AI recruitment tool vendor comparison template

Set the hiring task before you score the tools

Write the handoff you need at the top of the sheet: a list to review, interested candidates to contact, or confirmed interviews. A demo can make each of those look like a finished job. Your team may still have several hours of work left.

Write down the open role, target market, hiring manager, and review deadline. For each requirement, specify what evidence would satisfy it. "Senior backend engineer" is too broad to judge consistently. "Owned a production Go service and can explain an incident they resolved" gives the reviewer something to check.

Keep requirements related to the work. Record a missing fact as unknown. An AI summary is a lead to verify, and a public profile rarely establishes everything you need to know about a candidate.

Copy this brief into your evaluation document:

Start with a role you know. If you have recently hired a backend engineer, you can probably explain why a profile fits, which gaps matter, and which requirements you would relax. Save the unfamiliar role for a second test, when you can distinguish a search problem from an unclear brief.

Copy the vendor comparison scorecard

The suggested weights below total 100. Change them before the trial if your main problem is outreach or coordination. Keep the same weights for every vendor in that comparison.

Score each row from 0 to 4: 0 means the required task failed, 1 means major rework, 2 means usable with corrections, 3 means it met your requirement, and 4 means it exceeded a target you wrote down beforehand. Use "not tested" when you lack evidence.

For a completed scorecard, calculate each row as weight × score ÷ 4, then add the rows. Report a total out of 100 only after every row has evidence. If a row is untested, show the partial score and the tested weight, such as 48 points from 70 tested, and postpone the overall ranking. Do not rescale that result to 100; doing so rewards missing tests.

Write a measurable target beside each row. For candidate fit, this might be the proportion of reviewed prospects that meet your sourcing requirements. For effort, it might be a maximum number of review minutes per qualified prospect. Set those targets for your role before seeing results. The weights and scoring scale here are an editorial proposal, not an industry benchmark.

Write any requirement that must pass above the scores. If candidate export is required, specify the fields and inspect an actual export. If your team needs to review messages before sending, test that action. A score of 90 is little comfort when the missing feature is the one your recruiter uses every morning.

The table is meant to support a purchasing decision. It does not establish whether one vendor is better for every role, market, or team.

Review candidates with a shared rubric

Choose a review sample before you open the results. For a small pilot, the first 20 unique candidates per vendor is a manageable starting point. Twenty is a workload suggestion, not a statistically representative benchmark.

Keep the order each vendor returned and log every row you inspect until you reach 20 unique people or exhaust the results. Count duplicate rows and existing pipeline candidates separately. Use one stable profile URL or internal candidate ID to deduplicate, and keep those checking minutes in the effort total.

Report qualified prospects ÷ unique prospects reviewed as sample fit rate. Also report the number returned, duplicates, people already known, and new qualified prospects. A vendor that returns only 12 unique people has a sample of 12; do not silently treat the eight missing people as qualified or compare raw qualified counts without the denominators.

For each candidate, use a row like this:

Count someone as a qualified prospect only when your reviewer finds enough evidence for the agreed sourcing requirements. That label does not mean the person wants the job or will pass an interview. Keep interest and interview results in separate columns.

Have a second reviewer check disputed decisions, with vendor names hidden where practical. Save the reason for each disagreement. If both reviewers keep interpreting one requirement differently, improve the brief and re-review both samples against the same version before choosing a vendor.

Run a 14-day pilot

Fourteen days is a proposed schedule for checking a workflow. It is too short to establish quality of hire, retention, or a stable interview conversion rate. Some candidates will still be deciding when the trial ends.

If outreach is part of the pilot, confirm the message, audience, sender, and follow-up plan before sending. Do not contact the same person through two competing tools to see which gets a reply. Assign a candidate to one outreach path and flag any overlap.

Record sent messages, delivered messages where your provider reports them, replies, positive replies, and confirmed interview bookings separately. An accepted send request is not delivery, and a reply is not necessarily interest.

Give each contacted candidate the same observation window, such as seven complete days after the first message. Anyone who has not reached that window is pending. Report the counts and the chosen window together; an email sent yesterday should not be treated as a seven-day non-response. If delivery data is unavailable, label it unavailable and use sent recipients as the denominator for a clearly named reply-per-sent rate.

For a booking, check that a real calendar event exists and that the candidate has agreed to it. For an ATS handoff, inspect the receiving record. A screenshot of a success banner cannot show whether the record arrived with the required fields.

Test the current product instead of relying on its category label. Juicebox's May 2026 Agents announcement describes search calibration and outreach planning. A comparison that assumes every sourcing vendor stops at a search box would miss those claims. Ask the vendor to demonstrate the specific step you need, then record what happened in your pilot. Juicebox Agents announcement

Price the work you actually used

A subscription price does not tell you the cost of a usable candidate. A plan may meter seats, contacts, exports, agents, messages, or some combination of them. Use the AI recruiting pricing guide to identify which billing units you need to check, then verify current terms with each vendor.

For a worked example, the SeekOut pricing breakdown separates the annual payment, contact and export allowances, and seat questions to resolve before a pilot.

Juicebox's official pricing page, for example, lists contact and export allowances separately and presents Agents as an add-on. That is enough reason to ask which actions your pilot consumes. Do not assume two vendors use "credit" to mean the same thing. Juicebox pricing

Copy these rows into the cost sheet:

Keep two cost totals: cash committed to start using the product, and cost allocated to the pilot. A prepaid annual contract belongs in the first total even if you allocate only part of it to two weeks. Show your allocation method so a free trial does not conceal the cost of continued use.

Use allocated pilot cost ÷ unique qualified prospects to calculate cost per qualified prospect. Include the internal time your team spent checking the results. If no prospect qualifies, report the spend and zero qualified prospects; there is no finite cost per qualified prospect to publish.

For example, assume $120 in allocated software and usage costs plus four review hours valued at $50 an hour. The total is $320. If 20 unique reviewed prospects include eight who meet the agreed requirements, sample fit is 40% and cost per qualified prospect is $40. These are invented numbers to show the calculation, not Metix AI results or a market benchmark. Record how many of those eight were already in your pipeline before you estimate the value of newly sourced prospects.

Keep cost per interested candidate separate. A tool may return useful profiles while your offer, message, or timing produces few replies. Combining those stages hides the problem you need to fix.

If you need a longer budgeting view, use the same cost inputs for your expected monthly workload and your next hiring cycle. Keep the forecast apart from observed pilot costs. The calculations can inform the purchase without pretending two weeks of activity predict a full year.

Where Metix AI fits in this template

Metix AI's published monthly plans are $99, $199, $299, and $499. They include monthly Credits and AI Agents. The free 14-day trial includes 1 AI Agent and 500 Credits, with no card required.

Outside an AI Agent workflow, a contact unlock uses 10 Credits and a Deep Search list uses 100 Credits. An additional AI Agent above the plan allowance uses 3,000 Credits. Work inside an active Agent's workflow uses 0 Credits. These are usage units; they do not buy or guarantee a hire.

The listed plans are self-serve subscriptions. Delivery service, where Metix AI staff run the search, is separate and quoted. Apply the same rubric to Metix AI as to every other vendor, including the work your team still owns. Metix AI pricing, checked October 12, 2026