The short version.
We do not publish an open rate, a reply rate or a meeting rate for SYNC. We record opens, replies and meetings, and on 29 September 2026 we checked what those records are worth. They are not good enough to divide by anything.
For most messages with a recorded open, the first open arrives within a minute of sending. That suggests that software fetched the message and a person did not read it. Replies were incomplete until the middle of September. One meeting is recorded in the product, and a meeting agreed by phone or in a reply is recorded nowhere.
This piece lists what is recorded, why it cannot be trusted as a rate, what we changed, and what would have to be true before a rate could be trusted.
What an open is, and what we recorded.
An open is recorded when something downloads a small invisible image in the message. The record says that the image was downloaded. It does not say who or what downloaded it.
| Week from | Recorded | Counted | Excluded as our own views |
|---|---|---|---|
| 31 August | 41 | 33 | 8 |
| 7 September | 100 | 62 | 38 |
| 14 September | 96 | 96 | 0 |
| 21 September | 114 | 114 | 0 |
| 28 September (part) | 119 | 119 | 0 |
| Total | 470 | 424 | 46 |
When the first counted open came, for the 203 messages that have one:
For 145 of 203 messages (71%), the first open came within a minute of sending, and for 22 of those within ten seconds.[1] A person rarely opens a message that fast. Security scanners fetch a message on arrival, and the sender's own sent folder can fetch the image too. The timing suggests automatic fetching. We did not prove the cause.
The rest of the record does not settle it. Of the 424 counted opens, 379 came from what looks like an ordinary browser, and a scanner presents itself the same way. Another 34 came through a mail provider's image relay, which hides who asked, and 11 from a mail program.[1]
The filter for our own views is the second problem. It excluded 46 opens, all from the same source, and none since the week of 14 September.[1] The list behind the filter has not changed since 1 September. Either senders stopped looking at their sent mail, or the filter stopped matching. The data cannot tell which.
Two counts of the same thing also disagree. The counter kept on each message adds up to 518 opens, and the log of single opens holds 470. The difference comes from before 1 September, when opens were counted without a network address and without any filter.
What we recorded about replies and meetings.
Replies before 16 September are incomplete. Until then the product only learned of a reply when the mail provider reported one on a message sent through SYNC. When a seller wrote from their own mail program and the prospect answered, the product did not see it. Of the 26 incoming messages from that period, 3 were recovered later and 6 had to be relabelled by hand.[1]
Automatic replies are mixed in with the human ones: 9 of the 36 incoming messages carry that mark. A count of replies that includes out-of-office notices is not a count of interest.
One meeting is recorded. A meeting agreed by phone or in a reply is recorded nowhere in the product, so the record cannot show how many took place. We leave out the number of messages sent and the number of conversations on purpose, because a reader could divide by them and get the rates this piece says cannot be trusted.
Published benchmarks deserve the same caution.
The reply rates that circulate in sales come from companies that sell sending tools. One vendor reports an average reply rate of 3.43% across its own platform, and that 42% of replies come from follow-ups.[2] A second vendor publishes the same two figures and names the first as its source.[3] The two pages report one measurement.
A third vendor has published several studies of cold email. One rests on more than 30,000 emails, another on more than 28 million.[5][4] On the pages we read from the first and the third vendor, we found no description of how automatic replies were removed, or of how a reply that arrived outside the tool was counted. Those are the two problems we found in our own record.
We do not say these figures are wrong. A rate is only as good as its method, and these pages do not say how a reply was counted.
What we changed in the product.
- 01Since 1 September every recorded open keeps the network address it came from, and opens from addresses on our exclusion list are not counted. Before that date there was no address and no filter.
- 02Since 16 September SYNC reads the conversation in the seller's own mailbox. A reply is found there even when the seller wrote from their own mail program.
Three things are still open. The replies in the product have not been checked against the mailboxes over a full period. We do not know why the filter for our own views has excluded nothing since the week of 14 September. Meetings agreed outside the product are recorded nowhere. Until these are closed, a rate from this record cannot be trusted.
We make one promise. Once the record has matched the mailbox for a full month, we will publish our reply rate, also when the figure is poor.
What speaks against this.
A company that publishes no rates can also be hiding bad ones. Nothing in this piece rules that out. The reader has our word and a description of faulty records, and both are convenient for a company whose numbers are poor.
What would settle it is a rate with its method: the count of messages, the count of human replies checked against the mailboxes, automatic replies removed, over a period fixed in advance. We cannot give that today.
Our data is small. It covers 470 opens over about a month and 36 incoming messages over three and a half months, and most of the activity comes from very few workspaces. The reading that fast opens are automatic is an inference from timing. Some people do open a message at once.
The vendor figures rest on far more mail than we have sent. Our doubts about their method do not show that their numbers are wrong.[2][4]
We counted every recorded open since the log began in the week of 31 August 2026 (470, in 4 workspaces), and every incoming message in the conversations the product holds (36, from 15 June to 29 September 2026; the conversations are in 5 workspaces). The time to the first open is the time between sending and the first counted open, for the 203 messages that have one. The two middle bars are derived: they are the differences between the counts we took at one minute, five minutes and one hour. The share of 71% is computed from 145 and 203.
The split at 16 September uses the time a message entered our record, which can be later than the time it arrived. The periods before and after differ in length and in volume, so the two cannot be compared. We left out the number of messages sent and any split by workspace. No open, reply or meeting rate was computed.
The vendor pages were read on 29 September 2026. Figures like ours change daily and are valid for the day they were taken.
- [1]iSyncSO. Measurement of recorded opens, replies and meetings, 29 September 2026.our own measurementsee the method note
- [2]Instantly. Cold Email Benchmark Report 2026.vendor researchfigures from the vendor's own platform
- [3]Woodpecker. Cold Email Statistics.vendor researchrepeats the Instantly figures; not an independent source
- [4]Gong Labs. Does cold email even work any more? Here's what the data says.vendor research
- [5]Gong Labs. Increase Your Sales Email Reply Rate: 4 Data-Backed Methods.vendor research