How to work out which applicant tracking system is best for your desk

Nearly every bad software decision in recruitment starts with a shortlist assembled before anyone wrote down who the buyer is. Four buyer types dominate this market and they want almost incompatible things.

The staffing agency needs many clients, submittals as records, margin on the placement and a candidate database that keeps earning. The corporate talent team needs one employer, approval chains, requisition budgeting, structured scorecards and an interview process that survives an audit. The small employer needs a shared list, faster replies and a runner-up pool, and often needs nothing else. The sourcing-led firm needs contact discovery, outbound sequencing and a database that supports rediscovery more than it needs a pipeline.

Write your buyer type in one sentence and put it at the top of the evaluation document. Every product that does not serve that sentence comes off the list, no matter how good the demo was.

A demo shows you a tenant that was built to demo well. A trial on your own data shows you the product. Load a real, messy subset of your database, including the duplicates and the records with three attachments, and watch what the import does with them. That single step eliminates more products than any feature comparison.

Then run the work. Configure two or three genuinely different processes rather than one. Send real submittals or real applications through and inspect what the recipient receives. Book real interviews across a panel with real calendar conflicts. Ask a colleague who did not attend the demo to complete a task without help, because software that needs the salesperson present is software your team will route around.

Finally, break it deliberately. Search for something specific and see whether the answer respects all your constraints. Try to export everything. Try to correct a mistake somebody made three stages ago. The recovery paths tell you more about a system than the happy path ever will.

AI is where this market's pricing has become least transparent. Almost every vendor now sells matching, parsing, summarising or sourcing assistance, and the commercial models differ enormously: included in a tier, metered by credits, sold as a per-seat add-on, or quoted separately at renewal. Ask which of those four it is, in writing, and ask what happens when a busy month exhausts the allowance. Mid-month throttling of resume parsing is a genuinely disruptive event.

Data ownership deserves the same directness. Ask whether your candidate data is used to train models, whether that is on by default, and whether it can be turned off without moving tiers. Ask where the data is held and who the sub-processors are. Ask what an export contains, whether notes and attachments are included, and whether you can run it yourself. A vendor that answers all of these quickly and in writing is telling you something useful about how it will behave later.

For our part: AI matching credits are a hundred per user per month on the Professional plan, resume parsing and auto-shortlisting are on the same plan rather than an upgrade, and export is a workflow you run rather than a ticket you raise. The full sub-processor list and data residency are published rather than available on request.

The migration is the part of the decision most buyers underestimate, and it is the reason so many agencies stay on a system they dislike. A modern candidate database is not a table of people. It is custom fields added over a decade by people who have left, notes with authorship and timestamps that matter, several attachments per record, ownership and split arrangements, job orders, submittals tied to specific contacts on specific dates, and placements with rates, fees, guarantee windows and extensions.

A generic importer moves candidates and contacts and loses most of the rest quietly. The failures are predictable: notes arrive stamped with the import account so the audit trail is gone, second and third attachments are dropped, ownership collapses to whoever ran the load, and duplicates are created rather than merged. You discover all of it about six weeks later.

Ask each vendor how a migration is scoped, who runs it, what the deliverable is before cutover, and what the rollback position is. The answer you want involves a discovery pass, a written scope naming what is out as clearly as what is in, a test load into a sandbox your team works in, and a reconciliation report you sign off. Our own answer is that the first migrations are paid, founder-led projects rather than a one-click importer, which is slower and more honest.

The published per-seat price is the least reliable number in this market. Three structures move the real cost. Seat minimums set a floor: Crelate publishes a five-seat minimum on its eighty-five dollar Basic tier, which changes what a four-person firm actually pays. Annual escalators raise the renewal above year one by contract, and Crelate publishes one. Modular pricing means the number you were quoted covers a core and the thing you came for is a pack: Vincere publishes from sixty-nine pounds with modular packs, and Tracker RMS publishes seventy-nine and ninety-nine dollar tiers with separately priced modules.

Tier gating is the fourth. Bullhorn publishes a ninety-nine dollar entry tier and a hundred and sixty-five dollar Core tier that are applicant tracking rather than CRM, with the CRM in a custom-priced Pro tier and AI search sold as a paid add-on. For an agency buying specifically for the CRM, that is the difference between the advertised price and the real one.

Some vendors do not publish at all. JobAdder does not publish pricing, and Greenhouse, Lever and Ceipal pricing is either unpublished or not something we have verified. That is not a criticism; it is a reason to get a written quote early rather than at the end of the evaluation.

Agency recruitment platforms are built on the many-client model, treat candidates as long-lived relationships and carry submittals and placements as records. Bullhorn, Loxo, Recruit CRM, Recruiterflow, Crelate, Vincere, Tracker RMS, JobAdder and Surhires all sit broadly here, and they differ mainly in depth, price and how much of the product is gated above the tier you are quoted.

Corporate applicant tracking systems such as Greenhouse and Lever are strong on structured internal hiring: approvals, scorecards, interview kits, and reporting a talent director recognises. They are awkward for agencies because the model assumes one employer. Budget trackers such as Manatal and Zoho Recruit are honest about being lighter and are frequently the correct answer for a small employer. Sourcing-led platforms lead with contact discovery and outbound sequencing and often sit alongside a system of record rather than replacing it.

Surhires sits in the first category, positioned as a full recruitment software where matching and parsing are on the working plan rather than an upgrade, with WhatsApp as a first-class channel, a three-seat minimum and no annual escalator. It is not an HRIS, a payroll system, a background-check provider or an assessment platform, and it does not do pay and bill.

Week one: write the buyer sentence, shortlist three products, get written answers to the commercial questions, and start trials on real data. Week two: run the same three tasks in each product with the same people, and score them on completion rather than impression. Then decide.

The most common failure is a four-month evaluation that ends with the product that demoed best in month one. The second most common is choosing on a feature nobody uses. Score on the work your desk does every day, weight the migration and the exit terms properly, and accept that the second-best product chosen quickly usually beats the best product chosen slowly.