The model might not have created the failure. It might have revealed a flaw the old process had been quietly hiding. Andre was outside Nathan's office before 8 the next morning. I think I know what happened. Nathan read the appendix.
You have access to the pilot data? No, I haven't touched the system since I was removed. Good. Nathan examined the note again. What do you need checked?
Andre explained. Nathan did not promise to save the project. He said only. This deserves to be tested. Because Andre no longer had active project access, Nathan took the theory to the independent review team himself.
Finance agreed to inspect the source data before issuing a final failure report. Most of the day passed with no answer. Clare did not call Andre upstairs. She did not reassure him. That was exactly what she had promised from the beginning.
She had opened the door. She would not stand between him and the consequences of his work. Late that afternoon, Nathan came to Andre's desk. They found it. One distribution center was still sending location identifiers based on the old platform.
Several codes had not been fully mapped during the pilot. The new coordination logic interpreted some identical destinations as different routing points, causing duplicate handling and unnecessary transfers. Andre released a breath he had not realized he was holding. Nathan lifted one hand. This explains the failure.
It does not prove your savings estimate. Andre nodded. Then we test again. The second pilot used standardized routing data. This time, Andre could review the output, but could not manually manipulate results.
Finance wanted to know whether the system worked without its creator constantly rescuing it. Over the next two weeks, the numbers moved in the direction Andre had predicted. Truck waiting time fell. Duplicate processing decreased. Unnecessary transfers dropped.
Ontime departures improved, one metric underperformed. And Andre resisted the temptation to explain it away. He traced the cause, documented it, and let the independent team decide whether it changed the conclusion. At the end of the pilot, finance combined the validated performance improvement with a broader analysis of historical shipment volume across all four distribution centers. That longer data set allowed them to annualize the result more reliably than Andre's original sample.
His estimate had been approximately $3.7 million in avoidable losses each year. The verified number was roughly $4.15 million. The final report went to the board. One draft presentation described Andre's role as supporting analysis. Nathan corrected it.
During the meeting, he said that wording is inaccurate. Andre identified the pattern and built the original logic. The validation team improved and verified it, but the discovery was his. Andre sat several seats away and felt a strange calm. The man who had doubted him most was now insisting that his contribution be recorded accurately.