The Hiring Problem No One Talks About
Small businesses have a specific hiring problem that enterprise companies solved years ago with dedicated ATS software: when you post a job and receive 80 applications, someone has to manually read every resume to find the 5 candidates worth interviewing. At 4–5 minutes per resume, that is 5–7 hours of reading before you have spoken to a single candidate. For a business owner who is also handling sales, client work, and operations — that time simply does not exist.
The typical result is rushed skimming, which means good candidates get missed and mediocre candidates occasionally slip through based on superficial impressions. Or you hire a recruiter for a flat fee — which works, but costs $3,000–$8,000 per hire for a $40,000–$60,000 role.
The system I built for this client gives small businesses a third option: AI-powered resume screening that reads every application thoroughly, scores candidates against a defined rubric, and surfaces the top applicants in a ranked Google Sheet — with a Slack notification the moment a high-scoring candidate appears.
How the System Works
The Application Intake
Applications come in through a job application form — a dedicated page on the company website, a Typeform, or a Google Form. The form collects the candidate's name, email, the role they are applying for, and their resume as a PDF upload.
When a new application is submitted, the form webhook fires an n8n workflow. The first step is extracting the resume text from the PDF. n8n passes the PDF file to a text extraction API, which returns the full plain text content — stripping all formatting so Claude can read it cleanly regardless of whether the resume was designed in Canva, Word, or Google Docs.
The AI Scoring Step
The extracted resume text plus the job description (stored in the workflow configuration) are passed to Claude with a structured scoring prompt. Claude evaluates each candidate across five dimensions:
- Role fit (0–10): How closely does the candidate's experience match this specific role? Years of relevant experience, industry background, scope of past responsibilities.
- Skills match (0–10): Does the candidate have the specific technical or functional skills listed as required versus nice-to-have in the job description?
- Career trajectory (0–10): Is the candidate's career moving in a direction that makes sense for this role? Are they appropriately matched in seniority — not overqualified or underprepared?
- Outcome quality (0–5): Does the resume describe measurable results and specific achievements, or only duties and responsibilities? Candidates who quantify their impact tend to perform better.
- Overall recommendation: Strong Yes / Yes / Maybe / No, with a one-paragraph rationale explaining the score
Claude returns structured JSON with all five scores, a total out of 35, and the recommendation paragraph. The scoring takes 8–10 seconds per application. For a batch of 73 applications, the full screening completes in about 12 minutes.
CRM Logging — The Ranked Google Sheet
After scoring, n8n writes a new row to the hiring tracking Google Sheet: candidate name, email, role applied for, application date, all five dimension scores, total score, overall recommendation, and the AI's brief rationale for the score.
The sheet is sorted by total score descending, so the hiring manager can open it at any point and immediately see the top candidates ranked — without opening a single PDF. The AI summary for each candidate provides enough context to understand the ranking without needing to return to the original resume.
Slack Alerts for High-Scoring Candidates
For any candidate scoring above a configurable threshold (typically 28/35 or higher), n8n sends an immediate Slack notification — not batched at end of day, but within seconds of the application coming in. The notification includes: candidate name, total score, recommendation, the AI's rationale, and a direct link to their row in the tracking sheet.
This matters for competitive roles. A genuinely strong candidate who applies on a Tuesday afternoon can receive an interview invitation before they have finished submitting applications to other companies. For roles where good candidates are evaluating multiple offers, that speed is a real differentiator.
What the Output Actually Looks Like
After a typical hiring cycle with 60–80 applications, the ranked sheet shows a clear structure:
- 5–8 candidates in the Strong Yes band with total scores above 28 — these are the first-round interviews
- 12–18 candidates in the Yes band worth considering if the first group does not work out
- The remaining candidates clearly below threshold, each with the AI's rationale visible in their row
The hiring manager filters, sorts, and reviews the sheet without opening a single PDF. If they want to look at a specific candidate's resume, the PDF link is in the sheet. But they are now working from an annotated shortlist rather than an unsorted pile.
Technical Components
- n8n — orchestration, scheduling, all logic and routing
- Claude via Anthropic API — structured multi-dimensional scoring and recommendation generation
- PDF text extraction API — called via HTTP Request node, returns clean plain text from any PDF
- Google Sheets — application tracking and ranked candidate list
- Slack — real-time alerts for candidates above the score threshold
- Typeform or Google Forms — intake form with PDF upload support
Calibrating the Scoring Prompt
The most important step in setting up this system is writing the scoring prompt — specifically, making sure Claude understands what makes a strong candidate for this specific role at this specific company. A generic prompt that only looks at years of experience and keyword matching produces mediocre rankings.
For this client, the calibration process involved reviewing 10 past candidates together — 5 who were hired and worked out well, 5 who were rejected or who turned out to be poor fits — and adjusting the prompt until the AI's scores aligned with the known outcomes. After 3–4 iterations, the prompt was reliably surfacing the characteristics that predicted success in that specific role. This calibration session takes 2–3 hours but dramatically improves ranking quality.
What This System Does Not Replace
Resume screening surfaces candidates. It does not replace interviews or the final hiring decision. The AI evaluates applications against a defined rubric — it cannot assess communication style in conversation, personality fit, or cultural alignment. Those judgments still require human evaluation.
What the system eliminates is the bottleneck that comes before any of that: the hours of initial manual reading that most small business owners cannot actually afford. The AI reads every application thoroughly. You interview the top candidates. That division of labor works.
Results from the First Hiring Cycle
- Applications received: 73 over 3 weeks
- Owner time spent on initial screening: 45 minutes reviewing the top 8 AI-ranked candidates in the sheet, versus an estimated 6+ hours to read all 73 resumes manually
- Candidates invited to first-round interview: 6 — all from the AI's top 8
- Hire made: Yes — from the AI's top 3 ranked candidates
- Score of the hired candidate: 31/35 — the AI's top-ranked applicant for this role
Who This System Is Built For
This works best for small businesses that hire infrequently — 2–6 times per year — but receive enough applications per role to make manual screening genuinely painful. It is particularly useful when the role has defined requirements that can be clearly articulated and when moving quickly on strong candidates is a real competitive advantage.
It is less useful for highly specialized roles where AI cannot reliably assess domain competence from a resume alone, or for businesses hiring at very high volume where a dedicated ATS with built-in screening makes more economic sense.
Want This Built for Your Business?
If your next hiring cycle is going to cost you a week of reading resumes, an AI screening system like this can reclaim most of that time while ensuring no strong candidate gets missed because you were moving too fast. Book a free consultation and we can discuss whether this makes sense for your team size, hiring frequency, and the roles you typically fill.