One screening agent: every applicant assessed, nobody auto-rejected. International healthcare recruiting. Around 1,100 candidates a month arrive through Gravity Forms into HubSpot — 870 with a CV, 230 without. Three front-door gates stop roughly 20% today, all checked by hand. The agreed agent assesses the person (the job is context only), writes a match / possible match / not a match rating back to HubSpot, and takes over the no-CV chase end to end.
Acacium Group · introduced by Jamie Daykin, alongside Favorite · Demo agreed 18 Aug — being built
The numbers on the table
Candidates ~1,100 / month With CV ~870 · without CV ~230 Stops at the gates today ~20%, checked manually Chase today ~1,380 emails + 690 calls + 690 SMS/mo Instance Beam US · trigger from HubSpot, same pattern as Favorite POC timeframe 3 weeks (Jenna's answer) Where it stands
13 Aug — Jenna returned the scoping form: volumes, current chase process, 3-week POC window 18 Aug — Scoping session (Sam, Tom, Lauren on): flow agreed node by node, recap + deck sent same day 25 Aug — Espirita internal meeting on the data points; demo held until it lands 26 Aug — Jenna's mapping document received for review 27 Aug — HubSpot access granted to Sara and Mujtaba; connection testing with Craig begins The agreed flow — the menu's own modules, configured for Espirita Same modules, same steps as the menu — nothing custom-built. What changes per client is the yellow notes and the rulebook under each frame.
Intake — paid once$0.13 Trigger A new candidate lands in the mailbox; the run starts. 📌 Espirita: Fires from the HubSpot workflow — the pipeline itself is untouched.
Read the CV file The attached CV is opened and read — charged only when a file arrives. 📌 Espirita: 870 of ~1,100 arrive with a CV; the 230 without skip ahead to the chase.
Extract the facts Name, history and details become fields every module shares. 📌 Espirita: The 10 data points — final list from Jenna's mapping document (received 26 Aug).
What each step reads and writes Step Reads Writes Trigger application email in the mailbox · candidate record ID · candidate phone number · resume-attached yes/no flag new run for the candidate · steady conversation thread number (same candidate, same thread) · trigger note: with / without resume Read the CV file CV attachment (PDF) · page count · with / without resume flag full CV text for every later step · note of file quality (well- or poorly-formatted) Extract the facts full CV text · client's list of wanted details · banned-details list (age, gender, etc.) fact sheet: name, contact, location · work history: employer, title, duration, industry · licences & certifications with status and expiry · education & total years of experience · list of details not found
Entry Gravity Forms → Zapier → HubSpot — unchanged; a HubSpot workflow triggers the run (same pattern as Favorite) Instance Beam US Screen a candidate$0.375 Fetch the role The job requirements are pulled from the config table. 📌 Espirita: The job applied to is context only — Espirita assesses the individual.
Fetch the record The candidate's record comes up alongside.
Read the CV The CV is read in full, not skimmed.
◆ Condition · Front-door gatesResidency — not on the WHO safeguard list (~50 excluded countries)Occupation — a role Espirita actually recruitsEducation & experience — meets the per-profession minimumPASS ➜ All three pass — the sentence-by-sentence match runs next
FAIL ➜ Any gate fails — written back as 'not a match' with the reason; a recruiter still sees them. ~20% stop here today, all checked by hand
routing · not billed Compare and score Candidate meets criteria; every score comes with its reason. 📌 Espirita: Role-level must-haves per specialism — e.g. mental-health nurse vs clinical psychologist.
Compose the deliverable A recruiter brief, every sentence carrying its evidence.
Write the outcome Score and verdict are saved back to the system. 📌 Espirita: Writes the RAG value; tiers and who-to-call-first stay in HubSpot's hands.
What each step reads and writes Step Reads Writes Fetch the role role reference on the application · screening configuration table must-have checklist for the role · nice-to-have checklist · job title, location, facility type Fetch the record candidate name & record ID · recruiting-system record record with status & applied date · earlier-applications check result Read the CV the CV, every page · the role checklist alongside CV facts ready to compare: employers, titles, durations · licences & certificates noted Compare and score CV facts · must-have & nice-to-have checklists · licence type / state / status · years of experience vs the minimum per-requirement verdict: match / partial / gap / unconfirmed, each with its evidence · scores: licence, experience, certifications, overall · fit label: strong / good / low / alternative / limited · key strengths & key gaps Compose the deliverable scores & fit label with their reasons · matched strengths & open gaps · candidate facts & job requirements recruiter brief — 2-minute read, under 2,000 characters · phone-screen questions: 3 competencies + 4 role-specific · housekeeping notes: notice period, salary, availability Write the outcome job opening match ID · candidate record link · must-have & nice-to-have results · summary & recommendation record row saved: scores, summary, recommendation, source, task ID · status moved to screened
Matching Sentence by sentence against must-haves per specialism, visa and country rules layered on — no arithmetic scoring Write-back RAG rating only: match / possible match / not a match. HubSpot workflows own tiering, weighting and notifications Compliance Aligned with NYC Local Law 144 — no automated rejections, no inferences from protected characteristics, full reasoning and audit logs Engage a candidate$0.325 Pick the moment Decides who to contact and when it is worth it.
? Draft the message A short, precise question: are you still available?
Send The message goes out on the channel you chose.
Read the reply The candidate's answer is understood, whatever the wording. 📌 Espirita: Emailed attachments are processed directly — the manual upload step disappears.
What each step reads and writes Step Reads Writes Pick the moment screening score & open questions · time since the application arrived · candidate mobile number contact-now / wait decision · chosen channel and timing Draft the message open questions to confirm · role's licence & facility requirements · 160-character limit · message position in the sequence (1 of 3) short text asking to confirm requirements & availability · candidate summary noting what is still unknown Send the drafted text · candidate phone number · conversation thread number sent text on the chosen channel · delivery timestamp on the thread Read the reply candidate's reply text · the messages sent so far · earlier replies on the thread reply type: confirm / correction / question / decline / opt-out · confirmed fields: licence, experience, facility · hours per week & start window · corrections, e.g. licence state
Today Only the first chase email is automated; returned CVs are uploaded by hand — ~1,380 chase emails, 690 calls, 690 SMS a month Agreed The agent takes the chase end to end and logs every touchpoint on the timeline Cadence on the table Day 0 → reminder +2 days → final +3 days → no-response status (awaiting their preference) Channels Email first; SMS / WhatsApp via Dotdigital once set up — US & Canada by SMS, everywhere else WhatsApp, decided by the country field. Genesys is out of scope Qualify from the reply$0.225 Extract what was said The reply's claims are pulled out one by one.
Check against the CV Claims are checked against the CV — the CV wins conflicts. 📌 Espirita: When there is no CV yet, the answers are the record.
Set the outcome Qualified or not, the status is set in the record.
What each step reads and writes Step Reads Writes Extract what was said reply text plus the questions that were asked · full text-message history claim list, one line each · each claim with its exact quote · list of questions still unanswered Check against the CV claims from the reply · CV facts — the source of truth · earlier summaries & briefs contradiction list with quotes from both sides · new details not on the CV, marked as such · one combined evaluation summary Set the outcome validated claims & updated must-have count · record field length limit (2,000 characters) qualified / not qualified status on the record · compressed summary written back
Lauren's path Where the file isn't strictly needed, ask for the data instead — a short question flow / HubSpot form link fills the missing fields Result The candidate is assessed and RAG-rated without a CV; the physical file follows before client interviews ✋ Recruiters review everyone
No auto-rejects — every candidate reaches a recruiter; tiers and notifications drive who to call first Strong candidates on the 'wrong' job surface as alternative fits The previous Red/Amber/Green process is re-enabled in the background for testing (Jenna) — not surfaced to resourcers yet What it would cost at your own volumes Path Modules Per candidate Volume Monthly With a CV intake + screening intake $0.13 + screen a candidate $0.375 $0.505 870/mo $439 Without a CV intake + chase + qualify + the same screening once the data is in intake $0.13 + engage a candidate $0.325 + qualify from the reply $0.225 + screen a candidate $0.375 $1.05 230/mo $243 Both paths, at the volumes from the scoping form $682/mo
Standard model tier, menu rates. The two calibrated anchors on this exact pattern are Favorite's measured bills: screening $0.40 and qualification $0.25 per candidate, both exact menu matches. Channel costs (Dotdigital SMS / WhatsApp) come on top once chosen — never silently added.
Open questions
Mailbox & sending domain — Outlook or Gmail, which unified address Chase cadence & stop rule — proposal on the table: day 0, +2 days, +3 days, then no-response status The 10 data points — final list (internal meeting held 25 Aug; Jenna's mapping document received 26 Aug) Dotdigital — SMS package cost, WhatsApp capability, setup timeline (Lauren & Thomas, building on Ian's inquiry) Where this page comes from
Email: 'Espirita Screening Agent — Aug 18 Recap, Confirmed Points & What We Need for the Demo' (thread 1a015daf87f25c09) Email: 'Espirita Use Case Scoping Form for Beam AI' — Jenna O'Neill's answers, 13 Aug (thread 19fdb5683dbe5ff7) Email: 'Favorite / Acacium <> Beam AI Sync Summary — August 25, 2026' (thread 1a0394adac85ca09)