106 Things Disposition Data Can’t Tell You

(Happy Friday, Job Board Doctor friends! I am out learning and imbibing at RecFest UK this week. Fortunately for me (and you), Lou Goodman, in partnership with Jobiqo, has taken up the mantle this week and Lou is continuing her work addressing the current debate swirling around disposition data. As always I want to hear what you think, so let me know and I will catch up with you all next week.)

Or, why disposition data is an unreliable narrator.

My first encounter with this problem was in 2007. I was at MediaCom, leading the media planning and buying team for a very large, global client’s recruitment marketing, and we were doing something relatively rare for the time — manually matching ATS disposition data to click and post impression data every month. 

It was a significant operational lift and provided us with genuinely useful insights into media effectiveness. Despite having access to that data, we were still consistently asked to improve the quality of the applications we delivered. But the data didn’t give us the means to do so, because it didn’t tell us why hiring decisions were made the way they were. Rejection reasons were applied inconsistently across markets and told us nothing specific about what we needed to optimise for to improve quality. 

Now, that was partly because, as a media agency, we didn’t have access to candidate profiles. Job boards are in a different position; they have profiles, CVs, applications, and increasingly, AI capable of drawing inferences from that, but the lessons from that project almost 20 years ago are still relevant. 

So yes, the data available to a job board in 2026 is substantially richer than what we had in 2007, and that does make better correlation possible. But you can have the most sophisticated candidate profiles in the world and still be matching it against a disposition code that was selected by default, logged inaccurately, or reflects a process that changed three times after the job was posted. 

And, as any good data analyst will tell you, correlation isn’t causation. 

It’s why the focus on how job boards evolve must shift from unreliable hiring outcomes to measurable recruiting outcomes that track match quality before candidates are passed to the employer’s ATS. 

This article works through 106 things disposition data can’t tell you, grouped into 10 different categories. 

Gaps 1-8 all describe what goes wrong within the hiring funnel after the initial application, while Gaps 9 and 10 focus on the integrity of the input data itself. 

Gap 1: The “why” behind decisions (12 unknowns)

A disposition code that reads “rejected” isn’t a reason. It’s a label that may have had very little to do with the actual candidate, applied at the end of a process. For example, the difference

between an actual skills mismatch and a perceived skills mismatch is invisible in the data – and those two things point in entirely different directions if you’re trying to learn something useful. 

1. Was it a skills mismatch or a perceived skills mismatch? 
2. Was the salary invisible until partway through the process, and too low?
3. Were there cultural fit concerns (legitimate or biased)? 
4. Were there communication issues during the interview? 
5. Were there overqualification concerns? 
6. Were there concerns that the candidate seemed likely to leave quickly?
7. Was there unconscious bias (age, gender, race, background)? 
8. Was there a personality clash? 
9. Was there an internal candidate already chosen? 
10. Did the budget get cut after they applied? 
11. Did hiring managers’ priorities change? 
12. Was the rejection reason logged accurately, or was a default category selected?

Gap 2: Employer capacity constraints (6 unknowns)

“Not selected” and “not reviewed” can look identical in the data. A recruiter under pressure who stops at the first handful of viable candidates leaves a trail of decisions that appear rational. Research shows that doctors’ prescribing behaviour shifted from 75% generics to 98% simply by reordering a dropdown menu. The data doesn’t tell you how much of what you’re reading is driven by UI rather than judgement. 

1. How many strong matches were never reviewed by recruiters due to time constraints?
2. Did the hiring manager review only the first X applications and ignore the rest?
3. How many excellent candidates existed, but the team only had bandwidth to interview Y? 
4. Did they stop reviewing after finding “good enough” rather than the best match?
5. Were strong candidates missed due to poor search terms in the ATS?
6. How many “not selected” outcomes were capacity failures rather than actual poor matches? 

Gap 3: What changed during the hiring process (11 unknowns)

Disposition data records the final state. It doesn’t record what the role was when the candidate applied, how many times the requirements shifted, or whether the process was genuine from the start. 

1. Did job requirements change after posting or mid-process? 
2. Was the budget reduced, lowering the salary range? 
3. Did something about the role shift, for example, from full-time to contractor?
4. Did location flexibility change (remote to hybrid to office)? 
5. Did the hiring manager leave during the process, and did that change the candidate preferences? 
6. Did an internal candidate emerge partway through? 
7. Was there a well-qualified candidate pool, but the position was delayed?
8. Was the job delayed, but interviews continued to build a talent pipeline?
9. Did the team structure change (reporting lines, team size)? 
10. Was the required start date significantly pushed out or brought forward?
11. Was it a ghost role? 

Gap 4: Candidate experience and decision quality (12 unknowns)

When disposition shows “offer declined,” the data says nothing about why a candidate who made it through the entire process walked away. That’s one of the most commercially significant signals in recruitment – and it’s almost entirely absent. 

1. Did the candidate accept an offer to stay from their current employer?
2. Was compensation below market rate or candidate expectations? 
3. Did the interview process take too long, and did the candidate accept another offer?
4. Were interviewers unprofessional or disorganised? 
5. Were there cultural issues that emerged during interviews? 
6. Did the candidate receive a better offer with more flexibility? 
7. Was the commute or location unworkable? 
8. Did the lack of remote options cause the decline? 
9. Were growth and advancement opportunities unclear? 
10. Did the candidate research the company and find negative reviews?
11. Was the hiring manager’s rapport poor during interviews? 
12. Were there team dynamics issues revealed during a panel interview?

Gap 5: Time-Based Context and Engagement Decay (7 unknowns)

A timestamp tells you when something happened. It doesn’t tell you what the candidate’s state of mind was by the time it did. A six-week process looks the same in the data whether the candidate was engaged throughout or had mentally accepted another offer by week three. 

1. Did the candidate interview elsewhere during the weeks or months of silence?
2. Were they still genuinely interested during the interview, or going through the motions
3. Did communication gaps signal disorganisation, leading them to be sceptical about the company? 
4. How engaged was the candidate at each stage, versus simply staying in the process?
5. Did they withdraw interest informally without notifying anyone? 
6. Did other opportunities emerge that were more attractive? 
7. Did a too quick or too lengthy process cause top candidates to drop out early?

Gap 6: Batch Processing and Shortcuts (9 unknowns)

Mass disposition updates are a particular problem. When a recruiter closes out 40 applications in a single session, what you are seeing is an administrative event, not a considered assessment. Defaults are selected, queues are cleared, and the data appears to show decisions were made when they weren’t.

1. Did someone actually review all applications individually? 
2. Were candidates bulk-rejected after reviewing only the first X? 
3. Were the knockout questions poorly designed, filtering out good candidates?
4. Did the ATS auto-reject based on flawed keyword matching? 
5. Was there an integration error that misclassified candidates? 
6. Did a junior recruiter mass-reject without senior review? 
7. Were candidates rejected based on CV formatting issues from ATS parsing failure?
8. Did timezone or location filters exclude good candidates solely because of location?
9. Were salary filter thresholds set incorrectly? 

Gap 7: Systemic Patterns and Hidden Bias (16 unknowns)

Disposition codes exist. Bias audits largely do not. Without the context behind each decision, patterns that should trigger concern appear to be normal variation. The data doesn’t lie – but it doesn’t tell the truth either. 

1. Are women rejected for “cultural fit” at higher rates than men? 
2. Are older candidates screened out disproportionately? 
3. Are candidates with non-traditional backgrounds systematically excluded?
4. Is “cultural fit” functioning as code? 
5. Are certain names or ethnicities receiving lower callback rates? 
6. Are mothers penalised relative to fathers for employment gaps? 
7. Do certain interviewers cite cultural fit while others do not? 
8. Are candidates from specific universities favoured regardless of qualifications?
9. Are disability accommodations triggering rejections? 
10. Are accent or communication style differences causing bias? 
11. Are candidates with foreign credentials undervalued? 
12. Are career changers rejected despite transferable skills? 
13. Is “overqualified” functioning as a proxy for age discrimination? 
14. Are part-time work histories penalised unfairly? 
15. Do hiring panels show composition bias? 
16. Are LGBTQ+ indicators – pronouns, organisation memberships – affecting decisions? 

Gap 8: Parallel Processes and Competitive Dynamics (12 unknowns)

Disposition data is recorded per requisition, but candidate behaviour isn’t. The same person may be in five processes simultaneously, and you don’t know if the data captures that, or whether you lost them to a competitor, retained them despite competition, or were never really in the running. 

1. Did the candidate apply for multiple roles at the same company? 
2. Were they rejected for Role A but well-suited to Role B, a role they didn’t know about?
3. Did they interview with competitors simultaneously? 
4. Did they have other offers they turned down to take yours? 
5. Were they your second or third choice after top candidates declined?
6. Did you lose them to a competitor with a faster process? 
7. Did they withdraw from other processes when you hired them?
8. Were they interviewing with you as a backup while waiting for a preferred role?
9. Had they applied previously and been rejected? 
10. Are they a seasonal applicant who applies every year? 
11. Did they refer friends who also applied? 
12. Were they weighing an internal transfer against an external move? 

Gap 9: Job Posting and Advertising Accuracy (11 unknowns)

This gap addresses the problem of matching candidate profiles to a poorly written, misleading, or inconsistent job description, leading to disposition codes that appear rational but are based on flawed criteria. 

1. Did the job description misrepresent the actual role? 
2. Was the advertised minimum experience requirement contradicted by an unwritten, internal requirement (e.g., years of experience not listed)? 
3. Did the job description contain misleading language, such as encouraging applications from partially-qualified candidates, only for them to be immediately screened out for missing mandatory criteria? 
4. Were key requirements (technical skills, certification) listed as mandatory in the advertisement when they were only preferred? 
5. Was the location/flexibility status (remote/hybrid/in-office) inaccurately represented in the ad? 
6. Did the advertised salary range significantly misrepresent the actual budget for the role? 
7. Was salary included in the job posting? 
8. Did the job title accurately reflect the level and scope of the role, or was it inflated/deflated compared to industry standards? 
9. Was a short application deadline listed, but the role was not reviewed until weeks later
10. Did the ad fail to include legally required disclosures or critical information that led candidates to self-select out later in the process? 
11. Was the advertised “team culture” or “company mission” inconsistent with the reality revealed during the interview process, influencing candidate quality? 

Gap 10: CV and Application Accuracy (10 unknowns)

This gap addresses the problem of using the candidate’s application as an accurate data source when it may be inaccurate, exaggerated, or fabricated. Since the ATS disposition code reflects only a judgment about this information, it fails to capture the integrity of the data it is judging. 

1. Did the candidate exaggerate years of experience? 
2. Is their stated expertise, deep knowledge or surface-level familiarity?
3. Did they inflate their role or responsibilities? 
4. Are education credentials accurate? 
5. Did they fabricate employment dates to hide gaps? 
6. Are the reference contacts legitimate, or are they friends posing as managers?
7. Did they over-tailor their CV so it no longer represents their experience?
8. Are technical skills self-assessed accurately or inflated?
9. Did they copy job description keywords without experience? 
10. Are portfolio or project examples actually their work? 

What this means

The unknowns in this article are not edge cases. They’re the normal conditions under which hiring decisions get made and recorded. Taken individually, each one is a limitation you could argue around. Taken together, they describe a signal that simply isn’t reliable, and the stakes are rising. 

When disposition data gets fed into AI hiring tools, these gaps stop being analytical inconveniences and become something more serious. A model trained on flawed disposition data systematises past mistakes, the bias patterns in Gap 7 stop being organisational tendencies and become encoded decisions, applied at scale, with the appearance of objectivity. 

There’s also a more fundamental issue that the data matching work from 2007 made clear. Even when you do the hard work of connecting disposition outcomes to other data sources, you’re still working with a signal that’s not properly constructed. Rejection reasons are self-defined, inconsistently applied, and unvalidated across employers, roles, and hiring managers. Matching rich candidate data against that signal doesn’t make the signal more reliable; it just gives the unreliable narrator a better microphone. 

If disposition data is the only measure of an application’s quality, a job board’s role will simply be to pass candidates through a process and record employers’ decisions. Instead, job boards should develop an independent point of view on match quality – on both the candidate side and the job side – that doesn’t depend on employer disposition codes being reliable. 

This means shifting focus to recruiting outcomes – the measurable quality of connection and match quality before the candidate leaves the platform – rather than waiting for data on ultimate hiring outcomes. 

At RecBuzz Budapest in April, Alex Chukovski made the point that job boards need to take control of this and make disposition data irrelevant. That’s the right instinct. The goal should be to build a picture of candidate quality so complete that the employer’s disposition code becomes their opinion of the candidate rather than a verdict on the job board’s performance.

[Want to get Job Board Doctor posts via email? Subscribe here.]

[Got a tip, document or intel you want to share with the Doc? Tell me. Tip so hot you need it to be encrypted? Use Signal.]