Outside-in planning, explained by Lora Cecere: demand latency, demand sensing and the 80% of data you never use

About Lora Cecere

Founder of Supply Chain Insights and the most-read voice in supply chain research. She ran manufacturing plants for Procter & Gamble, Kraft, Clorox and Nestlé, worked for Manugistics, Gartner and AMR Research, and has published hundreds of reports on why supply chains underperform. She is followed by 345,000 people on LinkedIn. Listen to the episode.

The 80% of data you never use

Here is a number that should bother every planning team: 80% of the data that surrounds a supply chain is never used.

That is Lora Cecere's estimate, and she has spent four decades measuring it. Weather, ratings and reviews, supplier quality data, consumption data, EDI messages: it all exists, and almost none of it reaches the plan. "Our conventional architectures are really very focused on transactional data," she says. "We build our architectures with schema-on-write capabilities, putting everything into rows and columns, and then putting conventional optimisation engines on top of it."

Outside-in planning is her answer to that. This article covers what it is, the one concept inside it that most companies get wrong, and where to start.

Demand latency: the three weeks you don't know you're losing

Outside-in planning starts with the market and works back to the enterprise, instead of the other way round. Lora lists five things it makes possible, and the first is the one with a number attached.

"Demand latency is the time that something happens in the market where someone decides to buy a product and the supply chain sees the order. Most people don't realise that demand latency is at least three weeks and is often three to four months depending upon the kind of supply chain it is."

Every node between the shelf and your order book adds to that delay, and every week of delay is a week your plan runs on stale information. Her Project Zebra research tested the effect of outside-in data on the bullwhip effect and found a 9-10% reduction. "Most people don't measure the bullwhip," she adds.

The other four: manage the network rather than the enterprise; orchestrate demand and supply in both directions, because both are variable; build a planning master data layer that shows variation in lead time, conversion rate and price; and plan by flow, not as one monolithic supply chain.

Demand sensing vs demand planning: different horizons, different jobs

Ask whether outside-in planning is demand sensing with extra data and Lora's answer is a flat no. The two live in different horizons.

"When I think about demand management, I'm thinking about it in the tactical planning horizon, which is outside of lead time. I'm looking at the forecastability of an item. Demand sensing is in the operational horizon when I am in the order fulfilment period, where I'm looking at market data and I'm realigning the supply with demand in that executional period."

So demand planning, in her definition, is where you classify items by forecastability and latency and align the right engine to each pattern. Demand sensing is where you use market data to reshape the near-term plan inside lead time. Both use market data; they use it differently. "A lot of people confuse them, but I believe the respect of the horizons is important."

The practical consequence: a demand sensing engine bolted onto a plan that was built inside-out doesn't fix the plan. It makes a late plan react faster.

Measure the balanced scorecard, not the function

Why does any of this matter to a CFO? Because, in Lora's research with Arizona State University and Georgia Tech, the metrics that explain market value are not the ones planners are measured on.

"If you focus on cost, you'll throw the supply chain out of balance because it's a complex nonlinear system. You really have got to focus on operating margin." Her Supply Chains to Admire work scores companies against their peer group on four metrics: growth, operating margin, inventory turns and return on capital employed. Together they explain 60-70% of market capitalisation in most industries.

The trap she sees everywhere: a company improves a functional metric, such as OEE or transport cost, and makes the whole system worse. "You think you're improving something, but actually you might not be."

Don't AI stupid

Lora has a phrase for the current wave of agents: AI stupid.

"AI stupid is what I see a lot of people doing, where we put agents on top of existing supply chain planning technology and we ask them to do simplistic jobs and we celebrate. That agent may break because of lack of governance."

What she wants instead is a stack built to use the data that goes unused today: unstructured data brought up to a layer where language models, neural networks, machine learning and optimisation can work on it together. The goal is discovery. "How do I help companies not only deal with the known knowns, but the known unknowns and the unknown knowns, because that's really the world we're living in."

And a warning she repeats: "If you're just looking to replace planners, that's a really sad day. This should not be about workforce reduction. This should be about driving improvement and getting better answers at the speed of business."

The planner is an orchestrator, not a number generator

Lora's recent article on the supply chain engineer argues that the planner's job was never to push data through a system. It was to ask whether the plan is good.

"Most of our planners are spending six to seven hours in meetings. It's incumbent on all planners to not only push data through a system, but to ask themselves: do I have a good plan? Is it feasible? Have I architected it to improve reliability? And how do I influence the organisation to use this plan?"

She also pushes back on the way companies measure planners. "It isn't about forecast error, it's about forecast value add." (Our explainer on forecast accuracy, bias and forecast value add covers the arithmetic.)

Your next steps

• Get a cross-functional team in a room and review the last year. Where were the surprises: a launch that missed, a quality issue you didn't sense, a supplier that didn't deliver? Then ask which data you already have that would have shown it. "It is amazing how many insights come out of that kind of dialogue."
• Inventory the data around your supply chain. Lora's students are "often shocked that they have so much data that they don't use."
• Define demand planning and demand sensing separately, by horizon, and stop expecting one to do the other's job.
• Measure planners on forecast value add, not forecast error.
• Read "Staple Yourself to an Order" (Harvard Business Review), then follow one product from supplier to shelf. "Do your own discovery all the way to the shelf."

Lora's closing advice is the shortest version of the whole conversation: "Don't just accept the status quo, but continue to learn and ask hard questions."