The Challenge
Horizon Staffing Group operates from its headquarters in Dubai with branch offices in Riyadh, Doha, Singapore, Kuala Lumpur, and Mumbai. The agency specializes in mid-to-senior level placements across construction, oil and gas, healthcare, IT, and financial services — industries that pull talent from every corner of the world into the Gulf and Southeast Asian markets. On any given week, Horizon's 84 recruiters are working roughly 600 active job orders from 140 corporate clients who expect shortlists within 72 hours of briefing.
The math was punishing. Each job order attracted between 50 and 200 applications, depending on the role and the market. Across 600 active orders, that meant the agency was receiving somewhere between 30,000 and 40,000 resumes every month. These were not neatly formatted documents from a single country or language. A single job order for a project manager role at a construction firm in Qatar might attract applications from Indian engineers writing in English with Hindi-script certifications attached, Filipino candidates submitting resumes in mixed Tagalog and English, Jordanian applicants with Arabic-language CVs, and Pakistani professionals whose documents blended Urdu headers with English body text. The file formats were equally chaotic: PDFs created from Word, PDFs created from scanned paper, Word documents in both .doc and .docx, image files of handwritten CVs photographed on phones, and LinkedIn profile exports.
Horizon's recruiters were spending the first four to five hours of every working day doing nothing but reading resumes. Not evaluating candidates against job requirements. Not having conversations with hiring managers. Not negotiating offers. Reading. Scrolling through documents, mentally extracting names, nationalities, visa statuses, years of experience, certifications, language proficiencies, and salary expectations, then manually entering that information into the agency's database so it could be searched later. A senior recruiter estimated that she could properly process — read, extract, enter, and tag — about 12 resumes per hour. A junior recruiter managed eight. Across the team of 84, the agency was burning roughly 4,200 recruiter-hours per month on manual resume processing. That was 60% of total recruiter capacity consumed before a single candidate was contacted.
The downstream consequences were measurable. Horizon tracked time-to-shortlist — the number of hours between receiving a job order and delivering the first batch of qualified candidates to the client. The agency average sat at 56 hours, against a client expectation of 48. For roles requiring niche qualifications like specific safety certifications or bilingual fluency in uncommon language pairs, the number climbed past 80 hours. Clients were noticing. In the six months before Horizon contacted Workisy, the agency had lost four accounts representing $1.8 million in annual revenue. In each case, the departing client cited the same reason: competing agencies were delivering shortlists faster.
There was a deeper problem buried inside the manual workflow that the leadership team did not fully understand until they audited it. When recruiters manually processed resumes, they made judgment calls about what information to record and what to skip. Under time pressure, those calls became shortcuts. A recruiter processing her fortieth resume of the morning might note a candidate's job title and years of experience but skip the certifications section because it was written in Arabic and she could not read it. Another recruiter might enter "5 years construction experience" without capturing that the candidate had specific experience with modular building techniques — a detail that would have made him the top match for a job order that came in the following week. The agency's candidate database, which should have been its most valuable asset, was full of incomplete and inconsistently tagged records. When recruiters searched for candidates, they found some of the right people some of the time. The database was a partial map of the talent they actually had access to.
Horizon's CEO, Faisal Al-Rashidi, framed the situation in a board meeting: the agency was sitting on a reservoir of talent data and extracting value from only a fraction of it because every resume had to pass through a human bottleneck that could not scale.
The Solution
Horizon selected Workisy's applicant tracking system after evaluating the platform's AI resume parsing engine against three competing staffing agency software products. The deciding factor was not parsing speed — every vendor claimed fast processing. It was parsing depth across languages and formats. Horizon's test involved submitting a batch of 500 resumes that the agency had collected specifically because they were difficult: scanned Arabic CVs with mixed layouts, Filipino resumes with non-standard section headers, Indian CVs running eight pages with project-by-project breakdowns, and image-based documents captured from WhatsApp screenshots that candidates had sent directly to recruiters' phones. The other platforms returned structured data from 60% to 70% of the test batch. Workisy returned structured, usable data from 93%.
Workisy's parsing engine was configured for Horizon's specific operational needs. The system was set up to extract 34 distinct data fields from each resume: the standard fields like name, contact information, job titles, employers, dates, education, and skills, plus fields critical to Horizon's market — nationality, visa status, language proficiencies with self-assessed fluency levels, professional license numbers, safety certification types and expiry dates, willingness to relocate, and current and expected salary in the candidate's stated currency. For the Gulf market, where a candidate's visa status and nationality directly determine which roles they are eligible for under local labor quotas, these were not optional nice-to-have fields. They were the first thing a recruiter checked before deciding whether a candidate could even be submitted to a client.
The AI parsing engine processed each resume through multiple stages. First, document normalization — converting whatever format arrived into machine-readable text, including OCR for scanned documents and images. Second, language detection and processing — identifying the primary and secondary languages in the document and applying the appropriate NLP models. The system handled English, Arabic, Hindi, Urdu, Tagalog, Malay, Tamil, and Mandarin, covering 98% of the languages Horizon encountered. Third, intelligent field extraction — using contextual understanding rather than rigid pattern matching to identify and extract data. The system recognized that a section labeled "المؤهلات العلمية" in an Arabic CV was the education section, that "Mga Kasanayan" in a Filipino resume meant skills, and that an eight-page Indian CV with a project table on page six contained the most relevant experience details for a technical role.
Fourth, and most valuable for Horizon's workflow, the engine performed automatic candidate-to-requirement matching. As each resume was parsed, the structured data was immediately scored against all active job orders in Horizon's system. A recruiter opening her dashboard in the morning did not see a pile of unprocessed resumes. She saw a ranked list of candidates already matched to her open roles, with match scores and the specific attributes driving each score. A candidate scoring 91% for a senior mechanical engineer role in Qatar would show the match breakdown: 12 years of relevant experience, valid UPDA engineering license, Arabic and English fluency, current location in the Gulf, and salary expectation within the client's range.
Workisy's Recruitment CRM module tied the parsed data into Horizon's client management workflow. When a recruiter built a shortlist for a client, the system auto-generated candidate profiles in the client's preferred format, pulling the structured data directly from the parsed resume. No more manually rewriting candidate summaries. No more copy-pasting between systems. The CRM also tracked which candidates had been submitted to which clients, preventing the embarrassment of double-submitting a candidate who had already been rejected by the same hiring manager three months earlier — a problem that had occurred often enough to damage Horizon's reputation with two key accounts.
The Implementation
Workisy's implementation team and Horizon's operations leadership mapped out a four-week deployment designed around the agency's reality: 84 recruiters across six offices in five time zones who could not stop processing resumes while the new system was being rolled out. The approach was parallel operation, not cold switchover.
During week one, Workisy ingested Horizon's existing candidate database — 410,000 records accumulated over seven years. The quality of these records varied wildly. Some had full structured data entered meticulously by senior recruiters. Many had only a name, a phone number, and an attached resume file that had never been properly parsed. Workisy's engine processed every attached resume file in the legacy database, extracting and structuring data that had been locked inside unread documents for years. This backfill operation alone surfaced 28,000 candidates with qualifications that matched currently active job orders but had never appeared in search results because their records were incomplete.
Week two focused on configuration and testing with a pilot group of 12 recruiters — two from each office — selected because they handled the highest resume volumes and the most linguistically diverse candidate pools. These recruiters ran the new system alongside their existing manual process, comparing the AI's extracted data against their own work. The pilot group reported back on accuracy, missing fields, and parsing failures. Workisy's team used this feedback to fine-tune the extraction models, particularly for Gulf-specific document formats like UAE residence visa copies that candidates often attached alongside their CVs and Saudi professional license certificates that used a layout the system had not initially recognized.
Week three was full-team training, conducted as live working sessions rather than classroom presentations. Trainers sat with recruiters during their actual morning resume processing sessions and walked them through the new workflow in real time. The shift for recruiters was significant: instead of opening email attachments and reading resumes one by one, they opened their Workisy dashboard and reviewed pre-parsed, pre-matched candidate cards. Their job changed from data entry to quality verification — confirming the AI's extraction was accurate and making judgment calls on borderline matches that the system flagged for human review.
Week four was full operation across all six offices. Workisy's customer success manager remained available in a shared communication channel with Horizon's team leads, responding to questions and edge cases within minutes. The most common issue during the first week of full operation was recruiter trust. Recruiters who had spent years building expertise in reading resumes quickly were initially skeptical that an automated system could match their accuracy. The turning point came when one of Horizon's most experienced recruiters in the Dubai office discovered that the AI had correctly extracted a candidate's PMP certification from an Arabic-language CV that the recruiter herself had processed manually two months earlier and missed entirely. The certification made the candidate eligible for a project management role that had been open for six weeks.
The Results
The numbers moved fast. Within the first 30 days of full operation, resume processing time per document dropped from an average of 5.2 minutes of recruiter time to 45 seconds of review time. The AI parsed and structured each resume in under 8 seconds. Recruiters spent the remaining 37 seconds verifying the extraction and confirming or adjusting the match scores. That 85% reduction in processing time translated directly into recovered capacity: roughly 3,570 recruiter-hours per month freed from data entry and returned to the work that actually generated revenue — candidate engagement, client relationship management, and placement execution.
Time-to-shortlist, the metric that had been costing Horizon accounts, dropped from 56 hours to 19 hours. For standard roles with large candidate pools, recruiters were delivering matched shortlists within 8 hours of receiving a job order. The speed improvement was not just about parsing. It was about the AI's ability to simultaneously match a newly parsed resume against every open job order and surface candidates that recruiters would not have found through manual keyword searches. A recruiter working a healthcare staffing order in Riyadh received a matched candidate notification for a nurse whose resume had been submitted for a completely different role in Singapore. The nurse's qualifications, licensing, and stated willingness to relocate to Saudi Arabia made her an ideal fit — a connection that no recruiter working in a single-office silo would have made.
Candidate-to-client match accuracy — measured by the percentage of submitted candidates who progressed past the client's initial screening to interview stage — improved from 68% to 94%. The improvement came from two sources: richer data extraction that captured qualifications recruiters had been skipping under time pressure, and the elimination of human inconsistency in how candidate records were tagged. When every resume was parsed against the same 34-field schema by the same AI engine, search results became reliable. A search for candidates with specific safety certifications actually returned all candidates with those certifications, not just the ones whose recruiter had happened to enter that detail.
Placement volume increased by 40% in the first quarter after deployment. Horizon placed 1,680 candidates in Q1 2026, compared to 1,200 in Q4 2025 and 1,140 in Q1 2025. The additional 480 placements were worth approximately $2.9 million in gross placement fees. The increase was not driven by more job orders or more applications. It was driven by the agency's ability to find and present the right candidates faster than before and faster than competing agencies.
The legacy database reprocessing delivered an unexpected windfall. Of the 28,000 candidates surfaced by parsing previously unstructured resume files in Horizon's seven-year archive, 3,200 were contacted for active roles. Of those, 440 were still in the job market and interested. Horizon placed 112 of them within three months — candidates who had been sitting in the agency's own database, invisible, because their resume data had never been properly extracted. Those 112 placements represented $680,000 in revenue generated from an asset the agency already owned.
The multilingual parsing capability opened a revenue line that had not existed before. Horizon began accepting job orders from clients who specifically needed candidates with uncommon language combinations — an Abu Dhabi hospital group seeking Tamil-speaking nurses with Arabic proficiency, a Singaporean logistics company needing Malay-Mandarin bilingual operations managers. Previously, Horizon's recruiters could not efficiently search for language proficiencies that were buried in non-English resume text. With the AI extracting and normalizing language data from every resume regardless of the document's language, these searches became instant.
What's Next
Horizon is expanding its use of Workisy in two directions. The first is predictive talent availability modeling. Using the structured data now flowing through every parsed resume, combined with historical placement data, Horizon plans to build forecasting dashboards that show which skill categories and geographies are trending toward talent shortages. The goal is to proactively build candidate pipelines before clients come to them with urgent orders, shifting the agency from reactive order fulfillment to consultative talent advisory.
The second initiative is integrating Workisy's AI parsing with Horizon's candidate sourcing channels. Currently, resumes arrive via email, job board applications, and walk-in submissions at branch offices. Horizon is piloting a WhatsApp-based application channel where candidates in labor-sending countries like India, the Philippines, and Pakistan can submit their resume by sending a photo or document through WhatsApp. The AI parsing engine will process these submissions in real time, immediately extracting structured data and matching the candidate against open roles. For a market where millions of candidates apply through mobile-first channels, this removes every barrier between a qualified candidate and a placement opportunity.
Client Quote
"We spent seven years building a database of 410,000 candidates and could only search a fraction of it properly because most of the data was trapped inside unstructured files that nobody had time to read. In the first week after Workisy processed our archive, we found 28,000 candidates we did not know we had. We placed 112 of them. That is revenue we generated from talent that was already in our system, invisible to us because we were doing everything by hand. The speed is what the clients notice — shortlists in hours instead of days. But for me, the real transformation is accuracy. When my recruiters search for a candidate with a specific certification or a specific language pair, they find everyone who qualifies. Not just the ones we remembered to tag correctly. Everyone. In a staffing business where the difference between winning and losing a client is whether you can find the right person before your competitor does, that changes everything."
Faisal Al-Rashidi, CEO, Horizon Staffing Group