Lora Cecere: "Please do not AI this"
Lora Cecere, Supply Chain Shaman, wrote in her article last year, Please Do Not AI This:
"In this open letter, I beg you not to layer general AI over conventional supply chain planning. The issue? You might ask, why can what seems to sound so right be so wrong? Why am I so passionate about this? I feel that our current processes are broken, and generative AI has the risk of only helping us to make bad decisions faster...............The current state of supply chain planning, as shown in the taxonomy (below), is not sufficient. There is a reason why the most used technology by planners is a spreadsheet."
This is very sound advice given that the current supply chain planning model uses inaccurate forecasts/demand plans (world class 80% forecast mix accuracy hides the fact that c80% of item forecasts are >40% wrong) to generate inaccurate master production schedules that inevitably need expediting to avoid service misses - thereby setting off a variability generating vicious circle of further schedule interventions that explain why such supply chains always co-exist with excessive inventory, lead-time and cost, see SC Variability: what it is, why its bad and how it can be minimised
A little later in the article Lora writes
"I don’t know the answer on how to improve decision support for business leaders to improve supply chain decisions, but I am convinced that if you are honest with yourself, you don’t either. Together, we need to figure out the answer. // I focus on learning from history through research to understand what works and does not.........."
which is a somewhat surprising statement given that replenishment driven by demand (ie. enterprise-wide pull) is the obvious alternative to the conventional (inaccurate) forecast-driven model that she rightly criticises. By aligning material movements with the rate of consumption the demand-driven methodology eliminates the need for performance destroying schedule interventions and, in its perfect form, would provide 100% service with zero static inventory!
领英推荐
Supply Chains however, unlike water, cannot flow perfectly - for instance, some degree of batching is usually necessary and processing hold ups cannot always be eliminated, though Lean activities (eg. SMED, TQM, TPM, Standard Work, Poke-Yoke etc) can help minimise both. Importantly, Supply Chains can now also improve their performance significantly by replicating material flow with the use of software systems (1) that are designed to move materials in line with demand using enterprise-wide pull (ie. Demand Driven MRP) through legacy ERP. Typical benefits (2) are:
You can listen to Lora interviewing Nick Lynch, Global Supply Chain Planning Director at Shell Lubricants, about their transformative implementation of Demand Driven MRP here: Shell Lubricants - Lynch - Cecere or read about it at: Supply Chain Digital: how Shell Lubricants remains ahead of the curve.
And this article explains exactly why the current SCP model cannot operate effectively in manufacturing and why Demand Driven MRP / enterprise-wide Pull (with or without AI) does: Factory flow is non-linear so don't use master production schedules.
Simon Eagle great article as usual. the other part of the problem is the base model is inherently flawed. Lora is 100% correct of course - even near on 20 years ago a client of ours after an i2 implementation said "it sends the wrong message to the right person much faster". Not a great recommendation, and we wonder why all the millions spent on supply chain solution fail. we have adapted the model highlighted to show in a familiar context how to even approach the problem - a bit like the old joke, how would you improve your supply chain? well i wouldnt start from here! Key issue is Planned Orders from an inaccurate forecast are no basis for a supply chain plan. Please just STOP! No wonder the AI doesnt work, it is trying to solve the wrong problem. Keep going!
Building Tech Products | Operations and Supply chain | Theory of Constraint (TOC) evangelist
7 个月Simon Eagle as Eli taught us - objective should be to simplify the system, ‘inherent simplicity’. In depth analysis leads to ‘insights’ which help simplify the system not build one more layer of complexity ! Which in turn creates heaviness in the system and adds to overheads. One more layer gets added to the onion which actually takes one further away from the root cause ‘ insight’. Unfortunately, many managers feel that complex is the only way and outright discard the solutions which are simple but effective in solving the chronic pain. Many of them are blindly embarking on the journey of AI without looking at their effectiveness or relevance toh an area. Getting AI to their resume seems to be the objective. AI is good, but using it the right way is so much more important.