Algorithms are getting very good at reading factory data. They are no substitute for the people who understand it.
Walk the floor of almost any food or drink factory in the UK and data is everywhere: line controllers logging cycle times, checkweighers recording every pack, quality results in a laboratory information management system (LIMS), electronic point of sale (EPOS) and depot data flowing back from customers. Some live in the cloud, some on a server down the corridor, and a surprising amount in a spreadsheet only one planner understands.
Very little of it changes what anyone actually does on a Tuesday morning, writes Simon Clark, Founder and CEO at Julius & Clark.
That is the paradox facing FMCG operations teams: we have never collected more information and rarely felt less certain. Demand planning, supply planning, scheduling and procurement are each done well enough in isolation, then stitched together with email, a monthly S&OP meeting and the instincts of a few long-serving people. So when something moves – a heatwave, a promotion that overperforms, a late supplier, a filler below rate – the chain does not re-plan so much as react.
AI is the most credible answer to this in twenty years. But the framing that matters is not artificial intelligence (AI) replacing human intelligence (HI); it is AI bolstering HI. An algorithm is very good at reading eight million rows and spotting the unusual, and hopeless at knowing Line 3 struggles on its first run after a deep clean. Put them together and you have something powerful. Leave either out and you have an expensive disappointment.
Joining up the chain
The real prize is not a better forecast but a faster route from a market signal to a decision on the floor and a call to a supplier.
Consider a chilled ready meal producer: weather-sensitive demand, short shelf life, punishing waste. A traditional forecast leans on last year’s numbers and a planner’s overlay; a machine learning model layers on temperature forecasts, promotional calendars, competitor activity and live EPOS to sense demand shifting days earlier. What matters is what follows: that view flows into supply planning, the schedule and the raw material call-off, so the vegetable order changes before the sleeving film is committed.
This is not theoretical. Danone has run machine learning demand planning on its fresh products for over a decade, reporting a 15 to 20% cut in forecast error, a 25 to 30% drop in lost sales and obsolescence, and planner workload roughly halved, redirected to higher-value work. Nestlé uses AI and causal models to simulate how a demand shift, ingredient shortage or logistics constraint plays out across its value chain, so planners can adjust before a gap appears on shelf. Neither removes the planner from the loop; each hands them a scenario and the reasoning behind it.
Or take a soft drinks bottler with dozens of SKUs across a few lines. Sequencing is genuinely hard: flavour changeovers, allergen rules, CIP windows, labour, transitions costing forty minutes one way and four hours the other. Schedulers solve it weekly with experience and a whiteboard. An optimisation engine can lift OEE by several points and cut changeover time, but it will happily propose something the shift manager knows is undeliverable because engineering are on Line 2 all Thursday.
The best implementations I have seen treat the algorithm as a fast first draft and the scheduler as the editor. The same logic is moving upstream: PepsiCo uses digital twins and AI agents to design facilities virtually, testing capacity and layout before committing capital, and reports cost savings of 10 to 15%.
Quality is where the technology is moving fastest. Vision systems on a bakery or snacks line inspect every unit, not a sample – seal integrity, fill weight, bake colour, foreign body – and close the loop back to the equipment, nudging a depositor or oven zone into tolerance before a pallet goes out of spec. Edge processing matters here: a decision that travels to the cloud and back is too slow at 600 packs a minute.
PepsiCo runs vision systems on its snack lines, Nestlé inspects wrapper integrity and fill level across its factory network, and Tyson Foods grades by computer vision what it once sampled by hand. The gain has the same shape each time: defects caught earlier, fewer manual checks, people moved onto work needing judgement, and a data trail showing what causes quality drift.
Procurement is the fourth piece, usually the last connected. A dairy or confectionery business exposed to volatile commodities can link demand scenarios to raw material requirements, then test them against supplier lead times, price curves and risk indicators. The output is not “buy now” but a clearer view of exposure, so the buyer’s judgement goes to a well-framed question, not a hunch.
Where to start
Four things separate the businesses getting value from the ones running pilots forever.
Start with a problem, not a dataset. “How do we cut short-shelf-life waste by 20%?” is specific. “Let’s do something with our data” is a budget line with no owner.
Be honest about data quality. Master data, BOM accuracy and consistent downtime coding are unglamorous and non-negotiable. AI amplifies what you feed it, including the errors.
Buy before you build. Very few food manufacturers should be developing models in-house. The question is which of the proven demand sensing, scheduling, vision and knowledge platforms fits your process and systems, not which has the cleverest algorithm.
Invest in the people who will use it. A planner who understands how a model reaches its recommendation will challenge it well. One who does not will ignore it or follow it blindly off a cliff.
None of this makes experienced operational judgement less valuable; it makes it more valuable, because it removes the guesswork judgement has to compensate for. The businesses that win will stop asking what AI can decide for them and start asking what it can put in front of the people already deciding. Simon Clark is the CEO and founder of Julius & Clark, an innovation, operations and solutions consultancy helping organisations of all sizes turn ambition into measurable, lasting results.
Comments are closed.