AI and Data / IoT Strategy
Most AI strategies fail because they start with technology instead of business problems. Here is the order of operations that works.
Key Takeaways
- 80% of AI projects fail because they start with the technology instead of the business problem.
- The correct order: business problem first, data audit second, model selection third, implementation fourth.
- Before investing in AI, answer: 'What decision are we trying to make better, and what data do we need to make it?'
- Start with one high-value use case that has clean data. Prove ROI before scaling.
A CEO told me he needed an "AI strategy." I asked what business problem he was trying to solve. He paused. "I do not know yet. But our competitors are all investing in AI, and our board is asking about it."
This is how 80% of AI initiatives die. They start with the technology and work backwards toward a problem. The successful ones do the opposite.
The Right Order of Operations
Step 1: Define the Business Problem
Not "we need AI." Instead: "Our sales team spends 40% of their time on leads that never convert. If we could predict which leads will convert before they enter the funnel, we could redirect that time to high-probability deals."
That is a business problem with a clear outcome: reduce wasted sales time by 40%, increase revenue per rep, shorten the sales cycle.
Step 2: Audit Your Data
AI is only as good as the data it learns from. Before building anything, audit what you have:
- Is the data clean? (Accurate, consistent, complete)
- Is there enough of it? (Most ML models need thousands of examples)
- Is it accessible? (In a database, not in spreadsheets or people's heads)
- Is it labeled? (For supervised learning, you need historical outcomes attached to the data)
I worked with a company that wanted to build a churn prediction model. When we audited their data, we found that churn reasons had not been tracked for the first two years. They had usage data but no outcome labels. We had to build the labeling system first, collect 6 months of data, and then build the model. Had they audited the data first, they would have known the timeline was 9 months, not 3.
Step 3: Select the Approach
Based on the problem and data, choose the simplest approach that works:
- Rule-based logic: If the data patterns are obvious and the decisions are binary, you do not need ML. An if/then rule set is cheaper, faster, and easier to maintain.
- Classical ML: If you have structured data with clear features and labels, standard models (regression, decision trees, random forests) work well and are interpretable.
- Deep learning / LLMs: If you have unstructured data (text, images, conversations) or very complex pattern recognition needs, more sophisticated models are warranted.
The mistake is jumping to the most complex option. Start simple. Graduate to complex when simple is not enough.
Step 4: Implement with Measurement
Build the first version in 4-8 weeks. Measure the business outcome, not the model accuracy. An ML model that is 85% accurate but moves no revenue metric is a failed project. A rule-based system that is 70% accurate and saves 20 hours of sales time per week is a success.
IoT Strategy: Same Principle
IoT projects follow the same pattern. The question is not "how many sensors can we deploy?" It is "what decision do we need to make better with real-time data?"
A manufacturing client wanted to deploy IoT sensors across their fleet. I asked: "What will you do differently with the sensor data?" After some discussion, the answer was: "predict maintenance needs before equipment fails." That is a business problem. From there, we could scope the data requirements, sensor placement, and decision model.
Your First Step
Write one sentence: "The business decision we want AI to improve is [X], and the metric we will use to measure success is [Y]." If you cannot complete that sentence with specifics, you are not ready for an AI project. You are ready for a problem definition workshop.