Articles · Recruitment

Vacancies not getting filled? The CVs are read the day they arrive

12 October 2026

Applications arrive by the thousand, the consultants are busy from morning till evening, CVs go out to clients in batches, and few vacancies get filled. The owner of a recruitment agency sees the bottleneck at the start: applications are screened slowly, and the strong candidates find a job elsewhere in the meantime. A vendor is offering an AI agent for CV screening.

A recruitment funnel is usually read from the top down: applications, screening, CV to the client, interview, start date. From the top it always looks like a question of volume. We will read it from the bottom — from the filled vacancy backwards — at Farthingale Staffing Partners. The agency is invented: it is one of the companies we test our audit agent on, and the agent wrote the report on three months of its data. In that time the agency opened 150 job orders and received 5,041 applications.

At the very bottom: some clients' vacancies get filled, others' never do

Of the 150 job orders, 49 came from four clients on framework agreements. Twenty-six of those ended in the period, and not one was filled by the agency's candidate. Of the 41 other job orders that ended, 23 were filled — 56.1%.

A client releases a framework role to five or six agencies at once through its portal, and the fee on it is the lowest. The work is no less: framework roles drew 1,711 applications and 253 CVs sent, all of them read, rung and submitted like any other.

That also explains the desk that worried the owner: the most CVs sent and the fewest placements. On that desk, 42 of the 59 job orders are framework roles. On standard terms the same desk filled 4 of 9 ended job orders, 44.4%, against 38.5% for the others. The gap lies in the desk's clients, not in whoever runs it.

One step up: the CV is with the client. Nearly a third are lost here

Over the three months, 881 CVs were sent to clients. Of those, 280 ended with the candidate withdrawing — 31.8%. Twenty-three ended in a placement.

Of 881 CVs sent to clients, 280 ended with the candidate withdrawing and 23 with a placement.

The share withdrawing is about the same on exclusive, standard and framework work, and for every consultant. So it is a property of the step itself, not of anyone's work. Another 111 CVs never received any answer before the job ended; 85 of those were on framework roles.

One word describes what happens at this step: waiting. The consultant sends the CV and waits for the client, and nobody agrees an answer date with any client. How long the wait lasts, the agency does not know. The consultant says a week or more, and over two weeks on the framework clients' portals; the owner's impression was a day or two. The recruitment software records the date of every step, but the licence does not allow it to be exported. That the wait is what makes candidates withdraw is likely, but not proven.

Higher still: applications are read the same day

Now the step the owner complained about. About 1,680 applications a month are read by one resourcer, in three rounds a day: nine, noon and four. On the morning of the review 74 were unread, and every one had arrived after the last round on Friday. Rejected candidates get an email the same evening. About one application in eight goes on to the consultants.

The owner believed the consultants read CVs themselves, three hours each a day, and described that figure as an impression. The three hours a day are real, but they belong to one person: 55.4 hours a month, about a third of the resourcer's paid time.

At the very top stands the vendor's agent

The CV screening agent costs 650 EUR a month and 2,400 EUR to set up. It speeds up a step that holds nothing back, and it does not remove a single payment the agency makes. On the supplier's own 80%, which nobody has measured, the agent would free 20.3 hours a month of salaried time and add 24 hours a month of checking. No post can be reduced on that.

From here it is clear why the loss went unnoticed. Everything above the sending of the CV happens inside the agency, in plain view: applications arrive by the thousand and people are busy. Everything below it happens at the client's, and that wait has no time limit: none was agreed, and the dates cannot be exported. A candidate who withdrew while the client stayed silent is recorded as the candidate's decision. On top of that, the meaning of the stages in the software is written down nowhere: staff could not define several of them.

Answer dates cost nothing to agree; reminders come later

1. Agree an answer date when the agency takes a vacancy on. An answer on each CV within two working days, interview slots held in advance. It is free; start with exclusive and standard clients.
2. Switch on the export of step dates. The software supplier says about a week; ask for the price. Without the dates you cannot see whether the first rule works.
3. Decide what to do with framework roles. Get out the agreements first: nobody at the agency could say whether it must work every such role, or when one last ended in a fee.
4. Later, as a trial — automatic reminders to client contacts: 300 EUR to set up and 90 EUR a month. They do not reach clients who take CVs only through their portal.

There is a place for an agent in this funnel, only lower down than where it is being offered. Once the dates and the answer-date rule are in place, the report names an agent that lists the CVs with no answer and reminds client contacts. The limit of its usefulness is known in advance: chasing clients for an answer takes the consultants about 10 hours a week, and that is by one consultant's account. Calls with candidates and hiring managers, rejections and fees stay with people; how to draw that line is covered in the article on what to hand to an agent.

The report promises neither a saving nor extra fees. It names one number to watch: the share of sent CVs that end with the candidate withdrawing. It is 31.8% today; look again two months after answer dates start being agreed.

Three numbers an agency can pull today

1. How the CVs sent to clients ended. In the recruitment software, take every CV sent in the quarter and break them down by outcome: candidate placed, client said no, candidate withdrew, no answer. The warning signs are withdrawals outnumbering placements many times over, like 280 against 23, and a share that is the same for every consultant. Then the loss is below screening, and there is no point in speeding screening up.
2. Filled job orders among those that ended — separately by type of agreement. The list of job orders with status and terms. In the example, 0 of 26 for framework roles and 23 of 41 for the rest. Next to it, count how many applications and sent CVs went into the type with the worst result. The warning sign is a zero where a noticeable part of the work goes: that is a question about the agreements, and software will not answer it.
3. How many applications are unread right now, and how old the oldest one is. Open the incoming applications on a Monday morning. In the example there are 74, and all arrived after Friday's last round. If your oldest application has been sitting for a week, screening really is the bottleneck, and a conversation about a CV screening agent makes sense.

A fourth number you may not be able to get: how many days a client takes to answer a CV. Check whether your software exports the step dates. If it does not, that is your first finding: the main wait in your funnel is one that nobody can measure.

All articles