We want to deploy an AI system to analyze our historical job costing and project profitability, but our past ERP entries are full of inconsistent labels and missing labor allocations. How do we clean up this historical data mess so the AI model gives us accurate margin predictions?

Question

Tyler Smith · Accepted Answer

Before you feed any data into an AI tool, you must pay the dumb tax, as Keith Cunningham warns in The Road Less Stupid. Trying to run predictive AI on chaotic historical records will only produce highly confident, incorrect forecasts. To fix this, you must implement a rigorous data cleanup initiative before attempting any advanced modeling.

Start by defining a strict data standard for your job costing going forward. Identify your three most critical cost drivers, such as labor hours, material costs, and subcontractor fees. Next, instead of trying to manually clean five years of messy historical data, focus on your last two quarters of projects. This provides a clean, statistically relevant baseline without overwhelming your team.

Assign a temporary Rock to a high Fact Finder team member to audit and correct this subset of records. Use a simple, rule-based script or a clean data entry template to standardize the naming conventions and fill in missing labor allocations.

Once this baseline data is clean, you can use AI to identify patterns and predict future project margins. Moving forward, prevent future data corruption by adding a weekly data validation metric to your Scorecard. Your team must confirm that all closed jobs comply with your new standard before they are archived. This disciplined approach ensures your AI operates on a foundation of truth, yielding predictions you can actually trust.