A practical automation plan

Everything downstream depends on having your history in a structured form: roles with dates, achievements with numbers, skills with the evidence for each. An hour spent here makes every later step better.

Add the constraints that never change — location, right to work, notice period, compensation expectation. These are what stop automation suggesting or submitting things you cannot take.

Add the refusals as well, because nothing infers them. Sectors you will not work in, employers to exclude, whether you will take a management role, whether you will relocate — five minutes of this prevents most of the recommendations that make people abandon a tool in the first week.

Screening is the highest-value, lowest-risk automation available. Set up searches across the sources that matter in your field and let a tool filter to the ones worth reading.

Keep the filter transparent and check what it rejected occasionally, particularly in the first week. A filter that quietly excludes a category you care about is worse than no filter, and you will only find out by looking.

Spend the first week correcting it rather than consuming it. Ten explicit corrections — this one was wrong, this one you should have shown me — improve the feed more than any amount of settings tuning, and they are the only way the system learns what you actually mean.

• Structured profile and fixed constraints
• Automated discovery and screening
• Application tracking and follow-up reminders
• Assisted drafting, with review
• Automated submission, only for low-stakes applications

Automatic capture of what you applied to, when, and with which CV — plus reminders for follow-ups and deadlines — removes the persistent background load of a search.

This is the step people skip and regret. Three months in, a candidate with no record cannot tell which version of their CV an interviewer is holding, or whether they already followed up.

It is also the only source of evidence about your own search. Without a record there is no way to answer whether the tailored applications did better than the quick ones, which means no way to decide where next week’s hours should go.

Use a tool to draft and to point out which requirements your CV does not evidence. Read everything before it goes out, and remove any claim you would not defend in an interview.

The checks that matter: is every fact true, is anything specific to this role, and would you be comfortable if the recruiter knew exactly how it was produced.

That third question is the one that does the work. It is not a moral test but a practical one — if the answer is no, the document almost certainly reads as generated to someone who sees forty a day, and the reason you are uncomfortable is the same reason it will not land.

Full auto-apply is the last step and the one with real downside. Reserve it for structured forms with no free text, on roles you are exploring rather than targeting.

For anything you actually want, review before it goes. The applications that decide your search are the ones worth the ten minutes.

Set the caps before you turn it on rather than after something goes wrong: a weekly maximum, a per-employer maximum, and an exclusion list. All three are trivial to configure in advance and impossible to apply retroactively.

Some things should stay manual not because automation cannot do them but because doing them yourself is the point. Deciding which roles you want, messaging someone you actually know, and preparing for an interview all belong in this category.

Choosing your targets is the most important of these. A search where a tool decided what you were aiming at will eventually deliver an offer for a job you did not want, and by then several weeks have gone into it.

Networking is the other one worth protecting. It is the highest-return activity in most searches and the least automatable, and the practical risk of a well-automated pipeline is that it fills the time that would otherwise have gone there.

• Choosing what you are aiming at
• Messaging people you actually know
• Interview preparation and the interviews themselves
• Anything where a person will read it as personal

After a few weeks, look at what changed. More applications with the same number of responses means the automation traded quality for volume, and the right response is to send fewer and better ones.

The outcome to watch is responses per hour spent, not applications sent. That is the number automation should move, and the one that tells you whether the setup is working.

Be willing to switch parts of it off. Automation that is not moving that number is costing attention for nothing, and a smaller setup you actually trust beats a complete one you have stopped reading.