Decision Systems Under Constraint: Building AI Systems for Real Organizations

The Reality

Real organizations don't have unlimited budgets, unlimited data, or unlimited time. They operate under constraints. And that's actually fine. The best decision systems I've seen were built under tight constraints.

The key is designing for constraints, not pretending they don't exist.

What Constraints Look Like

Limited data is common. You don't have six years of historical data. You have six months. Or you have data, but it's fragmented across multiple systems. Or you have data quality issues that make it unreliable.

Legacy systems are everywhere. You can't rip out the ERP system that's been running your business for 15 years. You have to work with it. You have to integrate with it. You have to respect the technical debt.

Organizational constraints are real. People resist change. Competing priorities pull resources. Budget is limited. Skills are scarce. You can't hire 50 data scientists. You have three.

Time pressure is constant. You need to show value quickly. You can't wait for perfect data. You can't spend a year building the perfect system. You have three months.

How to Design for Constraints

Start simple. Use simpler models that require less data. Avoid complex architectures. Focus on core functionality. Iterate from there. I've seen teams build sophisticated systems that nobody could maintain. Then they rebuild with simpler approaches and everything worked better.

Leverage domain expertise. Use human expertise to augment data. Combine AI with human judgment. Build hybrid systems. Respect organizational knowledge. A manufacturing company we worked with had 20 years of production knowledge from their team. We built a system that combined that knowledge with six months of data. It worked better than a pure AI approach would have.

Integrate with existing systems. Work with legacy systems. Use APIs and middleware. Minimize disruption. Plan for technical debt. Don't try to replace everything at once.

Focus on ROI. Prioritize high-impact decisions. Measure business value. Show quick wins. Build momentum. I worked with a financial services company that spent $50K on a POC and generated $50M in additional revenue in the first year. That's what happens when you focus on ROI.

Plan for evolution. Design for flexibility. Build modular systems. Plan for upgrades. Document decisions. You're not building the final system. You're building the foundation for the final system.

A Real Example

A manufacturing company built decision systems under constraint. They had limited data—only six months of historical production data. They had a legacy ERP system that was hard to integrate with. They had a small technical team. And they had a three-month timeline.

Here's what they did. They used domain expertise from their production team to augment the limited data. They built a simple model that didn't require years of historical data. They integrated with the ERP via API instead of trying to replace it. They focused on production scheduling, which was their highest-impact decision.

The results were immediate. Equipment utilization improved by 15%. Production delays dropped by 40%. They saved $2M in the first year. And they built a foundation for future AI initiatives.

That's what happens when you design for constraints instead of pretending they don't exist.

When to Use This Approach

Use this approach when you have limited resources. Use it when you have legacy systems. Use it when you need to move fast. Use it when you need to show value quickly.

Don't use it when you have unlimited resources and time. Don't use it when you're building a greenfield system. Don't use it when you can afford to take a long-term approach.