A private university in Saudi Arabia that receives more than 12,000 applications annually for about 2,500 seats faced a challenge that seemed administrative but was actually commercial at its core. Applications came in on the website, by email, and in person. Transcripts needed to be reviewed by the admissions staff, test scores needed to be verified, the status of the attestation needed to be checked, the possibility of scholarships needed to be considered, offers needed to be issued, hundreds of students needed to be answered every day, and admitted students somehow needed to be kept warm long enough to turn them into actual enrollments.
The pace was too slow for the market they were in. Routine applications would still take approximately five to seven business days to process. In the high season, the answers can take up to two or three weeks. By that point, many excellent students had already accepted the competing offers. The university was also leaking money in a second way: The manual scholarship distribution in Excel generated arbitrary awards that occasionally exceeded the budget.
At aTeam Soft Solutions, we developed an AI agent to handle automated application intake, document extraction, eligibility checks, scholarship recommendations, offer generation, yield management, and applicant communication in both Arabic and English channels. The result was not only speedier admissions operations. The enrollment engine was tougher. Processing of applications fell to less than 48 hours in the majority of cases, the enrollment yield went up from 60 to 81 percent, scholarship management was brought under control, and the university gained close to SAR 12 million per year in tuition from students it would have been safer to bet it had lost.
At the start of the project, we spent time learning about the admissions office as a real operating environment, not merely as a workflow diagram. The result was a group engaged in a host of valuable activities but caught in a process that regarded admissions as an administrative gate, rather than a function critical to revenue.
The university provided undergraduate and postgraduate courses in business, engineering, IT, health sciences, and design. Tuition was the only source of revenue, so that meant every student who was admitted and enrolled was of real financial importance. Each seat equated to approximately 50,000 to 120,000 SAR of annual tuition value, based on the course. That is to say that all delays, every uncompleted follow-up, every slip-up in scholarships, and every lost offer-to-enrollment conversion could be calculated as a revenue consequence.
They had 15 staff members managing the entire cycle in the admissions office. They were liable for receiving applications; verifying eligibility; reviewing transcripts and test scores; checking for attestation status; evaluating merit scholarship applications; creating admission offers; and assisting the student through to confirmation and deposit. And on top of that, they were answering applicant inquiries on the phone and by email, on WhatsApp, and via social media.
The issue wasn’t low effort. It was about the volume and timing.
About 60% of the yearly applications were through the university’s web portal. Another 25% came by email with attachments. And the rest, 15 percent, were walk-ins that were received at the admissions counter. From January to March, throughout the high season, about 60 percent of the annual volume came in during just ten weeks. This required the system to process an increase of thousands of applications at a high speed and with high reliability while prospective students were engaged in active comparison shopping among universities.
A single application may appear simple until you consider the actual labor it involves.
Admissions staff had to review transcripts and assess if the GPA met the threshold for the program; look at standardized tests such as SAT, IELTS, TOEFL, Qiyas, etc., depending on the program; confirm that the school certificates were properly attested; make sure the student met any age, nationality, or ministry-imposed requirements; and so forth and so on before they could assess their chances of making it with a scholarship. A few of these were rules-based. Few of them were based on how many could be accommodated in the program, on internal scholarship policy, and on strategic priorities for enrollment.
Most of that work was done manually.
Transcripts were reviewed by the staff. Scores were cross-verified by staff. Staff identified any missing documents. Scholarship calculations were performed in Excel with a matrix model that factored both indicators of academic merit and financial need. Offer letters were produced once the team felt that the file was complete and that the student was suitable for admission. Then, the communication process commenced.
That’s where the university was losing actual value.
The admissions process was not over once an offer had been made. Actually, the most commercially sensitive phase started there. The university believes that somewhere in the region of 40 percent of students it admits are lost between offer and enrollment. A few of those students were almost certainly never going to turn up for lessons. But a lot of them were just going with speedier, more responsive rivals.
In the growing private higher education market in Saudi Arabia, speed of response is not preferred. It’s a competitive weapon. Students and their families are comparing universities at velocity. When one university gets back in 48 hours with a clean offer, clear scholarship data, and continuous communication, while another university takes a week or more to respond and then goes quiet, there is a pretty strong advantage for the first university.
The university’s own admissions numbers reflected the same problem. During the application season, they’d receive more than 500 inquiries a day on all platforms. The team was able to respond to maybe 200. The rest were left unanswered, or a response arrived too late to be of any use. Status questions were a particularly painful point. When an applicant asked, “What is the status of my application?” someone still had to look up the answer manually, since the old chatbot couldn’t tell anyone anything specific about their accounts.
There was also a hidden cost to the admissions office in managing scholarships.
Since the scholarships were being distributed manually through spreadsheets, mistakes were being made. A handful of students did end up with more aid than the policy dictated. Some programs are oversubscribing scholarship funds too early in the cycle. Other students who might have been persuaded to attend by a strategically determined award were not getting the right offer at the right time. The institution had previously seen scholarship overspending of SAR 1-2 million in the past years.
What was evident very early on at the company was that the university administration needed to stop considering admissions as a line-processing function. It had to be treated like a conversion engine. The university didn’t just have to process applications at a faster rate. It needed an AI agent to assess applications, to communicate promptly, to distribute scholarships more strategically, and to keep admitted students engaged until they really got in.
The university did have a Student Information System, but that system didn’t really exist until after the student was enrolled. It was not intended to assess applications, authenticate documents, handle incomplete submissions, or lead a structured communication track before enrollment.
They had also experimented with an admissions CRM. It helped track status at a high level, but it was not doing the real work. It can say that an applicant is “in review” or “offer sent”, but it cannot interpret transcripts, verify score requirements, identify missing documents, or calculate a scholarship recommendation based on policy and budget.
A chatbot had been deployed, too, but it was only good for very generic FAQ-type questions. It was able to answer questions such as “What programs do you have?” or “What is the tuition cost?” It would not answer “What is the status of my Computer Engineering application?” “Which document am I missing?” And those questions still have to be answered by humans.
That’s why the solution had to be more than a workflow tracker or a generic chatbot. The university wanted a genuine student enrollment AI agent that could read documents, assess admissions logic, communicate personally, and proactively support yield enhancement. This is to say that this was not simply admissions processing automation. It was a pricing and conversion problem in higher education.
We developed the system in stages as the university initially wanted confidence in application evaluation before it would trust the AI in scholarship and yield management. We began by addressing intake and eligibility, then progressed to offers and scholarships, then to enrollment support, and finally to high-volume inquiry management and analytics.
The initial phase focused on one of the greatest slowdowns: by the time an admissions officer even started evaluating an application, they had already spent too much time taking information and working out what was missing.
Upon receipt of an application through the portal, email, or physical counter workflow, the AI agent ingested the submitted documents and extracted the core data points such as applicant name, national ID or Iqama, transcript details, GPA, school, test scores, program selected, statement content, and some other structured information.
Then it went straight into completeness and eligibility checks.
Instead of a human taking days to realize a critical document was missing, the AI agent checked the submission to the program’s list of required files and sent a WhatsApp or SMS message telling the user what was missing. This was one of the easiest and most useful improvements in the whole process. Students no longer had to endure silent periods while their applications were unfinished. They were given exact directions almost immediately.
The system also verified if the applicant was above the minimum admissions thresholds for the program he or she was applying to, which includes GPA minimums, required test scores, attestation status, and any other program-based eligibility requirements. The applications were then classified into three general categories: obviously eligible, obviously ineligible, and review required. This did not eliminate human admissions discretion. It provided the team with a baseline to start from that was much more organized and drastically cut down on the time members spent wasting on easy cases.
The admissions officers could then look at the AI’s sorting, confirm or overturn where necessary, and devote more of their time to cases near the line or of particular importance.
This stage alone altered the pace of the office. An AI agent eliminated much of the repetitive triage work that was hindering the system during peak season.
Once the intake and eligibility layer was permanent, we shifted into the second crucial leverage point: that of offers and scholarships.
For the approved applicants, the AI agent created bilingual offer letters in Arabic and English. These involve the program information, conditions if applicable, tuition fees, scholarship amount when awarded, payment schedule, enrollment deadline, and further steps. That made the bids quick, transparent, and uniform.
The scholarship engine was one of the most financially crucial components of the project.
The university’s previous system was built around Excel, which left those scholarship decisions open to inconsistency and leakage from the budget. We programmed the AI agent to assess scholarship recommendations by academic merit, financial-need metrics, program priorities, available funds, and strategic enrollment targets. For example, if the university wanted to increase its engineering intake, the system could give percentage scholarship awards more aggressively to good engineering candidates without going outside the overall scholarship envelope.
We also implemented real-time budget guardrails. That let the university know how much of the scholarship pool was already spoken for by program, and whether its current award pattern could hold up for the rest of the cycle. If 80% of the budget is already committed when only 60% of the target offers have been sent, the system alerts leadership and forces the strategy conversation early instead of after the overrun that has already happened.
That substantially shifted the internal dialogue of the admissions office. The AI agent was now more than just an assistant to staff offering issuing. It was also running the business of the university, shaping its class economically.
And since the entire process was expedited, even talented, straightforward applicants found out from submission to offer in less than 48 hours. That in itself made the university seem a lot more professional and alert to the market.
The third phase is where the biggest financial impact comes into sight.
We didn’t stop at sending the offers. The AI agent tracked every admitted student’s path from enrollment offer and handled that path as a yield issue. Once an offer was accepted or even just opened, the system started handling the relationship.
Students were sent customized follow-up messages related to their specific program, rather than generic admissions emails. An admitted IT student received content about the first term, labs, projects, and IT-related student life. A health science student got a different sequence. The communication was via WhatsApp in Arabic or English according to the recipient’s preference, with a follow-up email.
The AI agent was also tracking these engagement signals: Did the student open the message, click on the payment or acceptance link, look at accommodation information, go to the curriculum page, or drop out? Students who seemed to be checked out were tagged as vanishing scenario cases. This allowed the admissions team to concentrate their personal outreach efforts on those students who were the highest value and most likely to go at risk of choosing elsewhere.
For certain cases, we also added peer ambassador matching. The system could, for example, suggest that a currently enrolled student in the same background or program reach out to an admitted student who appears to be uncertain. It gives the admissions process a more real feel without making the admissions team have to manually manage every interaction.
This stage reframed the admissions office internally. It wasn’t only reviewing applications and making offers. It was actively working on the conversion rate. The AI agent evolved into a structured yield-management system.
For us at aTeam Soft Solutions, that’s where the project clearly changed from “automation” to university enrollment management AI. The key important result was not fewer manual tasks. It was an addition to students really enrolling in.
When the inquiry-management phase was rolled out, the admissions office had already improved its internal processing speed considerably. But there was still one big pressure point that remained: the volume of communications from applicants.
During admissions season, prospective students were still asking questions in the hundreds daily. Some of the questions asked were general. Others are extremely particular. The previous system was not designed to accommodate this without overwhelming the staff.
So we gave the AI agent a more general conversational role over WhatsApp, email, and web channels. It could respond to common inquiries about programs, deadlines, tuition, and campus facilities, but most importantly, it could also respond to status-specific inquiries by connecting to the admissions process. When a student inquired about their application, outstanding documents, offer letter, or payment fee deadline, the system replied with the real-time status and next step action.
This is more important than many institutions understand. A student holding for a reply is not merely seeking information. They are evaluating gravity. Quick, unambiguous communication of status tends to alleviate anxiety and promote confidence in the institution.
We also incorporated multilingual capabilities in the system to make the Arabic and English conversations both feel natural. Most complex queries—such as transfer credits, unusual scholarship cases, special accommodations, or program edge cases—were routed to humans with context already attached. Now the admissions team didn’t have to start every escalated conversation from scratch.
The platform was developed using a Python FastAPI backend, with Celery and Redis orchestrating asynchronous processes like the extraction of documents, eligibility scoring, scholarship simulation, message scheduling, and status-triggered communications. PostgreSQL retained structured application, scholarship, and enrollment information, while AWS managed the infrastructure, secure document storage, and delivery of email.
Claude was employed for document interpretation, messaging customization, bilingual interactions, and thinking through disorganized application documents. The React admissions dashboard provided staff a single workspace for application review, scholarship visibility, yield monitoring, inquiry escalation, and program-level demand patterns. React Native was used to deliver the student-facing mobile experience where applicable.
We connected to the university’s existing SIS via API connections, allowing the new AI agent to update status, confirmed enrollments, and deposits without requiring the institution to swap out its core student system. WhatsApp Business API and Twilio managed the real-time communication flows that made the system feel responsive to students instead of bureaucratic.
One important technical design decision was that the scholarship optimization and yield management were not encoded as fixed rules. We created a simulation layer so the university could run different scenarios on how various scholarship policies might influence enrollment, revenue, and budget consumption before completely committing. That’s why the system was strategically useful rather than just operationally efficient.
International document identification was one of the difficult components of the project. Students arrive from different countries with distinct grading systems, certificate formats, and attestation prerequisite needs. We had to develop the country-specific logic for grading the conversion and document expectations instead of treating each certificate as if it were locally regularized.
The recommendation for scholarships also proved to be a more multifaceted issue than a straightforward matter. It’s not like “just give the biggest discount to the best students.” The university had to strike a balance between improving yield, budget constraints, program priorities, and the mix of students. This made scholarship maximization a constrained decision problem rather than a rule of generosity.
Another hurdle was the frequency of the communication. When the AI agent was able to send rapid, customized follow-ups, the risk became under-communication. Students can tune out if the messages feel unsolicited. We fixed that with adaptive communication. Higher-engagement students received lighter-touch completion of an action. More precisely timed follow-up sequences were sent to students who were less engaged. Students who indicated they desired less communication were transitioned to a reduced-frequency path.
We also needed to assist the university in embracing a mindset shift internally. Admissions employees initially viewed the project as a process automation project. As it gained greater success, it was becoming increasingly clear that admissions was really a sales and conversion function. That forced the leadership to think differently about what the technology was supposed to be for.
The initial clear outcome was the speed. Application processing was reduced from five to seven working days to under 48 hours for almost simple cases. During the peak season, this had a huge competitive impact as the students got answers before most of the competing institutions could answer.
Document completion rates increased from almost 70% to 92% as the missing items were recognized and tracked automatically. That decreased the dead files sitting in limbo and enhanced the number of applicants who could progress into actual consideration.
The biggest business result was that of the enrollment yield. The yield, or the percentage of admitted students who actually enroll, rose from 60% to 81%, a 35% relative improvement. In practical terms, this meant an additional 400 students enrolled and an additional SAR 12 million in annual tuition revenue at average tuition levels.
The management of the scholarship budget was also much better. The university experienced a shift from SAR 1-2 million overruns in a few cycles to zero budget overruns, as the AI agent regularly monitored commitments versus available scholarship pools.
Handling time for responses to queries went from 24 to 48 hours, with many students completely ignored, to less than five minutes for the majority of queries, with the system dealing with some 95% of the volume directly. It made a huge difference in the applicant experience and the reputation of the admissions office substantially.
The same 15-person staff managed 20% more applications without overtime. More to the point, six workers were shifted out of repetitive data processing and into student counseling, campus tours, and relationship admissions work. That is a far better application of human talent in admissions.
Satisfaction with the applicant experience increased from 3.2 to 4.5 out of 5. And interestingly, there was a high percentage of feedback from enrolled students who mentioned quick, professional admissions as a reason for selecting the university. That’s the sort of result the leadership was looking for when we reimagined the challenge as growth rather than merely workflow.
We developed the solution using Python with FastAPI for orchestration, Celery and Redis for asynchronous handling, PostgreSQL for applications and enrollment information, Claude for document awareness and customized communication, React.js for the admissions dashboard, React Native for student-facing mobile processes, WhatsApp Business API and Twilio for messaging, API integration with the university’s SIS, payment gateway integration for handling deposits, and AWS EC2, RDS, S3, SES, and Lambda for infrastructure, storage, messaging, and scheduled execution.
The greatest takeaway was that production improvement matters more financially than operational efficiency by itself.
Yes, the admissions office had become more agile and scalable. But the big economic win was turning more offers into enrolled students. And that was a result of a sort of combination of speed, personalization, and disciplined perseverance. The AI agent quickly responded, spoke directly to the student’s selected program and interests, and maintained consistent follow-up over six weeks, something human teams simply cannot do for 12,000 applications at scale.
We also found that the scholarship strategy is not merely a matter of policy. It’s a modeling problem. Because universities don’t have a real-time view of budget, yield, and program priority combined, they often overspend or underspend their scholarship budgets in certain areas and waste them in others. Once those elements resided in one system, scholarship got a lot more strategic.
And lastly, we also found that admissions technology works when the leadership views it as a revenue engine, not as an administrative tool. At the company, we adopted this as our guiding principle for AI university admissions automation: the aim is not just about getting through more files. It is to admit the right students and do so more quickly, more reliably, and with an improved experience.
