shown, and dropped, with every view logged.
See a sample alert →
Can we limit which mailboxes are reviewed?
Yes. You pick exactly which mailboxes are reviewed from a searchable list, and Aftercomm never reads anything else. Microsoft grants mail permission for the whole organization — there is no way to narrow them at the consent screen — so by default that limit is enforced in Aftercomm’s software. Organizations that want Microsoft to enforce it instead can apply Exchange RBAC for Applications, and Aftercomm will then prove the restriction works by trying to read a mailbox you didn’t select and showing you that Microsoft refused.
How the provider-level lock works →
How hard is it to turn off?
One click. Settings has a switch that stops all Microsoft 365 access immediately — Aftercomm stops reading mail, no further alerts are produced, and no one can open message content. No support ticket, no notice period. Settings then links you to the Aftercomm entry in your own Microsoft admin center to revoke the permission grant itself; doing it yourself means the revocation is enforced by Microsoft rather than by Aftercomm’s word. Your review history stays intact and exportable.
Do we have to tell employees their email is reviewed?
Often, yes. Some US states, New York, Connecticut and Delaware among them, require written notice before an employer monitors email, and in the EU and UK monitoring needs a lawful basis and a clear privacy notice. Most companies cover it in an acceptable-use or electronic-communications policy employees already acknowledge. Aftercomm reviews only the company mailboxes you choose. This isn’t legal advice: confirm your notice with counsel before you start.
Does the system ever contact the sender of a flagged message?
No. False positives happen. Your compliance, HR, or legal team decides what response — if any — is appropriate. Aftercomm never messages the sender.
What is your compliance posture?
SOC 2 Type II is in progress with a Big-Four auditor, and ISO 27001 is on the roadmap. Aftercomm provides a GDPR-ready DPA on request and completes standard security questionnaires (SIG, CAIQ). It doesn’t yet sign HIPAA business associate agreements, so get in touch before monitoring mailboxes that send patient information. Enterprise customers can run their own penetration test.

Know about the risky email before it becomes a legal matter.

Connect Microsoft 365 in about 15 minutes, and see what Aftercomm finds in the mail your team has already sent, the same day.

7-day free trial · No credit card required