Bolt AI screening onto your ATS if it has a working integration or open API, your candidate data is reasonably clean and your contract has months to run. Switch if the ATS can't write screening results back to candidate records, or its own AI tier costs nearly as much as a new system. Trial the bolt-on for 30 days first.
Most agencies underestimate what switching costs: moving candidate records, notes, consents and placement history, then a month or two of recruiters working more slowly in an unfamiliar system. The bolt-on route has its own hidden cost, which is running two places where candidate decisions are recorded. The right answer depends on which of those costs is smaller for you.
Three routes, not two
There are really three options, and the middle one often gets missed:
- Upgrade within your ATS. Many systems now sell AI in a higher tier. Bullhorn's small-agency pricing shows the pattern: Starter ($99 per user a month) and Core ($165) include no AI, while Pro adds AI search and match and Max adds agentic screening, both by quote. No integration work, one system of record, but you pay the tier price for every user.
- Bolt on a screening tool. A specialist tool reads applicants from your ATS, scores them against your criteria and writes results back. Brainner, for example, prices by volume of analyses (2,000 or 5,000 a month on its main plans, with overage at $9.95 per 100 extra candidates) and integrates with Greenhouse, Lever, Workable, Zoho Recruit, SmartRecruiters and several others. Workable's own screening agent works on credits: one per applicant screened.
- Switch to an ATS with AI built in. Some systems include AI scoring at every tier. Manatal includes AI recommendations and scoring from $15 per user a month billed annually ($35 on its Enterprise plan with unlimited jobs and candidates).
Notice what the bolt-on integration list tends to contain: in-house recruiting systems used by employers. Agency-focused systems are less often on the list. Check for your ATS by name before anything else, because "we integrate via API" can mean a supported connector or a developer project you pay for.
What bolting on looks like, before and after
Before (illustrative): a recruiter exports 180 applicants for a payroll role as a spreadsheet, uploads the CVs to a screening tool, reads the ranked list, then goes back into the ATS to move 25 candidates to "shortlisted" and reject the rest. Forty minutes of handling, and the scores live only in the screening tool.
After, with a working integration: applications arrive in the ATS as usual. The screening tool picks them up automatically, scores each against the role's must-haves and nice-to-haves, and writes the score, a two-line reason and a suggested status back to the candidate record. The recruiter reviews the suggestions in the ATS and confirms. No export, and the score sits on the record where the next recruiter will see it.
If the integration can't write back, you're living in the "before" version permanently, just with better rankings.
What switching actually involves
- Candidate records and CVs: usually migrated, but check duplicates, custom fields and attachments.
- Notes and activity history: often the part that migrates worst. Years of call notes may arrive as one text blob per candidate, or not at all.
- Consent and marketing preferences: must come across intact. A candidate who opted out of job alerts and then gets them from your new system is a complaint waiting to happen.
- Placements, clients and commission history: essential for business development and reporting; check the new system handles agency placements properly, not just employers' hiring.
- Integrations: job boards, email, calendar, e-signature, timesheets and invoicing all need reconnecting.
- People: recruiters are slower for the first few weeks. Plan the switch away from your busiest hiring months.
Before any vendor quotes for migration, take a sample export from your current ATS and check five records by hand. A filled-in check from an illustrative agency:
| Record | CV attached? | Notes with dates? | Consent and opt-outs? | Placement linked? |
|---|---|---|---|---|
| Candidate placed in 2023 | Yes | One text block, no dates | Yes | No, placement in a separate file |
| Active candidate, 3 applications | Latest CV only | Yes | Yes | n/a |
| Candidate who opted out of alerts | Yes | Yes | Opt-out missing | n/a |
| Contractor, 6 assignments | Yes | One text block, no dates | Yes | 2 of 6 linked |
| Client contact | n/a | Yes | Yes | n/a |
Three problems in five records is a normal first result, and each one is a question for the new vendor's migration team: can dated notes be kept, can opt-outs be mapped to the new system's consent fields, and can placements be linked to candidates and clients? The missing opt-out is the one that can't wait. If it would have been lost in a real migration, the first job alert from the new system goes to someone who asked not to receive one.
Upgrade, bolt-on or switch, compared
| Criterion | Upgrade tier | Bolt-on tool | Switch ATS |
|---|---|---|---|
| Set-up effort | Low | Low to medium, if a connector exists | High |
| Systems of record | One | Two, unless results write back | One |
| Cost structure | Higher price for every user | Volume-based; pay for what you screen | New seats, migration, lost time |
| Quality of screening | Varies; test it | Often strongest, as it's the whole product | Varies; test it |
| Reversibility | Downgrade at renewal | Cancel monthly on many tools | Hard to undo |
| Best when | Most recruiters would use the AI | A few desks screen high volumes | The ATS is also failing in other ways |
Signs your ATS can't carry AI screening
- No connector for any screening tool you've shortlisted, and no documented API your IT provider can use.
- Screening results can only be viewed in the other tool, not written back to candidate records.
- Duplicate candidates are common and parsing is poor, so any AI works on bad records. (Test this first: how to evaluate the AI features in your recruitment software includes a 30-CV parsing test.)
- The vendor's AI tier is quote-only and the quote is close to the price of a whole new system.
- Recruiters already keep side spreadsheets because the ATS is slow or awkward.
Two or more of these and switching deserves a serious look. One on its own usually has a workaround.
A workaround worth pricing carefully is joining the two systems with an automation platform. It works, but screening happens per applicant, and automation platforms bill per action. Zapier's Professional plan covers 750 tasks a month and Team 2,000; an agency screening 20 roles of 250 applicants, with two actions per applicant, needs 10,000. When Zapier gets too expensive shows how to find the tipping point.
Pricing the three routes for a six-recruiter agency
An illustrative agency: six recruiters on Bullhorn Core, which is $990 a month for six users at list price. They fill about 20 roles a month, each with around 180 applicants. Screening takes a recruiter about three hours per role. They've tested a bolt-on that cuts that to about an hour including review.
Value of the time saved: 20 roles × 2 hours × $40 an hour = $1,600 a month.
Route 1, upgrade to Pro: the cost is the Pro quote minus $165, times six users. If the uplift is under about $260 per user a month ($1,600 spread across six users), the upgrade pays for itself on screening time alone, provided the built-in screening performs as well as the bolt-on did in testing.
Route 2, bolt-on: 20 roles × 180 applicants is 3,600 analyses a month, within a 5,000-analysis plan. The bolt-on pays if its monthly price, plus about 10 minutes of double handling per role (20 roles × 10 minutes × $40 ≈ $133), stays well under $1,600. Check the busy months too. If a seasonal peak brings 30 roles at the same volume, that's 5,400 analyses, 400 over the plan, which at $9.95 per 100 extra candidates is about $40 of overage. Small, but worth knowing before a recruiter pauses screening mid-month because they think the allowance has run out.
Route 3, switch: moving to a system at $35 per user would cut licences from $990 to $210, saving $780 a month. Against that, a one-off cost (illustrative): a migration service quote of $3,000, about 15 hours of slower working per recruiter in the first month (6 × 15 × $40 = $3,600) and 40 hours of admin at $30 ($1,200). That's $7,800, paid back in about ten months, before counting any agency features the cheaper system lacks.
The agency chose the bolt-on, with a review at Core renewal in nine months. That date sets its own timetable. With a 90-day notice period on the current contract (check yours), the decision to stay, upgrade or leave has to be made by month six, so the bolt-on trial must finish by month five at the latest, leaving a month to get migration quotes if the answer is "switch". The deciding factor wasn't the arithmetic, which was close; it was that the bolt-on could be cancelled monthly while the other two routes locked them in. Your numbers will differ, but set them out the same way: the value of time saved, then the full cost of each route including your own people's time.
Running a 30-day bolt-on trial
- Confirm the write-back before the trial starts. Send the ATS vendor and the screening vendor the same question, in writing.
- Pick five live roles across different desks, plus three closed roles where you know who was placed.
- Screen in parallel for the first two roles: recruiters shortlist as usual, then compare with the tool's ranking.
- Log every role in a simple sheet.
- Run fairness checks before any score is used to reject: paired copies of real CVs that differ only in name, age signals or a career gap should score alike. AI CV screening setup and safeguards for agencies covers the wider safeguards.
- Decide on day 30 against pass marks agreed on day 1.
A filled-in trial log (illustrative):
| Role | Applicants | Recruiter's shortlist in tool's top 25 | Time, before / with tool | Issues |
|---|---|---|---|---|
| Payroll administrator | 210 | 9 of 10 | 3 h / 1 h | None |
| Site manager | 140 | 7 of 8 | 2.5 h / 50 min | One score not written back |
| Field service engineer | 95 | 4 of 7 | 2 h / 1 h | Missed career changers |
| Estimator | 60 | 5 of 6 | 1.5 h / 40 min | None |
| Customer service lead | 320 | 11 of 12 | 4 h / 1.2 h | Rejection status synced before review |
The last issue is the one that matters. On the customer service role, the integration wrote "rejected" back to 40 candidates' records before a recruiter had reviewed them, and the ATS's automation sent those candidates a rejection email. One of them was a strong applicant the recruiter would have shortlisted. A configuration change solved it: the tool writes a suggested status to a custom field, and only a recruiter changes the real status. Find failures like this in a trial, not in month six.
The field service engineer line needed a different fix, in the criteria rather than the software. The role's must-have read "3 years as a field service engineer", so the tool scored down anyone who'd done the same work under another title. Rewritten as "3 years maintaining or repairing electro-mechanical equipment at customer sites, in any sector", with "field service engineer" moved to the nice-to-haves, a re-run on the same 95 applicants put 6 of the recruiter's 7 in the top 25 (illustrative). If a bolt-on underperforms on one desk, rewrite that desk's criteria and re-run before blaming the tool.
Duplicates are the other thing a trial exposes. If a candidate exists twice in the ATS, once with a 2022 CV and once with this year's, the bolt-on may score both and write two different results back. In one illustrative check of 200 applicants to a single role, 14 were duplicates, and in three cases the lower-scoring old record was the one the recruiter opened first. Merge duplicates for the trial roles before screening starts, and ask the screening vendor whether it can flag likely duplicates on its own.
Here's the question to send both vendors on day 1, filled in:
Subject: Screening integration: write-back and status changes
We use [ATS name] with 6 users and plan to trial [screening tool]
on 5 roles this month. Please confirm in writing:
1. Which fields the screening tool writes back to candidate records
(score, reasons, status), and whether each can be switched off.
2. Whether any write-back can trigger our ATS automations, such as
rejection emails. We need suggested statuses kept separate from
live statuses.
3. How candidate opt-outs and deletion requests in the ATS reach
the screening tool, and how quickly.
4. What happens to candidate data in the screening tool if we cancel.Then check the answer to question 3 yourself rather than taking it on trust. Create a clearly labelled test candidate, let the bolt-on screen them against a trial role, then delete the record in the ATS. A day later, search for the test candidate in the screening tool. If they're still there with a score and a CV, deletion requests don't reach the tool, and you'd have to delete by hand in two places every time a candidate asks. That's workable for a handful of requests a month but easy to forget, so either get the vendor to fix the sync or add it to your deletion procedure in writing.
Contract and data points to settle before either move
- Notice periods on your current ATS. Know the renewal date and notice period before you trial anything; a decision after the notice date can lock you in for another year. AI software contracts: auto-renewals, price rises and notice periods lists the clauses to check.
- Data export. Get a sample full export from your current system now, including notes and attachments. It's the best evidence of how hard switching would be. More in keeping your data portable.
- Candidate transparency. Update your privacy notice to say AI helps screen applications, and say how a candidate can ask for a person to review a decision. If you recruit for employers in the EU, AI used to screen candidates is classed as high-risk under the EU AI Act, with those obligations deferred to 2 December 2027; ask each vendor how they'll support you.
- Training on your data. Get it in writing whether candidate data is used to train the vendor's models, and switch it off if you can.
Further reads
- Is AI CV Screening Fair? Bias Checks Every Recruiter Should Run — The fairness checks to run whichever route you choose.
- How Recruiters Use AI to Write Candidate Summaries Clients Read — What to do with screening results once you have them.
- How Much Time Does AI Save a Recruitment Agency per Role? — Realistic time savings to plug into your own sums.
- AI Recruiting Tools for Small Businesses: What's Worth Paying For — The wider market of AI recruiting tools and prices.
- Questions to Ask an AI Vendor Before You Sign Anything — General questions before signing with any AI vendor.
- How to Rescue a Stalled AI Project or Exit It Cleanly — What to do if a bolt-on trial drifts without a decision.
- How Staffing Agencies Automate Candidate Follow-Up With AI — The seven moments where candidates go quiet, which to automate first, message templates, AI reply sorting and the consent rules that keep texts welcome.
- Can AI Screen CVs Fairly? What Small Employers Need to Know — How AI CV screening goes unfair, what the rules expect of small employers, and a six-step set-up with tests you can run in an afternoon.
- Questions to Ask a Recruitment Agency About Its AI Screening — Stage-by-stage questions on an agency's AI screening, how to grade the answers, and how a grooming salon found good candidates its agency's filter had rejected.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Bullhorn small agency pricing page; Manatal pricing page; Brainner pricing and integrations pages; Workable press release on AI agent credits (14 September 2026); Zapier pricing page.