When a person reads a payroll agreement, they can often interpret abbreviations, infer missing context, and understand industry shorthand.
A machine has a different job.
It must identify specific pieces of information, understand the relationships between them, and place each value into the correct structured field.
That raises a practical question: does the way we write payroll information affect how accurately an AI model can extract it?
To explore that question, Jebra conducted a focused experiment using a deliberately simple payroll agreement. The goal was not to prove a universal rule about artificial intelligence. It was to test one practical idea: whether clear wording improves structured payroll data extraction.
The test used 1,000 text samples representing 200 unique workers.
The same basic payroll agreement was rewritten in five different styles. Each version contained the same underlying information, including worker name, worker classification, class type, base class rate, attendance bonus, bonus threshold, attendance calculation method, holiday multiplier, and session earnings cap.
The model's task was to extract 11 individual fields and return them in a predefined JSON structure.
Here is the exact JSON used as the target baseline for a test case:
By keeping the underlying payroll rules consistent, the experiment isolated one main variable: how the information was written.
S1
Label
Shorthand
Description
Abbreviations, symbols, no verbs
Example
Layla Silva [C]: Boxing $40/class. $4/head >8 att (CI+LC). Holiday 1.5x. Cap $100/session.
S2
Label
Spelled-out
Description
Abbreviations replaced with full words, colons as separators
Example
Layla Silva [contractor]. Rate: $40/class — Boxing. $4/head when attendance exceeds 8 (check-ins and late cancels). Holiday rate: 1.5x standard. Session cap: $100.
S3
Label
Explicit prose
Description
Full sentences, one clean statement per field
Example
Layla Silva operates as an independent contractor. She is paid $40 for each Boxing class she teaches. For every attendee above 8, she receives $4 extra, counting check-ins and late cancels. Public holidays attract a 1.5x multiplier on her $40 base rate. Session earnings are capped at $100.
S4
Label
Explicit redundant prose
Description
Same as S3 but key values repeated across sentences
Example
Layla Silva is hired as a contractor. Her Boxing rate is $40 per class. Each Boxing class she teaches pays $40. For Boxing classes, a per-head bonus of $4 applies for each attendee above 8, counting check-ins and late cancels. The $4 per-head threshold is 8. The Boxing rate of $40 increases to $60.00 on holidays (1.5x the $40 base). Per-session Boxing earnings are limited to $100 in total.
S5
Label
Explicit prose with filler
Description
Same as S3 but legal boilerplate text introduced.
Example
Layla Silva is contracted as a self-employed instructor. Her contractual rate for Boxing is $40 per class. For the purposes of this arrangement, she is entitled to a per-head bonus of $4 for each attendee above 8, counting check-ins and late cancels. As per the agreed terms, work on designated public holidays is paid at 1.5 times the standard rate. Total per-session earnings are capped at $100.
The results highlight a distinct hierarchy in how language style impacts extraction accuracy:
• Spelling out words is paramount: Replacing cryptic abbreviations with full words (S1 to S2) yielded the largest performance jump, nearly 11 percentage points.
• Syntax matters less than semantics: Moving from spelled-out shorthand to full grammatical sentences (S2 to S3) made no meaningful difference. If the vocabulary is explicit, the model extracts efficiently without prose.
• Redundancy degrades accuracy: Repeating key values across sentences (S4) underperforms. Instead of providing helpful "anchors," duplication introduces token ambiguity that misleads the model.
• Filler text introduces noise: Adding filler text like “subject to the terms of the agreement” also had a negative impact.
A machine-friendly payroll description might look like this:
| Worker type | Contractor |
|---|---|
| Class | Boxing |
| Pay type | Per class |
| Class rate | $40 |
| Bonus | $4 per attendee |
| Bonus threshold | Above 8 attendees |
| Attendance basis | Check-ins and late cancellations |
| Holiday multiplier | 1.5 |
| Session cap | $100 |
This format is not sophisticated prose. It is simply explicit, consistent, and easy to map to structured fields. Each value has one clear purpose and one likely destination.
Payroll arrangements are often more complicated than a simple hourly rate.
Fitness and wellness businesses may compensate workers using combinations of per-class rates, hourly wages, attendance bonuses, tiered rates, holiday multipliers, session caps, and different rules for different class types.
Turning those arrangements into structured payroll data requires more than identifying numbers. A system must understand what each number represents and how it relates to the surrounding rules.
Improving the quality of the source language can reduce ambiguity before payroll calculations begin.
This was a deliberately narrow experiment. It tested one relatively small model, one flat payroll agreement, and 11 predefined fields. More complicated agreements and different models may produce different results.
The findings should therefore be treated as practical evidence rather than a universal law for every AI system.
Still, the experiment offers a useful operating principle: when preparing information for structured extraction, spell out key concepts, state each rule clearly, avoid unnecessary repetition, and remove language that does not contribute meaning.
In other words, do not write more simply because the machine is less intelligent. Write more clearly because the machine has a more specific job.
Good machine-readable writing does not need to be long, formal, or technically impressive.
It needs to be unambiguous.
For structured payroll extraction, the best input may be the version that gives every rate, threshold, multiplier, and rule one clear meaning and one obvious place to go.
Write for the machine, not only for the expert.
Ready for the next step? Read Payroll AI Experiment #2: When Payroll Gets Complicated, the Machine Gets Harder to Please. This follow-up explores what happens when payroll agreements become more complex and compares how different model compression levels perform.