Real Remarka engagement · July 2026

Case Study: A 90-Minute Monthly Audit Reduced to Two Minutes

A St. Louis nonprofit needed to manually review thousands of employee time records every month. Remarka replaced the error-prone five-step manual audit with a single scripted step that saves time, eliminates missed records, and produces a clear report in about two minutes.

At a Glance

Client: A St. Louis-based nonprofit with more than 250 employees to audit across multiple locations
Project: An internal timekeeping audit app
Delivery: Less than two days from discussion to delivery
Cost: $500, all-in
Special Notes: Client data kept private, no AI needed by the app.

The Problem

This organization has employees at numerous locations. Employees are required to clock in from a webpage on a company computer at an approved location. The organization's older timekeeping system cannot prevent someone from using a phone or home computer, but it does record the internet address associated with every clock-in.

There is no budget or plans for updating the timekeeping system itself. That left the organization's Human Resources team responsible for reactive enforcement.

Each month, a senior employee exported a 150+ page report containing roughly 4,200 records and reviewed it by hand.

The audit depended on knowledge that lived mostly in one person's head:

Which internet addresses belonged to which locations (aka static IPs)
Which locations did not have a static internet address (aka dynamic IPs)
How overnight shifts were tracked and communicated in the log
How to handle employees working at different locations
Which unfamiliar internet addresses were legitimate (aka false-positives)
Which records represented likely violations and needed research

The work was difficult to delegate and vulnerable to missed records and inconsistent judgment calls. Replacing the timekeeping system or changing the organization's network would have cost far more than the problem justified. Coming up with some kind of cheap fix had been on the wishlist for some time.

The question was simple: Could the organization enforce the policy more reliably, in less time, without replacing the system or adding another subscription?

How We Worked

1. Uncover the rules

The client and I spent about 60 minutes walking through their current audit process, including looking at the export files and discussing known rules and exceptions. This turned the client's unwritten knowledge into clear business rules that could be scripted.

2. Build with real data

I used several export samples to build a working version of the report app. It processed the full exports and produced a findings report that the client could evaluate immediately.

3. Identify gaps and iterate

Upon looking at the first version of the new report, the client noticed problems. This triggered a second conversation, about 30 minutes, that revealed a slew of additional business rules, legitimate exceptions, and data format nuances that were not part of the initial conversation but were equally critical to being able to accurately process the timeclock data.

4. Improve the report

Once all business rules were accounted for and added to the program, we added useful views that had not been practical before, including month-over-month trends and how each location actually clocks in.

5. Test and package it

AI helped me write and test the program quickly. I stress-tested it with hundreds of thousands of synthetic records, clean data mixed with malformed, broken and edge-case records, deliberately trying to break it so the failures surfaced here rather than in front of the client. The tool was packaged as a simple double-click Windows application the client could run without technical help.

What Changed

The complete workflow originally had seven steps:

Export the timeclock data
Visually scan roughly 4,200 rows
Use internet tools to look up unfamiliar internet addresses
Make individual judgment calls
Investigate possible violations
Document violations and summarize the findings
Communicate the findings

The first and last steps remain, but the five-step manual process in the middle turned into a single step:

Export the timeclock data
Run the application on the export
Communicate the findings

The application now handles the sorting, internet address lookups, documented exceptions, historical comparisons and report creation.

The Finished Solution

Instead of reviewing a 150+ page report, the HR team now receives a short report that:

Lists likely violations with the evidence behind each one.
Separates records that need a human review, often revealing a business rule worth adding.
Shows month-over-month trends and how each location clocks in.

Each result includes a plain-English explanation. The client leadership still makes any final personnel decision, but no longer has to search thousands of rows to catch problems and document the evidence.

The program follows the same written rules every time, eliminating varying judgment and misses. It does not use any AI.

Results

About two minutes per month

Exporting the data takes about one minute, running the program takes roughly 20 seconds, and reading the report takes about 15 seconds. Allowing for normal handling time, the task takes about two minutes.

That reduces the annual workload from roughly 18 hours to about 24 minutes.

Missed rows and inconsistent judgments eliminated

The audit no longer depends on someone visually scanning thousands of rows or remembering how a similar record was handled last month. This eliminates accidentally missed records and inconsistent judgment calls from the audit process.

Previous violations uncovered

Testing six months of historical records found actual violations that earlier manual audits had missed.

Fewer false alarms

Legitimate exceptions are now part of the audit rules, reducing the chance that an employee is questioned over unusual but permitted activity.

Easy to delegate

The knowledge required to perform the audit now lives in the program instead of one employee's head. Someone else can run it without learning every rule and exception behind it.

No new system or subscription

The organization kept its existing timekeeping system. There were no network changes, ongoing AI costs or new monthly fees.

The Role AI Actually Played

AI did not decide how the audit should work. The client understood the business process. I uncovered and documented the rules, designed the solution and verified the results.

AI helped me write and test the program much faster. Work that could have taken several weeks was completed in less than one working day and delivered in under two days.

The finished tool is completely deterministic. The same data processed under the same rules produces the same result every time. AI made a small custom solution practical to build, but it is not involved when the client runs it.

Could Something Like This Be Hiding in Your Business?

A process may be a good candidate for automation when one or more of the following is true:

It must be done regularly
It follows mostly predictable rules
It requires copying or sorting raw data
It requires jumping between applications or cross-referencing the data between them
It takes more skilled time than it should
One experienced person knows most of the exceptions
Replacing the main system would be too expensive

Expense audits, license renewals, vendor reconciliation, compliance checks and recurring reports can all fit this pattern.

The answer is not always a major software project. Sometimes it is a small tool built beside the systems you already have.

If you have a repetitive task that everyone dislikes but nobody has fixed, describe it to me. A 30-minute conversation is usually enough to determine whether it is worth automating.