Recruiting automation: CVs classified the day they arrive, and a timing signal next to every name

Before recruiting automation: a week measured in placements

Sunday morning. One role, dozens of CVs in the inbox, and someone has only managed to open the first few. A recruiter running three roles at once knows that the day will end before the pile does, again.

The good candidate in the pile had already signed elsewhere by the time anyone reached them. The second interview was never booked, because nobody had time to get back to the candidate. A role the client offered was turned down, because the desk was full. And the CV from three years ago, of someone who fits exactly today, sits in the pool and nobody remembers it.

Anyone paid per placement does the arithmetic themselves: every hour spent on a pile is an hour with no placement in it.

In every one of these cases the information existed. It sat in the inbox, in the old pool or in a LinkedIn Recruiter alert skimmed in passing, and nobody reached it on the day it mattered. This is not a failure of judgement. It is a failure of timing.

Every section below answers two questions only: which part of the week disappears, and what you see on the desk instead.

Placement fit, not paper fit: what recruiting automation puts in order

Paper fit is what the CV says against what the role requires. Placement fit is paper fit, plus timing, plus a way to reach the person: the right person, at the moment they are ready to move, and reachable today.

The advantage in recruiting was always timing, not reading. The problem is that the timing signals are scattered.

The automation does three things first: it classifies every CV on the day it arrives, brings the pool you already built back to life, and puts every signal into one working list, ordered by timing, with the reason next to every name.

The system ranks, summarises and presents. A person decides who is interviewed and who is hired, and professional judgement and responsibility stay with you.

AI CV screening and job matching: the morning pile becomes a list with a reason per name

What disappears: opening every email, downloading every file, reading a page and a half to discover there is no licence. And keyword search too, which misses the candidate who wrote "statistical modelling in Python" instead of "machine learning".

Every morning the process pulls the CVs that arrived in the inbox and in the recruiting system, reads PDF, Word and an image from WhatsApp, in Hebrew and in English, and produces one uniform structure:

• Employers and years of experience
• Tools and systems
• Licences and certifications
• Availability and area, as written
• And duplicates against the pool, merged into one card

Matching to the role is done once, in words. The criteria are written from the job description, together with you, before the run, and kept in a log for every role. AI reads the CV as free text and recognises that the same thing was said in other words. That is the part rigid rules never could do.

What you see instead: one card per candidate, the same fields every time, and for every role a ranked list with a paragraph of reasons for every name. Each point in the reasons refers to the criterion it answers. A screening question left unanswered is marked as missing, not inferred. A candidate the system is unsure about is flagged for a person to check. Anyone not at the top of the list stays on the list.

Fields such as age and year of birth, graduation year, address, marital status, reserve duty and photo are hidden before ranking.

Tagula CV Finder does that same-day classification every morning at one placement agency. The details are in the next section.

Tagula CV Finder: automated CV screening running today at one placement agency

This is not a description of what could be built. Tagula CV Finder is a tool we built, and it runs today on the CV pool of one placement agency. We do not publish its name. What the tool does:

• Pulls the CVs that arrived by email, every day
• Analyses and classifies them, so that a reply can go out fast
• Cross-checks them against the old CVs already in the pool
• Infers the likely professional development since the CV was filed
• And proposes fit for newly opened roles

The fourth item is the point. A CV from three years ago, from someone who was a junior then. The tool says: it is likely that today they are at the next level, and here is the role opened this week that they fit. The pool the placement agency already paid to build stops being an archive and becomes a live source. That is placement fit in practice: approaching the person the time in between has made a fit. The inference is a suggestion you verify with the candidate, not a fact.

Placement in record time is what the tool makes possible, not what it promises: a CV classified today can be answered today.

In one process measured at one placement agency, the daily CV review took between two and a half and four hours a day on average. The tool does the daily classification run in about two minutes. This is the classification run, not the whole recruiting process: interviews, calls and decisions kept their length.

Since then, the company reports that the number of placements has grown by more than fifty percent, and attributes it to what it did with the freed time: it took on more roles and entered more fields. That is one client and one process, not what to expect at yours. What it is worth at yours is measured at yours.

Passive candidate sourcing: who is ready to move this week, and how you know

What disappears: the approach that arrives six months after the candidate has already moved, and the standing "maybe" list that nobody maintains.

There is no automated access to LinkedIn or to any other platform here. A timing signal comes from four places:

• LinkedIn Recruiter's own tooling, in the recruiter's hands. To the extent that LinkedIn Recruiter shows such signals, for example Open to Work or a saved-search alert, a person reads them there. The moment you record the candidate in your system, the process opens a task and a reminder to follow up, and nothing more. The cross-check against your history and the draft wait until the candidate replies and hands over their details.
• Your own pool. Who applied again, who updated a CV, who replied to an old message, whose process ended a few months ago, and whose CV is old enough to assume they have progressed.
• Public company-level announcements. Layoffs, a closure, a merger or a relocation the company published itself: a press release on the company website, an exchange filing or an item on a news site. Not from social networks. A weekly list of companies with a link to the source. Companies, not a profile of any person.
• A candidate's reply. The signal that vanishes fastest in the inbox. Every day the process surfaces the list of who is waiting for you, sorted by how long the reply has been waiting.

What you see instead: a daily ready-to-talk list, sorted by timing, with the trigger written next to every name. For anyone whose CV is in the pool, the first approach is drafted from the CV and the role, and you send it, from your own account. No first approach to a candidate is sent automatically.

A signal is a suggestion to reach out, not a decision. Timing is the advantage of whoever arrives first, and it is built from what the candidate themselves marked, handed over or published.

After the list: interview scheduling automation and the candidate nobody answered

What disappears: three emails back and forth to set a time, the summary written from memory two days after the interview, and the candidate left without an answer who went to a competitor.

A timing signal is worth something only if the approach goes out the same day, and what follows does not get stuck behind it. These are the steps after the list that can be built:

• Interview scheduling. From the interviewers' calendars, including rescheduling and reminders.
• A structured interview summary. Against the role's criteria, after a recording notice at the start of the call. The interviewer's impression stays a separate field.
• Status updates. To the candidate and to the employer client.
• A uniform screening questionnaire. Sent to everyone in the same wording, with the answers filed on the card.
• Offer and onboarding documents. From the record, with a reminder of the date.
• A WhatsApp AI assistant for candidates. It presents itself as such, and passes to a person any question it did not answer.

What you see instead: a full calendar, a summary in the system minutes after the call, and any name that has not moved in a week is flagged.

Recruiting automation for placement agencies: the full desk and the role turned down

What disappears: a client's role turned down because there is no capacity, and the one candidate who fits three clients that nobody cross-checked.

When you are paid per placement, manual screening is what sets how many roles can be held at once. When it comes off the desk, that limit can move. At one placement agency it moved, and the figures are in the Tagula CV Finder section above.

What you see instead: every new role from every client runs against the whole pool on the day it opens, including whoever reached the final round for another role and whoever was ruled out only because of timing. One candidate against several roles, within what they consented to, and one working list for the desk.

This is what the placement agency whose process was measured did with the freed time: more roles and more fields. What it did, not what is promised to you.

A placement agency's pool is larger and older than a single employer's, so your retention and deletion policy is applied inside the process.

Recruiting automation by sector: what differs in high-tech, IT, finance and industry

Timing is the same in every sector. What is checked in the CV differs, and that is what is tuned at the measurement stage:

• High-tech. The same skill is written ten ways, and many CVs were written with AI and look identical. What is checked differently: semantic fit against the role's criteria, and evidence the candidate attached themselves: a portfolio, code or a take-home assignment. Not keywords. And the signal that matters here more than any other: who replied.
• Technology: IT, infrastructure, systems implementers and engineering. The role rests on a specific system, Priority, SAP, Microsoft 365 or a particular cloud provider, and on the vendor's certification. What is checked differently: hands-on experience with the system named in the role and the certification's validity, as a filled or empty field on the card.
• Finance and insurance. The gate is a licence: investment adviser, portfolio manager, insurance agent, pension adviser. What is checked differently: the licence number the candidate provided is checked against the public register of the Israel Securities Authority or of the Capital Market, Insurance and Savings Authority, to the extent that the register allows such a check. That is checked at the measurement stage. Not a criminal record check. That is not built.
• Industry, construction, logistics and healthcare. A certification has an expiry date: work at height, forklift or a Ministry of Health licence. And the candidates live on WhatsApp, not on email. What is checked differently: the certification's validity and its expiry date, not only that it exists. At Metal Stone this is a control that was actually built: a real-time alert on training and refresher courses approaching their expiry. That is an onboarding and compliance control, not recruiting.

In every sector the list is ranked differently, and in every sector a person decides who is interviewed.

Candidate data and CVs: what goes to an AI provider and what stays in your systems

A CV is personal data, so this is a conversation before anything starts, including the part that is convenient not to say.

A process that uses AI sends data to a third-party provider. The provider may collect, store or use it. Each provider has its own policy, set by the provider and changed from time to time, and we neither control it nor answer for it.

So we first map what must go to the provider and what stays out. Name, ID number, photo and address can stay out. The less that leaves, the less depends on someone else's policy.

The CVs and the candidate data stay in your systems: the recruiting system, the mailbox, the folders. The process reads from them and writes back to them, and does not set up a candidate pool at Tagula. The access we ask for is limited to what the process must read and write, no wider than that, and is set in a written agreement.

The cross-check, the ranking and the draft work on people who have already handed you their details. What may be done with an old file is set by what the candidate consented to when they handed it over, and that is checked at the measurement stage. The duties to candidates under the Privacy Protection Law, including notice, access and deletion, stay with you.

Running on your own servers, with no third-party provider, is an option examined case by case, and not the default. What is binding, what is kept, where and for how long, is set in a written agreement and not on this page.

What do we not say, and what is not built?

This is not false modesty. We do not promise hours, placements or time to fill. What was measured at one client belongs to that client.

• There is no automated access to LinkedIn or to any other platform, and no process that discovers on its own that a candidate changed status. What comes from there comes through LinkedIn's own tools, to you.
• No profile is built of a person who did not approach you, and no further information is gathered about them from social networks, code repositories or job boards. No lists are bought.
• The system rejects nobody and invites nobody to an interview. A person does that.
• No criminal record check, and no emotion or personality analysis from video or voice.
• No "objective, bias-free ranking". The criteria are visible, the reasons are kept, and the ranking is checked against the decisions you actually made.
• No promise of compliance with the Privacy Protection Law, the Equal Employment Opportunities Law or European regulation. The process is built so that you have something to show.
• Not a new recruiting system. It runs alongside the one you have.

And if your question is what happens after onboarding, a leaver still being paid or still with access to systems, that is a separate subject: Continuous controls in the organisation.

Where do you start?

With the role that hurt the most last quarter. The one that closed elsewhere, or the one that was turned down. If nothing has happened yet, we start with Sunday's inbox.

Bring a description of the process to the call, and nobody's CV: how many roles are open at once, how many CVs a day and from which sources, which recruiting system you have (Niloosoft Hunter, Civi, Comeet, Greenhouse, or a mailbox and folders), and who decides.

Then the measurement stage:

• We measure how many hours screening takes today
• We check the connection to the recruiting system and the mailbox
• We write criteria for one role
• We do a parallel run on the last quarter: the system ranks, you keep deciding as usual, and we compare. We correct the criteria, not the result
• And only then do we decide whether it justifies building

The test question: how many of last year's placements were already sitting in your pool on the day the role opened. A process that would have changed nothing last year will probably not be built, and that is an answer we give the client. We measure, we simplify, and only then do we automate.

• Email: info@tagula.ai
• Phone and WhatsApp: 054-650-4053

More on automation services for business: how it works, and when we say no

Frequently asked questions