Building enterprise software is easy to get wrong. In 2025, we contracted to rebuild a core customer service pipeline for a $35B multinational. We met the technical success parameters, but the project taught us hard lessons about user adoption and operational boundaries.
Here are the mistakes we made and the principles we now enforce on every build.
1. Designing for the VP, Not the Agent
We spent weeks in conference rooms with the VP of Operations. We mapped the theoretical flow. We built exactly what they asked for: a fully unified control panel. But when we launched, we realized that the agents on the floor hated it. They preferred their old clunky tools because our panel required three extra clicks to resolve common billing issues.
Lesson: The person buying the software is rarely the person using it. We now embed alongside floor agents on day one of discovery.
2. Underestimating the Cost of System Sync Latency
We designed our LLM pipeline to trigger webhooks in real-time. But the client's legacy mainframe updated records in nightly batches. Pushing updates in real-time caused race conditions and data conflicts across systems. We had to rewrite the sync logic to match their mainframe physics.
3. Ignoring the 'Silent Obstructors'
In every enterprise, there are managers whose leverage depends on the complexity of current manual steps. Automating those steps threatens their headcount metrics. We built the system, but team managers quiet-quit using it, routing tickets around the automation.
Now, we build incentives for middle management directly into the roadmap — freeing them to lead strategy rather than managing data transcription.