When Your Response Rate Drops and Nothing Changed

You’ve been tracking for six weeks. Early on, replies were coming back at a rate you could live with. Now they aren’t, and you haven’t changed anything — same documents, same approach, same effort.

The instinct at this point is to rebuild everything at once. Resist it for one week, because a drop in a self-tracked rate has several possible causes and only some of them are about you. If you rewrite your whole approach in response to a measurement artefact, you’ve destroyed the one baseline you had and you’ll never know what happened.

Work through this in order. It takes twenty minutes.

First: is it a real change, or a small number?

The most common cause, by a wide margin.

Response rate on small samples is extremely jumpy. Suppose you sent nine applications last month and three came back; that’s a third. This month you sent seven and one came back; that’s a seventh. Those two figures look like a collapse and are entirely consistent with nothing having changed at all — you’re comparing two handfuls.

Two habits fix this:

Use a rolling window, not a weekly figure. Compute over the last twenty-five applications rather than the last calendar week. Weekly rates on a search that sends three or four a week are noise with a date on them.

Look at counts alongside the rate. Write both: “last 25 applications: 4 replies”. A rate on its own hides how much it’s built from, and a rate built from six events shouldn’t change any decision.

If the drop is fewer than about five events’ worth of difference, stop here. There’s nothing to explain yet. Send another fifteen applications and look again.

Second: did the denominator change?

The subtler measurement problem, and it catches careful people.

Early in a search, most people apply only to things they genuinely fit, because those are the roles they found first and were excited about. Later, the obvious matches are exhausted and the marginal ones creep in: a role a level up, one where you meet most but not all of the stated basics, one in an adjacent sector.

Nothing about your documents changed. What changed is the population you’re applying to. The rate fell because the mix got harder, which is a completely different problem from “my approach stopped working” and has a completely different fix.

Check it directly. Take the ten most recent applications and the ten earliest, and score each honestly: did I meet the stated basics, and was this squarely the kind of role I’d been targeting? If the recent ten are visibly weaker matches, you’ve found your cause.

Related, and worth checking at the same time: definition drift. Are you still counting the same things? If you started counting quick speculative submissions as applications halfway through, or started including roles you found via a recruiter alongside cold applications, the two periods aren’t comparable. Hygiene rules for what counts are in how many applications is the right number, and the important part is that they don’t change mid-measurement.

Third: did the channel mix shift?

Break the last twenty-five applications down by source and compare to the previous twenty-five.

This is where the source column pays for itself. A very common pattern: the first month contained two or three referral or recruiter-sourced applications, which converted well, and the second month was all cold applications, which convert less well. The blended rate fell without either channel’s rate changing.

If that’s what happened, the finding isn’t “my applications got worse”. It’s “I stopped generating warm introductions”, and the action is to go and generate some, not to rewrite a document.

Look at it per channel, always. A single blended response rate is nearly uninterpretable once you have more than one channel running.

Fourth: is it a stage shift rather than a rate shift?

Check where the responses stopped, because “response rate” bundles several transitions.

• Applications producing no reply at all is a targeting or reach problem.
• Replies arriving but all of them rejections is a different problem, and a milder one — you’re reaching humans.
• Screens happening but nothing progressing past them isn’t a response-rate issue at all, and no amount of application volume will move it.

A drop concentrated at one transition is much more diagnostic than a drop in the aggregate. If you recorded the stage each rejection arrived at — as suggested in the notes you take after every conversation — this takes two minutes.

Fifth: timing effects you don’t control

Some of these are real and none of them are about you.

Calendar slowdowns. Hiring processes visibly slow around major holiday periods and quarter boundaries in many places. Applications sent into a fortnight when half the hiring team is away often produce a reply weeks later or never, and the effect on a small sample is large.

Elapsed time isn’t uniform. If your rolling window includes applications sent nine days ago, some of those replies simply haven’t arrived yet. Always compute your rate on applications old enough to have resolved — past your own median time-to-reply, ideally past your ghost cutoff. Otherwise a productive recent week reads as a collapse purely because it’s recent. This is the single most common way people frighten themselves with their own data.

Market conditions in your specific niche. Sometimes the roles being advertised in your exact specialism thin out for a while. That’s a genuine condition, not a personal failing, and the response is usually to widen the search deliberately rather than to send more of the same — see widening the search one variable at a time.

If it survives all five checks

Then something did change, and it’s worth acting on. Two rules for acting.

Change one variable. Not the documents and the channel mix and the target roles and the volume. One, for a fixed period — three weeks or twenty applications — then measure. Changing four things means the next measurement is uninterpretable, and you will have burned six weeks to learn nothing.

Pick the variable by where the funnel broke. No replies at all points at targeting first, then at whether your material is reaching a human. Replies but no screens points elsewhere. Screens but no progress points somewhere else again. The site’s position is that targeting is the commonest cause and the cheapest to test, so start there unless your data says otherwise.

What to write down

In your weekly review, one line:

04 Jul — last 25 apps (all past cutoff): 4 replies, 2 screens.
         By source: referral 2/3, board 1/16, careers page 1/6.
         Prev 25: 7 replies, 3 screens. Referral was 3/6 then.
         Read: fewer referrals this month, not worse applications.
         Change: one referral conversation per week for 3 weeks.

Six lines, once a week. The value isn’t the number — it’s that when the rate moves you have a previous entry to compare against, and a record of what you already changed. Without it, every dip triggers a full rebuild, and a search where you rebuild everything monthly never accumulates any evidence about what works.