[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-body-the-naive-baseline-is-beating-you":3},"\nCompanies spend heavily on forecasting. Software licences,\nanalysts, planner adjustments, consensus meetings. Almost none of\nthem measure whether any of that activity adds value, because\nthere is nothing to measure against.\n\nThere should be. A control group is available, it is free, and it\ntakes a week to set up. Compute the naive forecast, same as last\nyear adjusted for trend, for every item you plan, and measure your\nactual process against it over the two years of history you\nalready possess.\n\nThe published evidence on what this reveals is uncomfortable.\nSteve Morlidge's research in Foresight, the International Institute\nof Forecasters' journal, studied 300,000 real forecasts across\neight supply-chain companies and found\n[52% were less accurate than the naive forecast](https://isf.forecasters.org/wp-content/uploads/gravity_forms/2-dd30f7ae09136fa695c552259bdb3f99/2019/07/Gilliland_Michael_ISF2019.pdf).\nFollow-up work across more than 20 companies found 30 to 50% of\nforecasts routinely losing to naive. Half the forecasting effort\nin industry is spent losing to a number that costs nothing to\nproduce.\n\n## The question is the process, not the forecast\n\nThe field calls the discipline forecast value added, and the name\nholds the insight. The interesting question is not whether the\nforecast is accurate. It is which step of the process improved it.\n\n- **The model against naive.** For products that sell steadily,\n  the statistical layer usually wins. For items that sell rarely\n  or in unpredictable bursts, frequently not. Knowing which is\n  which redirects the effort and the spend.\n- **Each human step against its input.** Did the planner's\n  adjustment improve on the model? Did the consensus meeting\n  improve on the planner? The human half of the chain has its own\n  uncomfortable evidence base ([half the forecast is a person, and\n  nobody scores that half](/insights/half-the-forecast-is-a-person)).\n- **The whole chain against naive.** The final number. If the\n  end-to-end process loses to the free baseline for a third of the\n  portfolio, that third is being planned at a premium price for\n  negative value.\n\nNone of this is a judgement about forecasters. It is a management\nfailure of the ordinary kind. An activity that consumes real money\nruns for years without anyone checking whether it works, because\nnobody installed the control group.\n\n## Run it before the next purchase\n\nAI forecasting claims are everywhere, and accuracy improvements of\n8 to 20% are routinely cited without stating the baseline. If you\nhave never computed your own naive benchmark, you cannot price\nthose claims. An \"85% accurate\" system may be brilliant or may be\nlosing to last year plus trend, and the sales material will not\nvolunteer which.\n\nRun the audit first and the conversation changes. You can demand\nproof on your own hard items, and a vendor must beat your reality,\nnot your anxiety. The same test then governs the system after\npurchase, and the steps that keep failing are retired without\nsentiment.\n\n## Some items should stop consuming forecasting effort\n\nThe most useful finding is often the least expected one. Some\ndemand cannot be forecast better than naive, by anyone, because\nthe signal is not in the data. Those items should stop consuming\nforecasting effort altogether. The right response to them is a\ndifferent operating posture (e.g. faster response and honest\nbuffers), not a better model, and only the control group tells you\nwhich items they are.\n\nContinuous forecast value added, per item and per process step,\nwith the naive baseline enforced as the floor, is standard in\n[the Prophesee Supply Chain Suite](/solutions/supply-chain/demand).\nInstall the control group on your own history.\n[Start here](/contact).\n",1786984937551]