Stop Cleaning Your Data. Use AI To Figure Out Which Info Matters
Enterprise data strategy is being challenged by a counterintuitive idea: maybe cleaning and organizing data first is the wrong way to unlock AI value. In a recent article for Forbes, John Sviokla, HBS Executive Fellow and co-founder of GAI Insights, argues that many organizations are over-investing in data preparation while under-investing in finding the actual signals that drive decisions. The article opens with a sharp critique of the common enterprise mindset: “get your data ready first.” While this approach feels safe, Sviokla argues it often delays real AI value and leads companies to perfect datasets that may not even contain useful insights. Instead, the article proposes a shift toward a “signal-first” strategy. The key idea is simple: businesses should first identify which decisions matter, and then work backward to determine what data actually influences those decisions. This is where concepts like expected value of perfect information (EVPI) come in — if better information wouldn’t change a decision, then cleaning it adds little value. Sviokla also highlights that AI itself is better suited to messy, unstructured data than traditional analytics. Customer feedback, call transcripts, sensor data, and other “dirty” inputs often contain richer signals than highly structured but sanitized datasets. To illustrate the idea, it points to companies like Verisk Analytics, which built its business by aggregating real-world insurance and risk data tied directly to underwriting and claims decisions — effectively treating data acquisition as signal acquisition, not storage hygiene. The broader message is a reversal of conventional wisdom: AI shouldn’t wait for perfectly governed data. Instead, it should be used to discover which data is actually valuable in the first place. In this view, clean data without signal is just overhead — while even messy data can be a competitive advantage if it helps improve real decisions.
