Proving 5x more vacancies with the same team: a PoC for an international recruitment agency

A vacancy sourcing process stuck at manual capacity

Each of the client's five sales consultants searched job boards across four countries – Norway, Sweden, Germany, and the Netherlands – every morning by hand, pasting results into individual Excel files. Scaling that process meant solving several problems at once.

• Slow vacancy search. Consultants browsed ~30 sources this way, managing only ~120 vacancies a month against a target several times higher. And since speed decides the deal, every delay risks losing it to a competitor.
• No shared database. Each consultant worked from their own regional Excel file with no consolidated view of pipeline, vacancy types, or activity. The move toward an ERP would only widen the gap.
• Slow contact discovery. Finding the right decision-maker – owner, ops manager, or department HR – could take hours per company, even with tools like Apollo.
• Intuition-based qualification. The client only wants employers open to hiring international staff, but spotting them relied on the consultant's gut feel, with no repeatable rule behind it.

All this manual work and unstructured processes made the whole operation too slow and hard to scale. The client came to DigitalSuits for a way to fix the workflow and start winning more leads.

AI & Automation Blueprint to validate the idea

Our answer was an AI & Automation Blueprint, which is a fixed-price discovery phase we use to help clients map their real workflow, score what's worth automating, and price out the build. In this specific case, we wanted to prove the approach with a Proof of Concept on live data.

To get it right, we had to answer a few fundamental questions:

Q1: How does the workflow actually run?

The blueprint starts with the people doing the work. We sat with the sales managers and data researchers who ran the process daily and walked their real workflow end to end:

• what they actually do
• where the data lives
• which manual workarounds keep things moving
• where the hours quietly disappear.

That's how the gap between the process as described and the process as performed comes to light, and it's where the automation opportunities usually hide.

Q2: Which sources can we pull data from automatically, and how?

We sorted the ~30 boards across the four markets into three groups:

• National government job boards with vacancies pulled through free APIs
• Large job boards like Indeed and Glassdoor, reached through the TheirStack aggregator
• Niche sector boards, covered by custom scrapers

No single method fits all three, but together they replace the entire manual daily search.

Q3: How to unify years of scattered data?

The hard part was making sense of what each Excel file actually stored and turning it into one consistent schema.

We designed an approach for it:

• document what every column means
• turn those definitions into a clean database structure
• advise which fields are worth keeping versus historical clutter
• generate the migration itself with an SQL script that moves the data from the Excel files into the new database.

The result would be a single database with a user-friendly dashboard for sales managers. This single source of truth would be ready to receive new vacancies daily and to feed the client's new ERP (Odoo) or any future platform.

Q4: How to keep the data fresh and clean?

Three mechanisms would keep the database current and clean:

• Daily incremental pulls. Instead of re-crawling entire sites, the pipeline runs on a schedule and pulls only the last 24 hours of postings. This way, consultants get new vacancies while they're still fresh.
• Deduplication at two levels. Each record is checked against the database at both vacancy and company level, with an LLM confirming it's real and catching re-posts.
• A human in the loop. Everything the pipeline collects – vacancy, company, contacts – lands in an intermediate layer where a researcher approves or declines it before it flows onward into the ERP.

Q5: How to qualify vacancies and leads?

The LLM layer (Claude API) does two jobs:

1. Intent analysis. It reads each vacancy against a custom prompt and flags whether the employer is open to hiring candidates from abroad.
2. Contact AI validation. Once Apollo returns the people at that company, the LLM filters them by function title to keep only those responsible for that kind of hire.

Since contact enrichment is where GDPR risk sits, we gave the client guidance on handling it right, including favoring LinkedIn social selling over cold outreach.

Q6: What does it cost to run at scale?

We priced out every moving part, so the client could see the real running cost before funding a build:

• National government APIs: free, covering over 60% of total volume
• Custom scrapers: ~$150/month
• Contact enrichment (Apollo): from ~$60/month, scaling with volume
• Hosting + database (Vercel, Supabase): ~$115–155/month

All in, the pipeline runs at roughly $365/month in production.

Q7: How to prove it works?

We tested the riskiest assumptions with a Proof of Concept on one live source – a Dutch agricultural board central to the client's sector.

• Data freshness. Using the designed architecture, the pipeline pulls only the last 24 hours of postings without a full re-crawl.
• Contact discovery. Apollo returns manager-level people even for small Dutch farms.
• Deduplication. Claude verifies each posting is real, normalizes fields, and catches reworded vacancy re-posts.

All three held, clearing the way to scale the same architecture across the remaining sources.

Put together, the full pipeline looks like this:

Automation model to scale sourcing with the same team

Within three weeks, the client got an automation model validated end-to-end on a live source. It covers the whole flow, from daily vacancy collection to qualified, contact-enriched leads, with a person kept in the loop as a quality gate.

Here's what the model is built to deliver:

~120 → ~600 vacancies/month

Roughly 5× the output, without scaling the team.

~30 sources → one daily flow

Every board merged into a single automated pipeline.

$365/month to run

Full operating cost for automation at scale.