What small organisations can realistically achieve, and what they cannot.
Choose a narrow decision
Small teams should begin with one repeated decision where historical examples exist and improvement can be measured. Forecasting everything or creating a general intelligence layer produces scope without a reliable learning loop.
Define the baseline first. A rule, spreadsheet or manual process may already perform well enough, and any model should demonstrate improvement against it.
Build the data path
Assign ownership for collection, definitions, quality and access. Keep training and evaluation sets separated, and prevent future information from leaking into historical predictions.
Use simple models when they meet the requirement. They are often easier to explain, monitor and retrain.
Operate the outcome
A notebook result is not a product. The team needs versioned data and models, repeatable training, deployment, monitoring and a fallback when inputs change.
Review false positives and false negatives with the people affected by the decision. Human feedback and operational impact should guide the next iteration.