Most AI projects fail before they even start.

Most AI projects fail before they even start. Not because of the tech, but because of why they’re started.

It's exciting to see companies jumping into AI. But too many projects I'm seeing lack real strategy behind them, driven by external pressure, FOMO, or just the urge to "do something with AI."

We help companies plan their AI initiatives, so I wanted to share a real use case that goes beyond summaries and chatbots.

A client came to us with a messy process. They receive tax documents in every format imaginable, digital files, scanned images, photos, and some that were practically drawn on napkins (slight exaggeration, but you get the idea). To get that data into their CRM, they'd hire interns for manual entry. Not scalable. Not cheap.

We gave them an alternative.

Using optical character recognition and trained models in AWS Textract, we fully automated the process. Files get uploaded to the appropriate client accounts, an integration fires, and the key values come back in structured JSON format.

Because accuracy matters, we also built in human-in-the-loop validation, a UI that displays the original document side by side with the extracted values so discrepancies get caught immediately.

I'm sharing this because it highlights a few things worth thinking about:

1. The ROI is easy to calculate. You can directly measure time and cost savings against what manual data entry was costing.     2. It makes employees more efficient rather than just putting AI agents in front of customers.     3.  It's not blindly using AI. Transparency, human oversight, and quality assurance are baked into the process.

If your company is exploring AI, start with problems like this. Ones where the value is clear and measurable.