Containers from Asia take approximately 2-4 weeks longer and costs are more than doubling.

no paywall https://archive.ph/2XA93

As it is common for the buzzword of the year in supply chain - a lot of noise, but very little substance.

As data were a byproduct and at best second concern for the businesses since the first introduction of ERPs and other enterprise software, it should not come as a surprise that the vast majority of data analytics and digital transformations fail. GIGO hits you when you don't know what are you looking for. Decisions, not the data should be primary focus when you start. Then it helps to narrow down your focus on improving only the data that matter.

Yes, continuous improvement is the key to make a new technology become economically efficient, but also it may become enabler as you've mentioned. Though, by breakthrough we usually mean a change of principle how the same objective is achieved better/faster/cheaper etc. This is qualitative change in the first place. Then continuous improvement starts.

If we look at other fields we can realize safety stock will never get buried completely.
We have modern medicine, but healers still do exist.
We have modern agriculture, but growing 'organic' movement (which is in fact less sustainable).
Etc, etc, etc.
Those advocating for "supply chain is not for tech, but for people" will always seek for dysfunctional but simplistic recipes.

Fully agree. The only case which might be relevant for "real time" data is dynamic repricing for SKUs pending stock out or promotion/discounting within the day for perishable goods. This might be relevant for e-groceries and other e-commerce businesses with tight competition where they closely tracking each other and have overlapping highly substitutable assortment.
For replenishment and purchasing "real time" data is overkill and waste of compute power. Daily batches are enough.

Probabilistic forecasting should be considered as state of the art for supply chain planning.
Though, as we recently have shown it is doable to explain supply chain decisions optimization with use of probabilistic forecasts (we explained it using purchasing decisions) even in Excel:
https://www.linkedin.com/posts/alexey--tikhonov_supplychainmanagement-probabilisticforecasting-activity-6962747294081298432-T-OH?utm_source=share&utm_medium=member_desktop
If there is a substance behind Demand sensing - looking forward to see the Excel with simplified approach showing what value this approach adds.

Demand sensing article on Wikipedia:
https://en.wikipedia.org/wiki/Demand_sensing

Compare it with probabilistic forecasting for instance:
https://en.wikipedia.org/wiki/Probabilistic_forecasting

Q5: There is a comment in the file recommending using this approach for educational purposes only, because such a forecast will be simplistic and will not account for seasonality or trends etc. How to use this approach beyond educational purposes for real inventory optimization problems?

A5: First, it should be noted that probabilistic forecasts in the file are synthetic probability distributions that are built using Excel functions.
It is possible to build probabilistic forecast that would be based on historical sales data and would take into account all systematic causes of variation like seasonality, trend etc.
We intentionally left it behind the scenes for two reasons:
1) We would have to provide historical sales data that probabilistic forecast would be based on. Here we have a problem that all users are from different industries and got accustomed to see different sales patters. We wanted to avoid "this doesn't look like my sales data" kind of impression.
2) Building production grade probabilistic forecast in Excel is technically speaking a nightmare and an open problem since we've never seen anybody done it so far. Taking into account Excel capability limitations it should be also very slow to rebuild the forecast once you update the data.
At Lokad we use domain specific programming language called Envision where probabilistic forecast can literally be build with several lines of code. Those who are interested about the procedure can play the recording of our past supply chain lecture:
https://tv.lokad.com/journal/2022/3/2/probabilistic-forecasting-for-supply-chain/

Q4: Can this approach of decisions prioritization with expected ROI be used for allocating scarce products from DC to stores?

A4: Absolutely.
This file was about purchasing optimization, but the same approach could be applied for DC to stores dispatch problem. ROI ranking perfectly works when there is limited stock at DC and inventory manager wants to make sure that dispatch plan is optimal.
The way it works is the first unit of the first SKU (can be arbitrarily chosen, but can also start from top sellers) and ask in which store (given current stock in each store and probabilistic demand forecast for this SKU for each store) it will have the highest ROI and/or fill rate gain? Then you rinse and repeat for all SKUs at a unit level for all the units (same as with purchasing). This enables to best allocate the existent stock on hand across all the stores in the network. Ranking procedure here is a bit different because at every step we make final decision to which store we send Nth unit of a given SKU. At the end of the process after considering all micro level decisions we chose termination criteria (no more stock at DC or target ROI or target fill rate etc.) and aggregate numbers per store per SKU to get dispatch lists.

Q3: What goes into engineering economic drivers? How critical it is to get them right from the very beginning or is it more like learning curve?

A3: Decision making using economic drivers in combination with probabilistic forecasts to some extent imitates human level decision making process. Costs and revenues are straightforward, but the stock out cover is not. It is completely context dependent and can be called heuristics. This is where human intelligence lies. Having them properly set enables to automate decision making so that there is no more need to say split 500k SKUs among 50 demand planners to manage 10k SKUs each. All decisions for all SKUs can be generated automatically by the single numerical recipe. Litmus test here is simple - if demand planner sees no insanity in decision recommendations, then it should be concluded that economic drivers are production grade, but for sure can be fine tuned over time.

The blog post talks about range forecasts and probabilistic forecasts. Also they talk about relatively small number of scenarios ("at least 5-10...") so it brings the question of accuracy with such small resolution.
Feasible decisions ranking and economic drivers are absent while they are essential components for choosing decisions with highest return on investment.
Level of details is underwhelming so far.
Though this is great news. They will definitely accelerate the adoption of the term.

Another metric of this kind would be stock levels expressed in dollar value or in days of sales. Very easy to cut and show "benefits", especially if you are in long lead time business. Cut them, update your CV and make diagonal jump to another business. Profit :)

One of the ways bureaucracies are ignoring difficult problems is by division of labor. Once the scope of the problem is broken up across multiple roles it becomes nobody's business. Calling cross-functional teams to rescue the business is not a solution as usually those teams are composed of multiple within-silo thinkers who can't abstract from their role and look at the problem as a whole.

Q2: After we made and executed purchasing decision there is no more uncertainty, then how would we evaluate whether the predicted probability forecast is the correct one?

A2: Even after we made and executed the decision in reality there is still uncertainty. We don't know when our order will be received because lead time is also probabilistic. Quality of goods can be probabilistic as well (consider fresh food for instance). Also we will not be able to estimate accuracy of our probabilistic demand forecast until the responsibility window (over which we built probabilistic forecast) becomes the past.
With respect to the metrics that are used for probabilistic forecast quality assessment you can check this lecture:

https://tv.lokad.com/journal/2022/3/2/probabilistic-forecasting-for-supply-chain/

Use time stamps to watch the lecture parts where the metrics were discussed.

This is brief summary of my Q&A with one of the users of Excel file. I will provide each question and answer as separate comment.

Q1: How are we creating the probability distributions for the products in this file? If I have historical sales data for a particular SKU, how should I get the probability distribution for that?

A1: We haven't made capabilities to build probabilistic demand forecasts using historical data in this Excel file. This is very hard to implement in Excel due to its limitations. Programming language like Envision is more appropriate for that, but one could also use Python or any other language. How to build Probabilistic forecast with historical data was discussed in this lecture:

https://tv.lokad.com/journal/2022/3/2/probabilistic-forecasting-for-supply-chain/

This file uses synthetic distributions via built-in Excel functions for normal and negative binomial distributions. Changing parameters you can change the distributions and the ranking of micro level decisions respectively. This is educational tool and the primary goal was to show how having probabilistic demand forecast and economic drivers demand planer can optimize purchasing decisions.
Though, the question of building probabilistic forecast based on historical data still remains valid. It is just not in the scope of this document because it is harder to understand, but also harder to explain and show through Excel.

Making noise and making money out of the noise, especially consistently, is not the same. I think that growing number of companies that adopt probabilistic perspective through appropriate software is the only adequate metric that tells anything about the adoption of the term.

That's is interesting. At some point there will be very few aircrafts of this model so servicing parts for them will become nightmarish as demand will be even more sparse than it is now meaning the parts rotation will slow down and carrying cost at some point will make it unprofitable for those remaining aircrafts to fly.

I have seen this so many times while working in FMCGs. Some people made that with multiple companies in sequence and landed CXO positions.

1-1-1-1
Looks easy-peasy. Where did they found so many well educated people who can't add numbers?

Let's show that supply chain practitioners can add.
Post your result in comment like X-X-X-X where X is either 0 or 1, where 0 means incorrect answer and 1 - correct one. So 1-1-0-0 would mean that only first and second questions were correctly answered.

another branch of discussion on LinkedIn
https://www.linkedin.com/feed/update/urn:li:activity:6966330168063725568/

Excel is the Swiss army's knife of any supply chain practitioner. While it is definitely not the most appropriate tool for managing supply chain it is important to be able to convey ideas through it. Lokad tried to build simplified educational version of decisions optimization with probabilistic forecasts in Excel. See the LinkedIn post by the link.
You can ask to receive the file in the comments either here or on LinkedIn under the post.

Great invention! A barcode can be seen as grandfather of modern QR-codes. Its introduction initiated series of inventions where for the same basic idea inventors just added new dimensions. For instance, regular QR-code can bee seen as two dimensional counterpart of a barcode. But inventors didn't stop there. Somebody added third dimension via color coding. It is interesting where this trend will end and how many dimensions can be added to flat QR-code?