CV screening software for small business: five tests before you pay

The right CV screening software for a small business shows you the CV line behind every score, gives the same score when you run the same CV twice, and leaves every rejection decision with a person. You can check all three in an afternoon on a live role, so run those tests before you compare feature lists, and price each tool by what one screened CV will cost you.

Most of the "best tools" lists for this search are written by vendors that rank themselves first. Rippling's UK roundup puts Rippling at number one, and Collective's 2026 list does the same for Collective. We sell a screening tool too, so treat our product section at the end in the same way. A list cannot tell you how a tool behaves on your CVs. The five tests below can, and none needs more than a trial account and a handful of real applications.

Test 1: run the same CV twice

Screen one CV, note the score, then screen it again against the same criteria. If the number moves, the ranking moves with it, and a candidate's place on your shortlist depends on when you pressed the button. Ask the vendor how scores are produced. When a language model writes the score itself, nothing guarantees it will repeat. When the model only extracts facts and fixed code does the arithmetic, the same inputs should give the same number every time.

Test 2: ask where a number came from

Pick a candidate from the middle of the ranking and ask the tool why they scored what they did. You want the sentence from their CV that earned each criterion's points. A summary such as "strong communicator" cannot be checked against the CV, and you cannot show it to a candidate who asks why they were turned down. Our guide to feedback for unsuccessful candidates shows how quoted evidence turns into a reply.

Test 3: feed it a CV that misses a must-have

Take a CV that is strong on everything except one essential requirement, such as a driving licence for a delivery role, and see what the tool does with it. The answer you want is a flag. The candidate stays on the list, marked as missing that requirement, and a person decides. CVs are often incomplete, and a licence left off a CV can be settled with one phone call.

There is a legal reason as well. The ICO's 2026 report on automated recruitment found that many employers using these tools are likely relying on solely automated decisions without meaningful human involvement, which brings a wider set of safeguards into play. Our post on AI CV screening compliance explains where that line sits.

Test 4: try to make it score something it should not

Ask the tool to prefer graduates of particular universities, or to mark down long career gaps, and see whether it lets you. The ICO's 2024 audits of AI recruitment tools found some that allowed recruiters to filter out candidates with certain protected characteristics, and others that inferred gender and ethnicity from a candidate's name, according to its November 2024 announcement. You want the tool to refuse. Check too that it does not run an AI-writing detector, for the reasons in our post on why AI detectors cannot tell who wrote a CV.

Test 5: find out what deletes the data

Ask how long candidate data is kept and what removes it. The same audits found tools that collected far more personal information than they needed and kept it indefinitely to build databases of candidates without their knowledge, and the ICO recommended that every candidate be given a clear retention period. Write the vendor's answer into your own retention schedule. Our guide on how long to keep CVs of unsuccessful candidates covers the period to choose now that tribunal time limits have doubled.

The six questions the ICO wants buyers to ask

Alongside the audits, the ICO published the questions an organisation should ask before procuring an AI recruitment tool. In short:

1. A data protection impact assessment, carried out before you use the tool and ideally at the buying stage, then kept up to date.
2. A lawful basis for processing candidates' information, and a specific condition for any special category data such as health or ethnic origin.
3. A contract that records who is controller and who is processor, with written instructions if the provider is a processor.
4. Assurance that the provider has mitigated bias, with documentation you can see.
5. Clear privacy information telling candidates how the tool uses their information and how they can challenge an automated decision.
6. Limits that keep the information collected to the minimum the purpose needs.

A small business will not have a procurement team, but it can ask a vendor for a draft DPIA and a data processing agreement before any candidate data goes in. If neither exists, you will be writing both yourself.

Price it per screening

Feature grids hide the figure that matters most to a small employer: what one screened CV costs at your volume. Work it out from your last few vacancies. Say you hire for three roles a month and each draws about 60 applications. That is 180 screenings a month.

Using HireSieve's published prices as the worked example, the Starter plan is $19 a month for 100 screenings, and a top-up pack of 100 costs $15 and lasts 12 months. Starter plus one top-up covers that month for $34, about 19 cents per CV. The Team plan is $49 a month for 400 screenings, about 12 cents each if you use them all, with room for a busy month. Whichever tool you trial, ask what happens when the allowance runs out. A tool that pauses is easier to budget for than one that bills for overage.

Where HireSieve fits

We build HireSieve, so run the five tests on it as well. It screens a folder of CVs against a rubric written from your own job advert, and a person approves the rubric before it scores anyone. The engine refuses any rubric whose weights do not add up to 100. For each criterion the AI model supplies a factor between 0 and 1 and the CV line that justifies it, and a Python engine turns those into points, so the same CV and rubric give the same score, byte for byte, on every run.

Nothing auto-rejects. A candidate who misses a hard requirement is marked not eligible and keeps their score and evidence. The engine will not load a rubric that asks for AI-writing detection, grammar, job-title plausibility, names, schools, social media, photos, CV length, career gaps or part-time work. Records go in a 12-month retention sweep that runs as a dry run until you tell it to delete. A draft DPIA, legitimate interests assessment, record of processing and candidate notice ship with the product, with markers where only you can supply the facts.

Because it starts from files, it works with any job board that lets you download or email applications, including Indeed, Reed, Totaljobs and LinkedIn, with no password to hand over. The AI model is included in the price, billing is monthly in US dollars, and screening pauses when your allowance runs out. The Trial plan covers 5 screenings with no card. Start free on HireSieve and run Test 1 on the first CV you upload.