The Stalled Outbound Search Read: Re-Target, Re-Pitch, or Reopen the Brief

This is the read I run when an outbound search has been live for two or three weeks and the replies haven't come. It's for in-house recruiters, sourcers, and founders doing their own hiring. You give it the live counts from one search, and it gives back one verdict: re-target the list, re-pitch the message, or reopen the brief with the hiring manager. It also names the one change to make and the count that will show whether the change worked.

What a stalled search read decides

A stalled search read works out which single part of an outbound search is broken, so the next batch changes one thing instead of everything. It runs three gates first (maturity, reachability, sample size), then scores five failure dimensions, then maps the scores to one verdict.

The scope is narrow on purpose. This read covers only the sourcer's own outbound funnel in the middle of a search: contacted, delivered, replied, interested, screened, submitted. It doesn't cover offers, approvals, or interview loops. It isn't a fillability check you run before the search starts, and it isn't a message split test.

Three terms I use throughout:

• Sequence: the full series of touches (the first message plus follow-ups) sent to one contact.
• Reply rate: contacts with any reply, divided by contacts delivered. Each contact counts once, across the whole sequence.
• Interested rate: contacts whose reply expresses interest or moves straight to scheduling, divided by contacts delivered.

The three verdicts:

• Re-target the list. You're contacting the wrong people. The fix is a new list, not new words.
• Re-pitch the message. The right people are seeing a message that doesn't land. The fix is the copy, the cadence, or the sender.
• Reopen the brief. The right people read the message and turn the job down, or they reach screen and fail a bar the brief never stated. Only the hiring manager can change that.

Two outcomes aren't verdicts. "Wait" means the sequences haven't matured or the sample is too small. "Fix the pipe" means mail isn't arriving. Both mean the same thing for now: change nothing about the list, the message, or the brief.

Why only one change? If you swap the list and rewrite the message in the same batch, the next read can't tell you which change moved the number.

Which outreach benchmarks can I score against?

You can only score against a benchmark that counts the same unit you count, for the same channel and role family. Published recruiting reply rates come in two incompatible units, per sequence and per message, and comparing across them invents failures that aren't there.

What sits behind those rows:

• The earlier CRM study covered roughly 8 million recruiting-email sequences, with a 78.3% open rate. The later study doesn't state its sample size.
• The ATS study covered more than 500,000 sequences over two years. Business roles came in at 18.7% and technical roles at 19.4%. Startups saw 22% replies, against 19% across all company sizes.
• The sourcing-platform study counted more than 4 million messages from more than 1,500 recruiting organisations, per message. By role family: product managers 9.94%, marketers 8.57%, engineers 4.64%.
• Interested as a share of replies, which I derived from the rows above: 37% in the earlier CRM study, 54% in the later one, and 37.7% in the ATS study.

No like-for-like series across years is publicly established, because each source defines a "reply" differently. Pick one benchmark that matches your unit and stick with it for the whole search.

Two consequences matter for scoring. First, email and LinkedIn differ by a factor of about 3.4 per message, so a blended average hides which channel is failing. Second, engineers reply per message at less than half the product manager rate, so grade against the role family, never the overall average.

What each follow-up adds

Follow-ups add replies faster than they add interest. In the earlier CRM study, cumulative reply went from 8.3% after touch 1 to 15.8% after touch 2 and 21.3% through touch 5. Over the same touches, interested went from 3.9% to 6.2% to 7.9%.

Touches 2 to 5 added 13.0 reply points but only 4.0 interested points, so only 31% of the replies they added were interested. At touch 1 the interested share was 47%. Later replies lean towards closing the loop with a polite no.

Sources disagree on where the curve flattens:

• The ATS figures, as reported secondhand, show about 7% after one email, 15% after two, 23% after three, then a plateau.
• In the sourcing-platform data, the first three touches captured 93.2% of replies, and a fourth brought the total to 97.7%.
• A second sourcing tool's study of more than 2.7 million emails found that three to four touches captured 90 to 95% of replies.
• The later CRM study says engagement starts to flatten after touch 5.

There's a useful check here: in the sourcing-platform data, message one earned about 43.8% of all replies. If your first message earns far more than that, your follow-ups are the weak part. Engineers lean harder on follow-ups. One small study of 172 software-engineer emails found only 37.9% of engineer replies came on the first email, against 45.9% for all roles. Treat that as directional only.

How many contacts before a rate means anything?

More than most stalled searches have sent. Telling a real halving of the benchmark apart from noise takes about 101 delivered contacts against a 21.3% sequence reply rate. Against a 7.9% interested rate it takes about 307, and against the 4.64% engineer reply rate about 540.

No recruiting source publishes a minimum sample. The numbers below are my own binomial calculations from the published rates. Treat them as derived, not as an industry standard.

How to read the two columns:

• Zero-reply n comes from the rule of three: with zero events in n trials, the 95% upper bound on the true rate is about 3 divided by n. If you have zero replies from fewer contacts than this, silence still fits a healthy search. Zero replies from 15 contacts tells you nothing.
• n to detect a halving is the delivered count at which a rate half the benchmark becomes distinguishable at 95% confidence and 80% power.

A worked example shows why this matters. Five replies from 50 contacts gives a 95% Wilson interval of 4.3% to 21.4%, which still includes the 21.3% benchmark. A 10% rate on 50 contacts proves nothing. Ten replies from 100 contacts gives 5.5% to 17.4%, which excludes it. The same 10% rate on 100 contacts is a real shortfall.

The PM and engineer rows use per-message rates, which run lower than per-sequence rates for the same people. Higher base rates need smaller samples, so if you measure per sequence, those rows overstate what you need. Treat them as a conservative ceiling.

Most searches never reach the interested-rate gate. When yours doesn't, reach a provisional verdict from the reply rate plus the reply intent tags, and label it provisional in your notes. On volume: at the engineer rate, ten replies take about 216 contacts, and at the PM rate about 101.

The five failure dimensions and how to score each

Score list fit, reachability, message, offer, and screen calibration as 0 (healthy), 1 (watch), or 2 (broken), using the evidence each one leaves in the counts. Any 2 is a candidate verdict, and the precedence order in the verdict section decides which one wins.

List fit

List fit tells you whether the people you contacted are the people the brief describes. To score it, audit 20 randomly drawn delivered contacts against the brief's must-haves, not the title. Score 0 for 18 or more matches, 1 for 15 to 17, and 2 for under 15. These cut-offs are my working standard rather than a published one. I anchored them on what title filters alone produce.

In Refolk's index of professional profiles, the current title exactly matched the filter for 22 of 25 US Senior Software Engineers, 19 of 25 UK, and 23 of 25 US Senior Product Managers. The profiles that didn't match were CTOs, founders, an engineering manager, and a researcher. People like that are unlikely to answer an IC pitch, so they inflate the contacted count and drag the rate down.

Small pools hide list-fit failures. The US pool of Senior Software Engineers in Refolk's index is 6.9 times the UK pool, and the UK sample also had the lowest title match. In a thin market, sourcers loosen filters to keep volume up, and fit drops. That shows up as a weak reply rate, not as a smaller list. A low-scoring audit means the list needs rebuilding against the role itself. Refolk takes the role as a plain-English description, so you can write the exclusions straight into the request instead of cleaning out a title-matched list by hand.

Reachability

Reachability tells you whether your messages arrive in inboxes at all. Score 2 if SPF, DKIM, or DMARC fails, or if the spam rate in Google Postmaster Tools reaches 0.3%. Score 1 if the spam rate sits between 0.1% and 0.3%. Score 0 if authentication passes and the spam rate is below 0.1%.

Google and Yahoo require authentication from all senders, not just bulk senders. The stricter bulk-sender rules apply at 5,000 or more messages to Gmail addresses in a day, and one-click unsubscribe is among them. Google has moved enforcement from temporary delays to permanent rejections. These rules change, so re-check the Gmail sender guidelines each time you run this read rather than trusting a number you remember.

On hard bounces, mailbox providers don't publish a single threshold. The 2% line you'll see quoted is a practitioner convention I couldn't verify against Google, so read bounces alongside the Postmaster reputation data, never on their own. As for timing, DNS fixes propagate within 24 to 48 hours, but reputation damage takes two to four weeks of clean sending to repair.

Message

The message score tells you whether the right people, reached cleanly, find the pitch worth answering. Only score it when list fit is 0 or 1 and the sample has cleared the gate. Score 0 if the interested rate is at or above the benchmark in the same unit, 1 if it's below the benchmark but above half of it, and 2 if it's below half at a cleared n.

As a secondary signal, look at the share of replies that came on touch 1. If it's far above about 43.8%, the follow-ups are underperforming even if the opener is fine. Some practitioners read the open-to-reply rate first. Open tracking is unavailable in some places because of GDPR considerations, so I don't build a score on it.

Offer

The offer score tells you whether people decline the job itself rather than the message. Tag every reply. Score 2 if one offer reason (comp, location or remote, or level) outnumbers every other specific decline reason combined. Score 1 if one offer reason recurs but doesn't lead. Score 0 if declines are generic, like "not now" or "happy where I am".

For a baseline, here's the reply mix in a sourcing-platform study of more than 400,000 replies:

• 31.4% expressed interest, and another 30.7% moved straight to scheduling.
• "Not interested" made up 28.9% of email replies and 26.7% of LinkedIn replies.
• One reply in ten flagged a competing process. That's normal noise, not an offer signal.

The later CRM study puts "thanks, but no thanks" replies at around 50%. Nobody publishes shares for comp, location, or remove-me replies, so your own tags are the only baseline. Don't take this score from rejection logs. In one study, salary appeared in only 1.0% of rejections logged during active conversation, while 72.2% were logged as "not interested", a label that soaks up the real reason.

Screen calibration

The screen calibration score tells you whether the bar at screen matches the bar in the brief. An ATS benchmark puts the recruiter-screen pass rate at about 35% across all sources, and 52% for referrals. You hand-picked your outbound candidates, so they should do at least as well. Score 0 at or above 35% with scattered decline reasons, 1 if the rate is below 35% or one reason recurs, and 2 for both together.

There's no public benchmark for screen pass rates of sourced candidates specifically, and none for the rate at which outbound submittals get a hiring-manager interview, so track your own. One CRM source reports sourced candidates passing from application to first screen at 86%. It only counts a sourced person as an application once they've expressed interest, so don't compare that figure with your screen pass rate.

The shares come from one study's rejection logs, and the right-hand column is my reading of them. About 75% of screening rejections there are fit reasons, which that study calls a statement about the search.

Step-by-step: reading a stalled search

The procedure takes one search's counts, one channel at a time, through the gates and the five scores to a verdict. Most of it is counting. The slow step is tagging replies by hand.

About the timing in step 1: in the sourcing-platform data, the median time to a software engineer's first reply was 4.1 days, with the 75th percentile at 9.0. Interested replies come faster, with 39.2% arriving within 24 hours and 54.3% within 72 hours. If your role family isn't engineering, check its own time-to-reply figure before you set the window. Nine days is a safe default.

If you log LinkedIn InMail replies by hand, make sure your sequence tool records them. Replies that never get logged make the LinkedIn channel look dead.

Search:
Channel: email / LinkedIn
Cohort send dates:
Last touch sent on:
Re-read window ends on:

REACHABILITY
Delivered contacts:
Hard bounces:
Spam complaints (raw count):
Postmaster spam rate:
SPF / DKIM / DMARC: pass / fail

SAMPLE
Benchmark used (name, unit, rate):
Minimum n for a halving:
Gate cleared: yes / no

LIST FIT
Audit matches out of 20:
Non-matches were (role types):

MESSAGE
Contacts with any reply:
Contacts with interested reply:
Reply rate (any reply / delivered):
Interested rate (interested / delivered):
Replies on touch 1 / all replies:

OFFER (reply tags)
Interested:
Not now:
Comp:
Location or remote:
Level:
Not a fit:
Remove me:
Competing process:

SCREEN
Screened:
Passed:
Pass rate:
Decline reasons (count each):
Submitted:
Hiring manager declines and reasons:

SCORES (0 / 1 / 2)
Reachability:   List fit:   Message:   Offer:   Screen:
Verdict:
One change:
Count to watch:
Re-read date:

Mapping the scores to one verdict

Apply the dimensions in a fixed order (gates, list fit, offer, screen calibration, message), and the first one scoring 2 is your verdict. Message comes last because a bad list and a bad offer both look like bad copy, so you only blame the message once you've ruled the others out.

Message is the leftover verdict. Rule out the list and the offer before you rewrite a word.

For a re-target, cut the non-matches the audit found, such as founders, managers, or adjacent roles, and build the new cohort from a fresh list rather than patching the old one. Asking Refolk for the role with the exclusions written in gets you a clean cohort that you can audit again before you send.

For a re-pitch, change one element only. If touch 1 takes far more than 43.8% of replies, rewrite the follow-ups and leave the opener alone. If both are weak, start with the opener.

For a reopened brief, bring counts, a verbatim example, and two options, not a complaint about the market.

Subject: {Role} search: one decision I need from you before the next batch

Hi {Name},

Where the search stands after {n} delivered contacts, all sequences complete:
- Interested: {x} ({rate}%) against a role benchmark of {benchmark}%
- Screened: {s}, passed: {p} ({pass rate}%)
- List check: {audit} of 20 random contacts match the brief as written

What the declines say:
- {count} of {total} specific declines cite {comp / location / level / a must-have}
- A verbatim example: "{quote from a reply or screen note}"

Another batch against the same brief is unlikely to change this. One decision would:
- Option A: {the change, for example widen the band, open to remote in a named region, accept an adjacent industry}
- Option B: keep the brief as it is and move the target date to {date}, at roughly {n} contacts per 10 replies

I'll measure the change on the next {gate n} contacts and send you the screen pass rate and decline mix on {re-read date}.

How the read goes wrong

Most bad verdicts come from reading a real number against the wrong comparison, not from bad arithmetic. These eight failure modes produce false message failures and send the blame to the wrong owner, and each comes with a check.

1. Judging before the sequence ends. After touch 1, a search shows only 39% of the final benchmark reply rate, so it reads as a message failure when it isn't one. Check: use the date of the last touch for each contact, not the date of the first send.
2. Unit mismatch. Comparing a per-message email rate of 4.96% with a per-sequence rate of 21.3% produces a false message failure. Check: recompute your rate as contacts with any reply divided by contacts delivered, and compare like with like.
3. Reply rate inflated by "no thanks". Follow-ups add replies faster than interest, so the reply rate rises while the interested rate stalls. Check: score the interested rate, never the reply rate alone.
4. Title-only lists. 24% of the UK title-filtered sample in Refolk's index weren't ICs. They inflate the contacted count and pull the rate down, which looks like a message failure. Check: audit 20 random contacts against the brief before you score the message.
5. Bounce threshold treated as official. The 2% bounce line is a convention I couldn't verify against Google. Check: run Postmaster Tools reputation alongside your bounce count, and don't rely on bounces alone.
6. Comp problems hidden in rejection logs. Salary appears in only 1.0% of logged conversation rejections. Check: tag comp mentions in the reply text itself.
7. Below-volume spam rates. The 0.3% line was built for bulk volumes. At 10,000 delivered emails it takes 30 complaints to reach it, but at recruiter volumes, one complaint in 300 sends is already 0.33%, a figure I derived. Check: report the raw complaint count next to the percentage.
8. Screen failures blamed on candidates. About 75% of screening rejections in one study's logs were fit reasons, which say more about the search than the person. Check: line up screen decline reasons against the brief's must-haves before you say anything about candidate quality.

Where evidence is thin, say so in the verdict. Several gaps aren't publicly established: a minimum sample for recruiting, a screen pass benchmark for sourced candidates, a submittal-to-interview rate for outbound, and published shares of comp or location declines. For each one, your own counts across searches will become a better baseline than anything published.

Keeping the read current after the verdict

A verdict is a hypothesis with a date on it. Set the re-read for the moment the new cohort's last touch has had its full reply window, and judge only the new cohort against the same benchmark you used before.

Three habits keep the read honest over time:

• Re-check benchmarks by unit, not by headline. When a source publishes a new edition, check whether it counts per sequence or per message, and what it calls a reply, before you take its rate. The latest CRM edition covers 6.2 million email sequences and 15.5 million messages, but I couldn't establish its headline rates, so I don't score against it yet.
• Re-check sender rules every read. Mailbox provider thresholds and enforcement change. Postmaster Tools and the Gmail sender guidelines are the authorities, not a figure from memory.
• Keep your own archive. Save every tally sheet. After a few searches in the same role family, your own interested rate and decline mix will be the most relevant benchmark you have.