How to verify scholarship applicant documents at scale
Document verification is where scholarship programmes quietly lose weeks. A thousand applications with four documents each means four thousand files, and in our experience the problems (a missing page, an expired income certificate, a marksheet that doesn’t match the stated marks) surface at interview stage, when fixing them is most expensive. This post gives you a verification matrix you can adapt, and shows where software and AI help. For the full lifecycle and tool comparison, see our scholarship management software guide.
Separate three questions
Verification mixes three different checks, and teams that treat them as one end up re-checking everything:
- Is it present and readable? Every document uploaded, right type, all pages, legible. This is clerical and can be automated.
- Does it say what the applicant said? Marks, income, institution, name. This is extraction and comparison. Software can do most of it with a confidence score.
- Is it genuine and does it meet the rule? Has it been altered? Is the income certificate from an authority you accept? Does hardship justify an exception? This is judgement. A person decides.
Build your workflow in that order. Only applications that pass check 1 move to check 2, and only uncertain or failed check-2 results reach a human at check 3.
The verification matrix
| Document | What to check | How | Who | Evidence to keep | Keep for |
|---|---|---|---|---|---|
| Marksheet / transcript | Name, board or university, year, marks against threshold | Extract marks; compare with application; flag mismatch beyond a set tolerance | System, then verifier for flags | Extracted values, confidence, verifier decision | Life of award plus your audit period |
| Income certificate | Issuing authority, date within validity, income below ceiling | Check authority against an accepted list; check date | Verifier | Authority, date, income band (not the full document, if you can avoid it) | Life of award |
| Enrolment / admission letter | Institution, course, year, academic session | Match institution to a master list; check session dates | System, then verifier | Institution ID, course, session | Renew each year |
| Identity proof | Name and date of birth match | Compare with application; in India, use masked Aadhaar or another ID | Verifier | ID type and last four digits only | Life of award |
| Bank details | Account holder name, IFSC, account active | Bank account validation before first payment | Finance | Validation result and date | Life of award plus finance retention |
| Guardian consent (under 18) | Guardian identified, consent recorded | Consent form or OTP-based record | System | Consent record with timestamp | As long as data is held |
A note on the identity row. In India, applicants are used to giving Aadhaar because the National Scholarship Portal requires it. Your programme probably doesn’t need the full number. Practitioner guidance from the Avni project advises against storing full Aadhaar numbers without an Aadhaar Data Vault, and suggests masked copies or the last four digits instead (Avni). The DPDP compliance checklist covers consent and retention for the rest of the matrix.
Worked example: 1,000 applications, two verifiers
Say you have 1,000 applications, four documents each, and two staff verifiers for three weeks.
- Check 1 (presence and readability) runs automatically on upload. Applicants with missing or unreadable files get a message within the hour asking them to re-upload, while the deadline is still open. On a portal we built for a sponsor-funded scholarship, these confirmation and status messages needed repeated refinement in testing. Applicants read a missing confirmation as failure.
- Check 2 (extraction) pulls marks, income and institution, and compares them with what the applicant typed. Say 70% of applications match on every field with high confidence. They go straight to the eligibility rules.
- Check 3 (judgement) gets the rest: 300 applications with a low-confidence field, a mismatch, or an authority not on your list. Two verifiers spend their three weeks on 300 hard cases instead of 4,000 files.
The numbers above are illustrative. Run your own through the admin-hours calculator on the hub.
Where AI helps, and where it must not decide
AI extraction is good at reading marksheets in several formats and languages and saying how sure it is. It is not good at judging hardship, spotting a well-made alteration, or deciding an exception. Our DozAI eligibility screening mission is designed around that line. Each application gets one of three results (meets, does not meet, needs a document or review) with the rule version and evidence attached. Panels make the final selection, and appeals go to someone who did not make the first decision.
Before you switch on any automated rule, test it on past applications with known outcomes, and check whether it rejects one group of applicants more often than others. A rule like “marksheet must be from a recognised board” can quietly exclude students from open schooling.
Common mistakes
- Verifying everything for everyone. Verify in stages. Don’t ask for bank details until a student is shortlisted.
- Keeping full documents forever. Store the extracted fact and the verifier’s decision; delete what you no longer need under your retention policy.
- No verifier record. If you can’t show who checked a document and when, you can’t defend a decision to a donor, an auditor or an applicant.
- Re-verifying at renewal from scratch. Year two needs a new marksheet and enrolment letter, not a new identity check. Our disbursement and renewals workflow shows which documents renew and which carry over.
Map your scholarship programme before you choose a tool
Tell us how your programme runs today and we’ll show you where a system would save the most time, and whether a specialist product would serve you better than a build. Take the free technology assessment →