← Help centre

AI assessments, fit scores, flags and scorecards

Signed in? Ask Lovelio this question inside the app - it answers from this same page.

Lovelio automatically assesses every candidate who lands on a job. The AI reads the CV, the job, the job's criteria and the client's Client DNA, then produces a Job Fit score, a Client Fit score (when Client DNA exists), a short plain-English summary, and a Yes / Maybe / No verdict that pre-sorts the candidate for the consultant to confirm. The AI never takes a consequential action on its own: it suggests, a human decides.

How an assessment runs

  • Assessment fires automatically whenever an application is created: a candidate applies through the careers page or widget, a consultant adds them to a job, or a bulk "Assess against a role" run adds a shortlist to a job. All of these use the same single assessment path.
  • The must-have check runs first, and it is free. Before an inbound applicant is assessed, Lovelio checks them against the job's hard requirements (the must-haves that come from the job's structured requirements, the same ones a search uses). An applicant who misses a must-have is not assessed at all. Their row shows "Not assessed" instead of a verdict, and hovering the chip says which must-have they missed, in the job's own words (for example "Does not meet the job's must-have: Location: Sydney."). The deal file says the same. They stay in the Funnel, nothing is rejected, and the consultant marks them Yes, No or Maybe like anyone else.
  • The must-have check only applies to people who applied themselves, and only when the job actually has must-haves. A job with no structured requirements assesses every applicant. Candidates a consultant added by hand, from a pool, an import, the extension or a bulk "Assess against a role" run are always assessed in full: the consultant chose them on purpose. Asking for an assessment by hand ("Assess now" or "Re-assess" on the candidate's Assessment tab, or asking Lovelio to assess a candidate) always runs the full assessment, whatever the check would have said, and a verdict then replaces the "Not assessed" chip.
  • The check is a plain rule, not a judgment. Everyone who passes it gets the full assessment, unchanged.
  • Assessments only run against Active jobs. A job that is Closed or Filled has nothing left to fill, so nothing on it is ever assessed or re-assessed - applications on closed jobs are history, and an unassessed historical application never shows an "Assessing" spinner. Reopen the job to assess against it again.
  • The AI scores every criterion defined on the job. Criteria come in two kinds: "Essential" criteria are pass or fail, and scale criteria score 1 to 5 (3 means meets expectations). Each criterion carries a weight up to 3.
  • Job Fit (0-100) is the weighted average of the criterion scores. An Essential pass counts as 100, a fail as 0; scale scores multiply by 20.
  • Client Fit (0-100) measures how well the candidate matches the client's Client DNA (the hiring culture captured for the job's client). It only exists when that client has a completed Client DNA. If there is no Client DNA, the score is simply absent. Lovelio never invents a Client Fit number.
  • Client Fit is only scored once Job Fit clears the reject line. A candidate whose Job Fit lands at or below the company's Job Fit reject threshold (from the same Settings block as the verdict thresholds) is a No on Job Fit alone, so Lovelio does not spend a Client DNA comparison on them at arrival. Their Client Fit shows "Not scored - job fit below the review line" and fills in automatically the moment a consultant opens their assessment. The same applies to their summary: candidates above the line get their summary written immediately; below the line it is written on first open, in a few seconds.
  • Every criterion score comes with a one-sentence AI reasoning, plus a summary of the candidate's fit written for the consultant, and a recommendation badge: "Advance", "Consider", "Cafe" (the AI thinks this person suits a talent pool rather than this role - it appears when Client Fit is strong but Job Fit is a clear miss) or "Reject". On a client job the assessment judges fit for the client, never fit for the agency. The one exception is the agency's own internal hires, where the agency is the employer.
  • Each assessment carries a confidence score (0-100) with the factors listed: whether a CV was parsed, skills listed, a cover letter provided, criteria defined, a job description present, Client DNA captured, and interviews on record. Thin data means lower confidence, and the factors say exactly what is missing.
  • Re-assessing runs a fresh version. On the candidate's Assessment tab the button reads "Assess now" when no assessment exists yet and "Re-assess" once one does. Previous versions are kept as history and the version number increments.
  • While a candidate is being assessed, the row shows an "Assessing" state and its actions are locked. If scoring ever gets stuck (an AI outage, for example), an hourly background sweep finds stuck rows and retries them automatically; anything stuck past 24 hours is escalated internally. Nothing needs to be done manually.
  • On a busy job the row may first show "Reading CV" instead. The first ten applicants on each job have their CV read the moment they apply. Every applicant after that on the same job is read in the half-price batch lane, so their row appears straight away with name, contact details and the CV file, marked "Reading CV", and the parsed profile, summary and assessment fill in when the batch lands, usually within minutes and always within the hour. If the batch is ever late, the hourly sweep reads the CV in line so nobody waits past the hour. Adding a candidate to a job yourself, uploading a CV while looking at a candidate, or pressing re-parse always reads the CV immediately. Pressing "Assess now" on an applicant still marked "Reading CV" reads the CV straight away and then assesses, so nobody waits on the batch when a consultant wants an answer now.

The verdict and what happens next

The two scores are compared against thresholds to produce a verdict:

  • Yes: both Job Fit and Client Fit clear their advance bars.
  • No: both scores at or below their reject bars.
  • Maybe: everything else.

One strong score alone can never produce a Yes - a great Job Fit with a poor Client Fit lands in Maybe. The verdict chip on the job grid shows the Job Fit alone, so hover it: the "Lovelio says" card leads with which score held the verdict and the bar it needed ("Job fit 92 clears the bar. Client fit 54 does not: a Yes needs 60 or more."), and the deal file's Review card says the same. When there is no Client Fit score, the verdict is decided on Job Fit alone, and can only be No (a clear miss) or Maybe, never an automatic Yes.

Thresholds live in Settings, under the Hiring group, on the Workflow tab, on the "How strict is the AI?" card. It is one dial with three positions. Generous: Yes at Job Fit 65 and Client Fit 50, No at 40 or lower. Balanced, the default: Yes at 75 and 60, No at 50 or lower. Strict: Yes at 85 and 70, No at 60 or lower. Each position also sets the hot candidate bar (75, 80 or 90 on both scores) and the phone screen bars. "View / edit the numbers" under the dial edits any one threshold on its own; the dial then shows Custom until you pick a position again, which puts every number back. A reject bar always stays below its advance bar.

The verdict is a suggestion only - it never moves a candidate. The candidate stays in the job's Funnel wearing the suggestion chip (with the score) until a consultant marks them Yes, No or Maybe. No outward-facing action, including any rejection email, ever fires until a human decides. This human gate is deliberate and permanent.

Two automatic follow-ons run after scoring: if the job has AI voice screening switched on, promising inbound applicants (verdict Yes or Maybe, applied themselves) get the screen invite automatically - consultant-added candidates are never auto-emailed and re-scores never re-send. And every talent pool with matching switched on evaluates the candidate; re-scores never re-add or re-notify.

AI flags stay scarce and always explain

  • Only genuine grey-zone candidates (verdict Maybe) who applied themselves get flagged for the consultant's call, at normal priority, once per application.
  • A clear No is never flagged - it lands in the job's No column for the human to confirm the rejection.
  • A clear Yes on an inbound applicant is recorded as a strong match on the review queue (it reaches you in the daily Slack digest), not a flag; the Yes suggestion chip sits on their row in the job's Funnel.
  • Every flagged row carries the real reason: the assessment summary and scores when they exist, otherwise the exact reason written when the flag was raised (for example a failed assessment or flagged reference quotes). A bare "AI flagged this" with no explanation never appears.
  • Candidates a consultant added themselves are still scored and staged, but never generate attention flags or bell notifications - the consultant chose them moments ago. Inbound applicants do get a bell notification with their score and verdict.

Client fit score across the app

The Client Fit score follows the candidate beyond a single job. When ranking or answering "who are our strongest fits", Lovelio uses the best available Client Fit: a candidate-level fit assessment when one exists, otherwise the best Client Fit across the candidate's per-job assessments. Strong fit means Client Fit 75 or higher, and strong fits rank first.

Human scorecards

After an interview, humans submit scorecards: a score and note per criterion, overall feedback, and an overall recommendation. Confirming a scorecard requires the overall recommendation plus at least one criterion score. The recommendation flips the interview outcome automatically: strong hire or hire marks it Passed, no hire or strong no hire marks it Failed. Scorecards work from the dashboard, Slack, and the API, all through the same path. If a job has no rankable scorecards, ranking views show empty rather than made-up numbers.

Limits and edge cases

  • Bulk CV imports are matched to open roles by similarity only - they are never AI-assessed and never create applications. Assessing thousands of imports against every role would be slow, expensive noise.
  • The AI observes and reports facts. Summaries, reasons, and flags state what is in the data, never career advice or instructions to the consultant.
  • No Client DNA on the job's client means no Client Fit score and a lower confidence score. Capture Client DNA for that client (notes on the client record's DNA tab): the moment it completes, Lovelio automatically re-assesses everyone on that client's active jobs who is missing a Client Fit score. Only the missing scores are filled in - candidates who already have a Client Fit are never re-scored by this, and candidates in No or on closed jobs are left alone. No clicking Re-assess candidate by candidate.

Common questions

  • Why is there no Client Fit score? Two possible reasons. Either the job's client has no completed Client DNA - add DNA notes to the client; once the DNA completes, missing Client Fit scores on that client's active jobs fill in automatically (within about 15 minutes at most). Or the candidate's Job Fit landed at or below the reject line, so Client Fit was not scored at arrival - open their assessment and it fills in on the spot.
  • A candidate has been stuck on "Assessing" for a while. Stuck scoring is retried automatically every hour. If it persists past 24 hours it is escalated internally.
  • A row says "Not assessed". Why? The applicant missed one of the job's must-haves, so the assessment was not run. Hover the chip (or open the deal file) to see which must-have. If you want the full assessment anyway, open the candidate's Assessment tab and click "Assess now".
  • Why was a candidate I added not checked against the must-haves? The check only applies to people who applied themselves. Anyone a consultant adds is assessed in full.
  • Can the AI reject someone? No. It sets a provisional No that a consultant must confirm before anything happens, and rejection emails only send after human confirmation.
  • Can I change how strict the scoring is? Yes. Settings, Hiring group, Workflow tab, the "How strict is the AI?" dial: Generous, Balanced or Strict, or open "View / edit the numbers" and set any threshold yourself.
  • Why did a mediocre candidate not get flagged for a call? Flags are reserved for genuine Maybe verdicts on inbound applicants. Clear misses go straight to the No column for confirmation instead.
  • What does the "Cafe" badge mean? The AI thinks the person fits the client's culture but not this role, and suggests the talent pool instead. Nothing is added automatically off the badge alone.