
Testing a CV parser with your own documents
A well-filled candidate profile isn't proof of correct extraction. Check each field against the source, then inspect what the system saves or changes.
Remo Vloet6 min.

Establish whether two records describe the same person before bringing contact details or applications together.
Matching information, such as a name, contact detail or work history, can flag a possible duplicate candidate profile. Treat that match as a reason to investigate. First establish whether the system shows two people, two records for one person or two separate applications. Then decide what needs correcting and what information must remain.
Two search results with the same name can create confusion. A colleague wants to send a message while another record already contains a conversation. Quickly deleting one result may look tidy, but could remove the agreement you needed to follow up.
A duplicate person record exists when the same person has unintentionally been entered as separate candidates. Two applications for different vacancies aren’t automatically an error. Your ATS’s structure determines how those relationships appear and which records you need to compare.
Keep the person separate from their applications. One candidate can be interested in several vacancies, each with its own conversation and outcome. Deleting a second application as though it were a duplicate person record would remove valid recruitment history.
Suppose you see the same candidate against an account manager vacancy and a sales manager vacancy. They’ve declined one role and have an interview scheduled for the other. Those different outcomes need to remain, even if general contact details are managed against one person.
Ask your administrator which identifier belongs to the person and which to the application. The name at the top of a screen doesn’t always tell you. Agree how colleagues can find other relevant applications within their access permissions.
A match may be strong or weak depending on context. A common name gives little certainty. An identical email address can be useful, but could also be a typing error or a shared address. Examine relevant details and their sources before drawing a conclusion.
This table is a review aid, not an automatic decision rule or statistical model.
| Signal | What remains to be checked |
|---|---|
| Same name | Could these be different people? What relevant additional information is available? |
| Same email address | Was it entered correctly, and does it belong to this person? |
| Similar telephone number | Is only the formatting different, or is it a different number? |
| Same profile reference | Do both records point to the same intended profile? |
| Overlapping work history | Do roles and dates agree, and what are their sources? |
| Different current employers | Is one detail older, or do the records concern different people? |
A changed surname or new email address doesn’t necessarily mean a different person. Equally, a shared employer doesn’t establish that two people are the same. Record the evidence supporting your conclusion and what remains unresolved.
Don’t gather unnecessary personal information just to settle an administrative suspicion. Start with relevant records you’re permitted to use. If appropriate verification is needed, agree who will carry it out and what it requires.
Compare identity first, then content and active commitments. Don’t immediately choose a main record because it’s newer or more complete. One may hold current contact details while the other contains next week’s promised call.
In this fictional example, two records may describe the same candidate, Robin Davies. The first contains an older CV and a personal email address. The second came from a later import, has the same work history and lists a different email address. The first also contains a scheduled follow-up call.
Begin by establishing whether both concern the same Robin. Shared work history helps the investigation but doesn’t provide complete certainty by itself. Examine the source, date and relevant other details. Until identity is clear, keep merging as an open decision.
Once identity is confirmed, assess the fields individually. Which address does Robin now use? Is the older employer historical experience that should remain? Does the promised call relate to an active vacancy? The newest record doesn’t automatically have the right value for every detail.
Assign one colleague to finish the review. Tell the recruiters involved where ongoing follow-up belongs so nobody sends another copy of the same message from a different profile during the investigation. Record the decision and its reason in the agreed place.
Check whether your system supports merging and what the action does to fields, documents, communication and applications. Ask about permissions and external references too. Start with a fictional test, because merging isn’t easily reversible in every product.
Before merging, agree who checks the result and how a wrong decision could be recovered. Without a confirmed recovery route, treat the operation as potentially irreversible. You can put ongoing follow-up with one colleague while keeping both records until the decision is settled.
For your configuration, ask to see at least:
An export may help you investigate information, but doesn’t establish that it can restore every relationship and state. Ask about the actual recovery process, and keep any test files in an appropriate, controlled environment.
Simply’s ATS and CRM connects records through relationships. When investigating a duplicate, keep each candidate’s vacancy connections visible. For Robin, the current contact details and promised call both need to survive. Agree with your administrator how to carry out the correction once identity is confirmed.
If merging isn’t supported or identity remains uncertain, agree a process appropriate to your system. Prevent duplicate follow-up without blindly deleting records. Ask your administrator or supplier how to preserve relevant information and relationships.
Investigate how the second record appeared. Was an import run twice, could a recruiter not find the existing profile, or did the candidate arrive through another source? A concrete cause suggests a useful process change. Reducing the current duplicate count alone says little about new information entering the system.
Try a small import file containing a known existing profile and a changed contact detail. Observe what happens and record it. Check whether a user with fewer permissions is allowed to recognise the same profile. Don’t create a process that bypasses access controls to investigate a possible duplicate.
Agree who reviews uncertain cases and how colleagues raise them. Keep it practical: references to the records, the matching signal and the open question are more useful than a broad statement that the database is messy.
Want to work through the check for your setup? Request a Simply demo with two fictional candidate records and separate applications. First ask whether and how possible matches become visible. Then examine which action is supported and what your team still needs to check.
About the author

Remo Vloet is a co-founder of Simply, the AI Operating System for recruitment agencies: inbox, meetings, sourcing, CV parsing, search and matching, documents and automation in one system. With a background in building complex software, he contributes to the technical vision behind Simply.
LinkedInArticles by Remo VloetNo. Different people can share a name, and names can be written in different ways. Use it as a signal to compare relevant additional information and its sources. Don't merge while you're uncertain whether both records describe the same person.
An email address can help recognition, but it can be entered incorrectly, shared or changed. Investigate the context and the system's rules. A technically unique field doesn't establish that all the content belongs to the right person either.
No. Some systems use separate profiles or records per vacancy, containing their own conversations, assessments and outcomes. Check the data model before treating those records as an error. Separate applications' histories need to remain recognisable.
That depends on the information and your software's behaviour. Don't automatically choose the newest or oldest record. Establish which values are current and which historical information and relationships must remain. Have someone with the right authority carry out the appropriate action.
Don't assume so. A suggestion based on similarity can be wrong, and an incorrect merge may be difficult to recover from. Ask about the signals, human review and consequences of the action. Test the available function with fictional cases where information conflicts.

A well-filled candidate profile isn't proof of correct extraction. Check each field against the source, then inspect what the system saves or changes.
Remo Vloet6 min.

Measure whether your team can use the information, without treating a populated field as automatically correct or current.
Remo Vloet6 min.

Decide what belongs to a candidate, a job or an application before adding more fields. Your records need to make sense when those relationships change.
Remo Vloet7 min.

Find out how Simply can completely evolve your workflow. No slides, just product.