925 cold emails from Microsoft 365, zero human replies

The numbers

We run outbound for our clients and for ourselves, from 22 sending domains. Five of them run on Microsoft 365, seventeen on Google Workspace. Same sequences, same prospect lists, same sending windows. Only the mailbox provider changes.

Method: 3 Aug to 9 Sep 2026. Each prospect counted once, in the week of their first email. Replies counted as distinct people, not messages. Our own test inboxes excluded. At the Google rate we would have expected about 8 human replies from the Microsoft side. We got zero.

What we ruled out

Our first reflex was the usual checklist. Every item came back clean, checked in DNS on 9 September.

So fixing the config would not have fixed anything. There was nothing broken to fix.

Your deliverability scores are measuring something else

Two tools told us everything was fine. Neither was measuring where the email actually lands.

The warm-up score is a closed loop between inboxes of the same tool's users. It proves your inbox can talk to other warm-up inboxes. It says nothing about a real prospect's Outlook.

mail-tester gave 8.3/10 to an email that Outlook put straight into junk. We tested it on 8 September: two identical emails, same minute, same recipient. One from a clean domain landed in the inbox. One from a domain listed on SURBL landed in junk. Both scored the same.

If you want to know where your email lands, send it to a real inbox at the provider you target and look. It takes five minutes and beats every score.

Théo Pascard, co-founder, Enjyn

The free placement test nobody uses

An out-of-office reply only fires when your email reaches the inbox. It does not fire from the junk folder. So the auto-reply rate measures placement, and nothing else: it does not care about your subject line, your copy or your targeting.

It also arrives in hours, not weeks, and it is about twice as frequent as human replies. On our data it separated the two providers from the first week, while human replies were still in the noise: 1.46% on Google, 0.22% on Microsoft (p < 0.001).

Limit: an auto-reply still fires from Gmail's Promotions tab. It tells inbox from junk, not Primary from Promotions.

The question we couldn't answer in September

Here is the honest part. We route email so that Microsoft inboxes write to Microsoft recipients and Google to Google. That is standard practice, and it means 895 of our 925 Microsoft sends went to Outlook recipients. Only 2 went from Google to Outlook.

So two stories fit the data equally well. Either our Microsoft 365 inboxes are the problem. Or Outlook filters cold email hard, whoever sends it. From this data alone, we can't tell them apart.

The random split we started on 9 September never reached a usable size: on day two Microsoft restricted two of the inboxes on that side, and it never restarted. So in October we changed the method and asked Microsoft directly.

What we found in October

Every email Microsoft receives carries a spam score in its headers, the SCL: 1 means inbox, 5 means junk. From 13 to 16 October we sent the same email from six sending setups to the same eight Microsoft 365 business inboxes, on three separate days, and read the score on every message. Three sending inboxes per setup, one dedicated domain each.

Method: eight Microsoft 365 business seed inboxes per test (MailReach), one test per setup on 13, 14 and 16 October, the same six-line email every time. Every inbox placement scored SCL 1, every junk placement SCL 5. The one miss in the 23/24 row is an email that never arrived, not one in junk. A setup passes if at least 5 of the 8 land in the inbox on at least 2 of the 3 days. Four setups passed, two failed, and the gaps were large.

So, back to September: Outlook does not junk cold email whoever sends it, since fresh dedicated inboxes reach it. Our five summer Microsoft domains had been sending 30 emails a day for weeks, and most likely wore down the same way. We can't prove that last part after the fact.

Limits: eight seed inboxes per test, not twenty. One domain per setup, so a pass says "this domain passes", not "every domain from this provider passes". One placement tool. Seed inboxes are not real companies with their own filtering rules.

What we still don't know

Whether the winners hold once they carry real volume. Every setup that passed was measured at rest, and the only one that had been sending cold email failed within a week.

So we are restarting Microsoft in steps: a few inboxes first, re-measured every Monday, and an inbox comes out the day it slips. We will add the result here in November.

What we changed