Operationalizing Analytics in Forestry

Editor's Notes

• #3: Really, operationalizing analytics just means making O.R. and other analytics technology more accessible to people who are not subject matter experts so that they can do their jobs better. The way to achieve this is through the effective combination of a Decision Management approach with a robust, modern analytic technology platform. Such a combination focuses analytics on the right problems and effectively integrates analytical results directly into operational systems for faster and more profitable decisions.
• #4: In terms of requirements for meeting the definition of operational analytics we have 3 items. Collaboration - In the forestry sector, we are well versed in collaborative planning environments because the forestry problems have always been large in scope (large geographic areas, long planning horizons, lots of data). Big data may be the current fashion but foresters have dealt with big data for decades. Reliable architecture – The proposed solution is based on Remsoft’s flagship Woodstock software that has been around for 20 years, has been deployed on six continents by over 200 clients managing a 500 million acres of forest land. I’d say that’s a pretty robust platform. Repeatable, industrial-scale process – This is new to Remsoft. For the past 20 years, we have focused on developing powerful, flexible stand-alone applications that can address a wide range of problems, and we’ve been quite successful. Application flexibility is important when you’re trying to appeal to wide audience and organizations are considering different alternatives, including yours. However, once your application has been adopted, flexibility becomes far less important as standards and business practices come into play, models become standardized and updating them efficiently becomes important. This is particularly true as planning horizons shift from decades, to years, to months to weeks, resulting in very short lead-times for making decisions.
• #5: So, in order to devise a repeatable industrial scale process, you need real data systems, you need to engage less technical users in the process without requiring modeling or OR expertise, you need to make the interfaces accessible (that is, users change a few indicators - they do not have access to EVERYTHING), you need to make the thing fast enough to be useful, and the underlying model needs to be something that can be adjusted relatively easy (no software application recoding). 20 years ago when Remsoft was first getting started, this type of solution was simply not possible. The technology just wasn’t there to support it. So why is it possible now?
• #6: The importance of computer technology to operations research and analytics cannot be overstated. In terms of how computer technology relates to problem solving, one of my professors years ago used the following simple analogy: Most real problems have a time component and they have a detail component, and you’re supposed to make some important decisions. The length of time you’re planning for and the amount of detail you want to consider determine the size of the model you want to build. However, available technology imposes a constraint, and it is generally true that what you want is always more than what you can have.
• #7: If the area of the box in the previous slide represents the maximum problem instance technology affords us, then we can adapt the instance to our needs. If a long planning horizon is important (say for sustainability measures) then we are forced to reduce detail. Conversely, if we have to consider a lot of details, we need to shorten the planning horizon commensurably.
• #8: Luckily, technology has been improving over time, and using our analogy, it means our boxes are bigger now. Computers are faster and access more memory, solvers are much faster due to things like interior point methods and improved mixed-integer algorithms. We still have to make compromises to fit the box, but they aren’t as onerous as they were in years past.
• #9: Over the last 20 years, Remsoft has issued almost 700 licenses of its forest management software technology. Granted, not all of these licenses have been maintained due to company failures, mergers and other changes in the marketplace This slide shows how Remsoft has changed and adapted over time, in terms of license share by application, geography and organization type of the licensees. Look at the top graph. Woodstock remains the flagship product of the company and accounts for over half of all licenses issued to date. But if you look closely Spatial Optimizer (formerly known as Stanley) accounted for 40% of licenses issued in 1997, and almost all were in the United States. The reason was that in 1994, the American Forest and Paper Association adopted the Sustainable Forestry Initiative, which obligated member companies to impose spatial constraints like adjacency and maximum opening size on their forestry operations. Spatial Optimizer provided a means to address these spatial constraints, but to use it, you had to have a Woodstock harvest schedule and so the company got a toehold in the US market.
• #10: In the middle graph you’ll notice that sales in Canada grew tremendously after 1997 when provincial ministries of natural resources adopted Woodstock; Canadian licenses to government agencies and forest products companies remain the largest component today. But you’ll also notice that market share in Australia and New Zealand increased after 2000, after Remsoft moved to the Windows platform while FOLPI remained DOS-based; this was a major factor to Remsoft adoption in that part of the world. In the bottom graph, you can see what many have called the Rise of the TIMO… during that time market share took off as U.S. large integrated forest products companies divested their lands to the investor class who lacked legacy planning systems of their own but wanted the stability of commercially supported software.
• #11: Returning to the top graph, you’ll notice that market share of Allocation Optimizer coincided with sales in South America (the light green in the middle graph). These new users wanted to be able to schedule deliveries to particular mills along with the schedule of harvests, and so Remsoft developed the AO module. As crew scheduling started becoming important, Remsoft released the Regimes module in 2008, and merged it with the base Woodstock product for new licenses in 2009. In the top graph you’ll see sales of Analytics apps starting in 2011, and these applications have been key to the recent move into the European market in the last 2 years. An important point to make that is not evident in these graphs is that when Remsoft has moved into new areas, it has usually resulted in new functionality to the modeling platform, or more recently in the development of new Analytics applications. However, once introduced, these new features have been well received by existing clients as well.
• #12: As I said, Remsoft licenses software to over 200 clients on six continents. On this slide you can see the breakdown of licensing according to the geographic location in the top pie charts and the type of client in the bottom pie charts. By modeling platform I mean (Woodstock, Allocation Optimizer, Spatial Optimizer) and by Analytics Apps I mean (Tactical Planner, Model Explorer and Analyst for Excel). 89% of current modeling platform licenses are in North America and Australia/New Zealand-these are Remsoft’s most mature markets. TIMOs, government agencies and forest products companies each comprise about 25% of licenses for modeling platform applications. The Analytics apps have only been available since 2009 and they have been particularly popular among TIMO clients. The org charts of these clients are flatter than traditional corporations and so decision-makers have more interest in the details that make up forest planning models. That said, 25% of the licenses are with government agencies and the software is gaining traction as it becomes more well-known. Over 90% of the Analytics apps are licensed in North America, but European clients are already 5% of the total.
• #13: So over time we have had big improvements in technology along with a software company that has learned a lot from its clients around the world over 20 years. So now let’s talk about the work Remsoft has been doing with Coillte, the Irish Forestry Board. Coillte is a state-owned corporation charged with managing public forest lands on behalf of the Irish people. It manages over 400,000 ha of forest land, comprised mainly of lodgepole pine and sitka spruce, but comprising many species from around the world. The land holdings are geographically dispersed throughout the country. Coillte operates 2 pulp mills and an OSB mill, and also supplies forest products to 16 or so other private mills, plus exports.
• #14: Remsoft has been working with Coillte for the last two years, first to implement strategic forest estate planning and more recently to address more operational planning problems. The slide illustrates one of the problems that Coillte is trying to address using improved analytics. Right now, these tasks are being handled manually, largely using spreadsheets, and their goals for this project are three-fold: 1. Improve on a manual and time consuming process Narrow scope of plans Low agility – difficult to respond Personnel limits 2. Breaking down the silos among the production planning, delivery planning and sales planning managers Misaligned plans Low transparency Lost opportunities 3. Be more in sync with strategic plan Further plan misalignment Increased business risk Solving this type of problem is an instance of operationalizing analytics.
• #15: What is listed here are the major milestones Remsoft has achieved since inception. The items in bold text represent capabilities needed to address Coillte’s operational planning problem. These include A robust modeling platform that can adapt to many types of planning problems (Woodstock) The ability to model harvest decisions as well as decisions about where to deliver harvested products (AO module) The ability to represent sequencing decisions such as harvesting felling units over multiple planning periods A high performance mixed integer solver is required because crew allocations and harvests are binary decisions Facilitators to draw the linkages among the various analytics models being employed. A means to pull data from corporate databases, populate a foundation model and then after solution, push data back A means of exposing the underlying model details to end-users who are decision-makers not modeling experts And of course, computer platforms that offer fast processing and large memory spaces to host large analytical models and distributed databases While some of these technologies have been available for a long time, some have only become available in the last few years. So let’s look at how the pieces come together.
• #16: The problem essentially is to minimize harvest and hauling cost by allocating crews to compartments such that weekly volume demands of Coillte’s customers are met, crew and trucking capacities are not exceeded and compartments are completely harvested before moving on. The planned approach is to formulate the problem as a mixed-integer programming problem using Remsoft’s Woodstock and Allocation Optimizer (AO) software, paired with the Gurobi solver.
• #17: Coillte is contracting with another service provider to develop a forest information system that will track land records, provide wood procurement/log accounting, mill yard inventory systems and other corporate database systems for tracking weekly customer demands and prices, tracking deliveries to mills, and issuing work orders to harvest and trucking crews. Because these systems are housed in different databases, processes are needed to bring all the information together in one place, called a DataMart. This shared data warehouse contains all the information required throughout the Operational Analytics solution, and is populated via business processes that source both Coillte's corporate database system and the operational model(s). The DataMart isn't a database we provide them that queries other systems, but rather a schema and set of processes Remsoft helped Coillte develop that enables a systemic approach to data management. With other clients, the DataMart will be different since it is a result of an elicitation process: ultimately it might be one database, a series of databases, or a cloud source - and both the client and Remsoft need to populate it. Remsoft’s Integrator technology will be used to provide dynamic, bi-directional data transfers between the optimization engine and corporate data systems via the DataMart.
• #18: A standardized Woodstock foundation model was devised that incorporates Coillte’s corporate naming conventions for felling units, harvest crews, harvest methods, customer IDs, etc. within their database systems. Using Remsoft’s Integrator technology, this Foundation Model can then be updated with the most recent prices, mill demands, crew capacities, and available felling units from the DataMart to generate a new planning model that simultaneously assigns crews to felling units and schedules the delivery of products to various mills. The template model is important because it incorporates overall corporate goals that cannot be changed by field managers. While they will have access to volume and revenue outputs to help with their decision-making, they do not have the ability to violate constraints linked to corporate standards. Instead, only a fraction of the model is exposed through the Operational Planner.
• #19: Once the initial model is solved, the results will be automatically published and made available to regional managers for verification and/or modification through a Remsoft Analytics application. Using an Excel interface, managers can examine the crew schedule (via tabular lists or Gantt charts) and make any needed changes to the schedule (e.g., change the crew on a block, change the cutting method on a block, advance or delay the harvest of the block, drop a block or crew completely for that time period, or change destinations for products. Any changes are saved and posted back to the DataMart. Presumably, if the changes are minor and do not violate any constraints, the schedule can then be uploaded to corporate servers for issuing work orders. Otherwise, if constraints are violated, the model will need to be updated with the changes, and re-solved before final verification and posting back to the DataMart.
• #20: Excel was chosen as the interface for several reasons. First, it was familiar to all the managers who will be working with the system. Second, Analyst for Excel provides access to all cost and yield coefficients as well as all performance indicators in the base model. Coillte can design its own Excel templates to display the most relevant information to its managers, and these can be changed in-house, unlike a customized desktop application. Users interface with the schedule in different ways. In a tabular view of harvest units scheduled for harvest, a user double-clicks a particular harvest unit to activate a dialog box that lists the harvest method, crew assigned, the period when harvest is to commence, expected product volumes and where they are to be delivered. He or she can change any of these items, and the impact of the change is reflected on an overview sheet of KPIs. Users can also interact with the schedule by double-clicking bars in the Gantt chart, volume delivery histograms or crew tabular lists, depending on his or her preference.
• #21: After 20 years in business, technology advances along with Remsoft’s application development have finally progressed to the point where these requirements are now possible. A collaborative environment based on Remsoft’s Woodstock technology forms the core of the proposed solution. This technology has developed over 20 years and is deployed around the world. The proposed solution will integrate directly with Coillte’s corporate database systems providing near-real time updates to prices, customer demands and crew capacities on demand. Using a familiar Excel interface, the Operational Planner allows managers to view and modify schedules, post the changes and to re-optimize the schedule without need to understand the underlying foundational model. Because the foundation model is built using standard Woodstock syntax, Coillte can leverage their forest planning staff who use Woodstock for strategic estate modeling if changes to the foundation model are required; there is no need for recoding of applications themselves.
• #22: Remsoft started out providing strategic forest planning tools. By and large, our interactions with clients were mainly at the technical level: introductory training for planning foresters, supplemental training, and ongoing technical support. Dealings with upper management were largely questions of signing authority and purchase orders. But over the past 20 years, there have been wholesale changes in timberland ownership, and along with globalization and increased competitiveness, the business of forestry is scrutinized by management much more than in years past. Our clients are trying to leverage the tools they have to eke out savings in the supply chain wherever they can, and as a result, they’ve turned to Remsoft for help. As a result, Remsoft created a formal Client Services group within the company to not only deliver training and technical support but to actively work with clients on implementing new solutions.
• #23: Going forward, the evolution of Remsoft from software vendor to solutions provider is likely to continue. In some ways, the recent media attention on big data and analytics is amusing because these latest shiny, new things don’t seem particularly earth-shattering to the forestry community. But they are influencing corporate decision makers and they raise the bar for decision-support tools. In years past working in the timberlands department might have been considered a backwater in the corporation; that is no longer true. Instead, budgets are increasingly scrutinized, and they are set more by potential payoff than historical precedent, and departments that can best justify projects through detailed analyses are winning out.
• #24: Modeling the entire enterprise as a monolithic model is impractical and therefore hierarchical planning with linked business models is the best alternative. Business trends and operational realities are demanding better decisions throughout the supply chain to improve the overall bottom line, and one way to deliver this is through analytics tools focused on increasingly more particular planning problems. This is the gist of operationalizing analytics. Going forward, the ongoing challenge for Remsoft will be how to maintain the open, flexible modeling environment that has been the basis for the company’s success to date, while providing intuitive working environments for non-modelers.