Category Archives: Forecasts

How Do We Harness The Wisdom of Crowds In Supply Chain Forecasting?

As per a recent article in Strategy + Business on cleaning the crystal ball, which discussed the challenges associated with forecasting, we are reminded how the old game of estimating the number of jelly beans in a jar illustrates the innate wisdom of the crowd. In a class of 50 to 60 students, the average of the individual guesses will typically be better than all but one or two of the individual guesses. Furthermore, not only can you not identify the best guesser in advance, but that “expert” may not be the best individual for the next jar because the first result likely reflected a bit of random luck. Then there’s the fact that research by James Shanteau, professor of psychology at Kansas State, has shown that expert judgements often demonstrate logically inconsistent results. For example, medical pathologists presented with the same evidence twice would reach a different conclusion 50% of the time.

However, teams of forecasters often generate better results (and decisions) than individuals as long as the teams include a sufficient degree of diversity of information and perspectives. This is partially because a naive forecaster often frames the question a different way and thinks more deeply about the fundamental driver of the forecast than an expert who has developed an intuitive, but often overconfident, sense of what the future holds. But you can’t just throw a group of people together in a room and ask them to come up with a consensus because the most vocal or senior person might dominate the discussion and overly influence the consensus because most people put too much confidence in the most senior or highest-paid person.

So how do we harness the wisdom of the crowds and insure that no one voice dominates the forecast when the forecast is inherently risky and unpredictable? We look in the place that we are probably most familiar with — e-Sourcing. A blind RFX survey sent out to an interdisciplinary team of carefully chosen and randomly chosen individuals who, given a scenario description, past sales, and expected market trends (from third party analyst firms) are asked to provide their input to the short-, medium-, and long-term forecasts at the SKU and group level. Then, we simply average all of the responses, giving slightly higher, but individually equal, weighting to the carefully chosen respondents (who collectively complete an interdisciplinary team and do it as part of their jobs) and slightly lower, but individually equal, weighting to a section of random organizational individuals asked to weigh in with outside opinions. It won’t be perfect, but it will be substantially better than all but a few guesses — and since you won’t know what guesses will be good or bad in advance, it will substantially reduce your risk.

Thoughts? Comments? Criticisms?

Share This on Linked In

Pre-Defined Forecasting Models — Are They Worth It?

It seems all of the inventory management and service management platforms these days are touting the myriad of forecasting models in their latest and greatest release, but are they worth it?

It’s a hard question to answer. Here are some of the main arguments from both sides of the fence. You be the judge.

Supporter: You should check out this great new software for inventory and service management (SISM) from Hyper Yard Product Engineering (HYPE). It supports time series, casual, and combined forecasting models that allows you to define forecast types, horizons, target replacement rates, smoothing constants, variance parameters, seasons and trends, target costs, target fill rates, thresholds, and a slew of other parameters. It will save your buyers a lot of time, experts and novices alike.

Detractor: I don’t know. Sounds like a bunch of hoopla to me. Not only do most “experts” believe that only their models are the right models, they are likely to be suspicious of the results unless they can tweak every parameter and analyze the assumptions and algorithm in detail. And even then, if the resulting curve doesn’t adhere to their intuition, they are likely to call it a pile of junk and override it with a manual forecast. As for novices, it’s not very useful either as they aren’t likely to have the competence necessary to understand the inputs or outputs or the reason why one method is preferable to another for a given commodity or category and will likely choose the wrong method and screw-up the parameters.

Supporter:Oh no, not hoopla at all. It’s extremely useful to both parties. It’s a great help to novices as it automatically selects the right model given the commodity or category, sets the parameters based on contract data and historical inventory patterns, and correlates to similar commodities in the system. It greatly increases the accuracy of their results. And since it provides dozens of algorithms, where each parameter and assumption can be overridden, experts will love it as it will contain their algorithm, allow them to tweak the parameters, and override results with a manually derived model where needed.

Detractor:So, the expert can override the system at any time and the novice doesn’t have to choose anything?

Supporter:Correct. An expert can even define a fully manual forecast if the situation is unusual (such as a one of a kind promotion or rapid uptick in the economy) and a novice doesn’t have to do anything.

Detractor:So, the novice gets a free ride until the black swan decides to attack and cause a major catastrophe, and the expert gets to bypass the system entirely. In the first case it’s misused, as no system can compensate for human intelligence and every forecast needs a sanity check, and in the second case it’s not used at all. Sounds like a waste of money to me.

Supporter:No, no. You miss the point. By automating the process, it gives the user time to focus on learning the category so that she can tweak the parameters when she notices an unusual uptick or downtick and so that she can focus on analyzing the forecast and performing the sanity check instead of wasting hours or days trying to manipulate formulas and software to produce the forecast. And since it has so much power, the expert won’t need to override with a manual forecast very often.

Detractor:In theory, yes, but in practice no. Human nature is what it is. Most novices are tactical procurement personnel who don’t understand the math behind forecasting and don’t want to. They’ll drive whatever forecast the system spits out until they drive off the cliff. And most experts are arrogant know-it-alls who don’t believe a dumb system can ever come close to their years of experience and massive ego. They’ll tweak or override every forecast even if there isn’t a compelling reason and even if the forecast is probably right (because only they can be right) and tweaking it makes it wrong. Both ways, the company loses. Not only does it lose the investment in the system, but it loses out when it gets stuck with excess inventory or fails to meet demand.

Supporter:Well, you have to use the system properly to take advantage of its power.

Detractor:Right. And neither party will. Furthermore, if forecasting really was a science and if these tools always worked as advertised, the world would be an easy place. But the reality is that most demand is impulse and spike driven and most forecasting is nothing more than dumb luck in the end. As a result, all those continuous and curve fitting algorithms turn out to be worthless as they can’t model spikes. The reality is that you can get as much accuracy from a kindergartener’s hand drawn curve, especially if it looks “close” to historical demand curves.

Supporter:That’s why these tools also have extensive comparative graphing capabilities that allow you to compare current forecasts against past behavior for similar horizons. You can see whether or not the forecast is in line with what normally happens and whether or not there are any spikes consistent with your expert’s “gut feeling”.

DetractorThat’s nothing more than warm fuzzies, and warm fuzzies mean nothing. It’s as useful as Wall Street’s “risk” algorithms. Heck of a lot of good they did us.

Supporter:That’s why the tools allow you to update the forecasts at any time to take into account the latest changes in demand.

DetractorSo what? Spikes are, by their very nature, unpredictable. All you can do is “smooth” the curve to the new data. An unexpected spike is an unexpected spike is an unexpected spike and you lose millions in sales because of it. Just ask Apple or Nintendo or Sony who seriously underestimated initial demand for a number of their recent products.

Supporter:Look, we both know that no tool will solve all your problems and that spikes only occur in a few categories. For example, for many consumer goods, and many consumables in particular, demand is relatively constant year after year. Some brands or models may sell a little more or less than predicted, but most categories of household goods rise steadily with the population. We’re not going to suddenly start washing our clothes twice as often, start drinking twice as much coffee a day, or buying twice as many CDs. You need to start somewhere, and a tool that’s likely to get most of it mostly right out of the box is a great time saver. Plus, the best way to detect a spike or a drop is to instantly be able to see whether or not demand is following the plan. That’s the true power of these tools.

DetractorMaybe, but if that’s the case, what do you need all the advanced modeling for? Wouldn’t a simple time series with the ability to fully define a forecast by hand be just as good?

Share This on Linked In

Five Common Inventory Management Mistakes from Demand Solutions

Demand Solutions recently released a white paper on “managing inventory for optimal advantage” that overviewed 10 common inventory mistakes and how to correct them. Of these, the following five can cost an organization dearly if not corrected.

  • Forecast Management without a Process

    All stakeholders have to agree on the process and the forecast that results and someone needs to own the process to insure it’s implemented properly. Otherwise the budget will be padded and the end result will be obsolete inventory and associated losses.

  • Not Talking to Customers

    Good inventory management is more than just the right volume, it’s the right volume at the right time in the right place. Be sure to understand what is driving customer replenishment patterns to insure that production is synched to customer needs. Otherwise, inventory can build up for months at a time, which will incur additional storage costs.

  • Forcing the Budget

    Don’t overlay the budget on top of the sales forecast. Both are approximations and both need to change to reflect reality. Attempting to synch them will result in production patterns that don’t match actual demands.

  • Too Many SKUs in Too Many Places

    This greatly decreases warehouse efficiency and increases fulfillment costs.

  • Never Trying New Things

    New technology provides better capability for ongoing, collaborative improvement. Avoiding new technology will limit operational efficiency and cost savings opportunities.

Share This on Linked In

If You Want to Control the Bullwhip …

… and its effects that can be very detrimental to your inventory levels, associated costs, and overall supply chain revenue, don’t do any of the following:

  1. Second Guess the System and Overcorrect

    This typically happens when a buyer orders early, thinks he needs more “just in case” inventory, or believes that the system has under-estimated inventory requirements. When this happens at each stage of the supply chain, the original order requirements end up increasing significantly. For example, if a buyer at a retail store adds 10%, then a buyer at the local warehouse adds 10%, then a buyer at the central warehouse adds 10%, and then the distributor adds 10%, the supplier will get an order for 146.4% of the original order volume and the supply chain will become saturated with excess inventory.

  2. Last-Minute Unplanned Promotions

    This goes for both buyers and sellers. Buyers, don’t allow marketing or sales to do last-minute unplanned promotions that were not taken into consideration during the forecasts. Without re-running all the forecasts, you can’t know how much more inventory you’ll need, and you’ll over-order “to be safe”. Furthermore, this surprise over-order will cause bullwhip second-guesses up the chain. Suppliers, don’t offer last-minute enticements to get a buyer to buy more. The net effect will be that your buyers will have too much stock, and then drastically cut their orders next time around. These unexpected cuts across the board will result in distributors overcorrecting downward, and then there won’t be enough inventory in the system and sales, and revenue, will be lost by all.

  3. Tweak the Order

    If you have a good, modern, forecasting and inventory management system that can make use of all of your historical data, multiple forecasting algorithms, and run multiple what if scenarios that can take into account multiple assumptions, then, as long as the forecast and inventory plan was generated by a seasoned pro, on average, it’s going to be much better than anything your gut tells you. Tweaking just leads to uncontrolled overcorrections throughout the supply chain.

  4. Increase the Batch Size

    Just because you can get an additional volume discount on order volume or shipping doesn’t mean that you can arbitrarily increase the batch size without consequences. Total cost of ownership, which will have been minimized by your inventory management or strategic sourcing decision optimization system, can involve dozens of variables. For example, there’s the inventory storage cost which could exceed the volume discount, especially if your warehousing cost is high. If you’re currently at FTL, it might put you over to FTL and LTL, and the LTL costs could be much higher. And, of course, the over-order will be followed by an under-order, which could lead to two devastating over-corrections by your distributor who was unprepared for the large swings up and down in order sizes.

Share This on Linked In

MCA Solutions – Bringing the Aftermarket Forward, Part II

In Part I, we re-introduced you to MCA Solutions, a Philadelphia, PA company that specializes in after market service (and service parts) optimization, and noted that they were still going strong despite some recent shake-ups in the market (and the noteable acquisition of Servigistics and Click Commerce by Marlin Equity Partners, who also acquired Emptoris not too long ago). We noted that, in addition to completing a strong SAP integration, they’ve also added a considerable amount of new functionality in the last two years around reporting, plan analysis, and reporting management.

Since we covered their new reporting and plan analysis solution in the last part, today we’re going to cover their performance management solution. Since you can’t manage what you can’t measure, and the best way to measure is often with a balanced scorecard, it’s based on scorecards, but since managers don’t like columns of numbers, it’s implemented using a dashboard, but since MCA agrees with me that traditional dashboards are inherently dangerous and dysfunctional, they realized that the only way the application would be truly useful was if it clearly identified not what was right, but what was wrong (since a goal of after-market service is exception-based management so that you only expend resources where needed). More importantly, the scorecard dashboard would only be useful if it allowed you to quickly discern what was wrong and do something about it. So what MCA built is a dashboard scorecard that not only highlights any metric that is out of bounds in red, but an interactive graphical scorecard that allows you to drill down into the metric retrieve all of the data associated with that metric in a single click.

Just like you can drill into a spend cube, you can drill into any metric on the scorecard. The first level drill will bring up all of the metrics the high level dashboard is composed of, and highlight which metrics are a problem. You can then drill into those metrics and bring up all of the associated raw data. So, if you brought up the scorecard and saw on-time delivery was only 80%, when anything under 90% is unacceptable, you could drill in and see the problem ports are LA and New Orleans and that San Diego, Washington, Vancouver, Boston, and Halifax were all meeting or exceeding their on-time delivery targets. You could drill in again and see that at these ports, most of the late deliveries were from West Coast Warblers and East Cost Easies and instantly know that either these suppliers have performance problems or that you’re not allowing them enough time in your inventory network design to transport the parts require to replenish your North American stock from your foreign suppliers. But since you can also drill into the application and the underlying model associated with any part, location, or supplier you can quickly determine if it’s a performance problem or a network design flaw. For instance, lets say you only allow 14 days for replenishment of goods in your LA warehouses from Shenzhen. Considering that sailing time is typically 12-15 days, and that it probably takes at least a day to get your goods unloaded at the port, and another for them to clear customs, get loaded onto the truck, and transported to your warehouse, there’s no way you’re going to get that part in less than 14 days by sea and it’s probably going to take at least 17 days on average, especially if these carriers are running slower ships. Then you know you need to adjust your model, and measure the supplier against a more reasonable delivery time. But if you are allowing 21 days, and your third party carrier is consistently late, then you have a supplier performance problem.

Moreover, the scorecard dashboard is completely customizeable. Each component is actually a dashboard report, and with their new flexible reporting capability, you can build any report you want. So you can design the dashboard to focus only on reporting problems. That way you can ignore the 90% of your network that is running smoothly and dive right into the 10% that isn’t running right, analyze the situation, revise the model, analyze the revision, implement an improvement, and see if the situation improves over time. If not, you can dive right in and try again. And if everything looks too good, you can define more metrics, more sanity checks, and find new problems to work on. Which is precisely what an actionable scorecard should allow you to do!

And your suppliers in China and Japan can use it too. The product is double-byte Unicode compliant and, in addition to a number of European languages, has also been translated into Mandarin and Japanese. With these recent improvements, you should be able to plug it right into your follow-the-sun operation and, once it’s configured and your data is complete, close the loop on your end-to-end after market service (parts) operation.

Share This on Linked In