forecasting demand and planning

This post reflects on the valuable working time I had at ADVN in the demand planning role, in which I forecasted demand and delivered demand-supply plans for retail information solutions.

Avery Dennison Corporation is a multinational materials science company that specializes in the design, manufacture and distribution of a wide variety of labeling and functional materials. Avery Dennison has 2 business groups: Materials Group (LGM - Label and Graphic Materials and IHM - Industrial and Healthcare Materials) and Solutions Group (RBIS - Retails Branding and Information Solution). From February 2024 to July 2024, I was a Data Analyst for ADVN - Avery Dennison RBIS Vietnam. My work focused on forecasting and demand planning, and here are a few things I’d learnt from this extremely cool job.

Gathering forecast inputs

ADVN’s business includes providing tags, labels, embellishments, and RFID solutions to serve sportswear markets. Because of its B2B business model, the company’s demand is strongly driven by retail brands, and understanding demands of retail brands becomes key to good sales forecasts. This requires knowledge about regular products in the market, unique products of each brand, seasonal and cyclical order trends of products, and new product programs, etc. However, it is a challenge to obtain these information in advance since brands are not willing to share their plans early, plus customer buy plans and market trends are rapidly changing due to the competitiveness of the industry.

Generating good sales forecast requires engagement from global and local commercial teams, as they are involved in frontline customer service. The commercial team understands cutomer businesses and their behaviours like no other team. They have visibility about market leading accounts, new win/loss, and account nomination status. Forecast inputs from commercial teams are collected and linked to the finance AOP forecast. The AOP forecast is a forecast of $ value growth generated by the local finance team. It is aligned on quantity level growth rate and sign-off for capacity planning. Besides, combining statistical forecasts can enrich forecast inputs with past sales and time series patterns. Statistical analyses are particularly useful in cases of stable demand and available history.

Demand planning process

In the demand planning process, gathering forecast inputs is the first step to achieve the consensus forecast. When forecast inputs are already available, demand planners need to convert all these inputs that are in different dimensions and units of measure into one universal measure. At this stage, forecast models get messy easily. To ensure efficient planning, it is vital to have a well-developed logic and process for demand conversion. Conversions between finance forecasts (in value) and production forecasts (in volume) and its breakdown per site, product line, and brand level often need manual allocations and adjustments to specific programs and technologies.

After demand conversion, the next step is to hold a stakeholder meeting for cross-functional discussion and alignment. Meeting attendees include PICs from commercial, finance and operations teams. General managers should also join to align on strategic plans for the business. Then, the final sales forecast per FG per location will be used for supply planning. In supply planning, supply planners often perform some analyses on top of the sales forecast they receive to ensure the most cost-effective purchase plan for each material required. These analyses take into account multiple factors: SOH, MOQ, scrap rate, availability of materials, order costs, and lead time for delivery, etc.

In a superb fast-paced industry like sportswear, planners need to find a flexible and adaptable way for managing ad-hoc changes. Demand planners should possess an analytical mind and sound business acumen to leverage market insights (e.g., pulled-forward effects, time-phasing delays, transitions in demand) to adjust sales forecasts timely. In addition, they should own strong collaborative skills to empower supply and operations teams to comfortably react to sudden spikes or drops in sales. In summary, the last step is to regularly review and update forecasts and ensure ad-hoc changes are effectively shared for visibility and follow-up actions.

Forecast performance measurement

An essential and indispensable task for demand planning is measuring forecast performance. Depending on natures of industries and internal businesses, different organizations use different terminologies and metrics to measure forecast performance at different levels. But overall, hit rate and forecast accuracy are probably the most commonly used measurements. They are very easy to compute and comprehend.

Hit rate

The first indicator, denoted as H, basically tells the achivement rate of the actual sales versus forecast. Its formula is given in equation (1). If H <100%, the order intake volume is smaller than the forecast volume. This low order trend results in excess stock and high inventory costs. If H >100%, the order intake volume is larger than the forecast volume. This strong order trend (also-called over-forecast issue) leads to material shortage and service impacts. Ideally, H is within [80%,120%].

\[H = \frac{Actual}{Forecast} * 100\% \tag{1}\]

From a high-level perspective, a hit rate provides a quick view on the total sales achievement. However, it is not able to reflect the variance between the actual sales and forecast at lower levels.

Forecast accuracy

The second indicator, denoted FCA, shows the overall accuracy of the forecast data. It is generated based on the absolute error between the actual sales and forecast by equation (2). If there is no actual sales while the forecast is positive, then the forecast accuracy is zero. Otherwise, the higher the FCA, the better the forecast data. So the goal is to bring the FCA number up as high as possible.

\[FCA = max[1-\frac{|Actual - Forecast|}{Actual}, 0] \tag{2}\]

The forecast accuracy let us know how successful a forecasting task is and help us determine areas for forecast improvement. Its limitation is not being able to tell in which direction the intake quantity deviates from the forecast.

Demand planning challenges

First, a distinguishing feature as well as difficulty of the fashion industry is extremely large SKUs. Products vary by size, color, pattern, category, and material. One category of product can have a combination of thousands of SKUs, adding to complexities in demand, supply, production and inventory planning and management. The fashion industry is also characterized by change, which obliges planners to keep up with continuous development, obsolescence, and transition of products. Common problems faced are having limited or no data for forecasting sales of new products, over-forecasting of one-off or obsolete products, and missing sales patterns of transitioned products. For example, last season Adidas ran a large sales promotion of long socks. This season, they decided to promote short socks. Even though these are complete different products, they target the same customer segment, season and sales value. Understanding there is a large demand of new products coming for the season for material and capacity preparation while not overstocking outdated products is to plan right things and plan things rights.

In the fashion industry, the commercial and production team often use different “languages” to determine and distinguish between FG items. The commercial team tells items apart by their visual design, program, category or customer group. But the production team prefers to tell them aparts based on their specification and production method. The language of the commercial team is practical and friendly, whereas the language of the production team is coded and specialised. To put it simply, one language can be refered to as customer item, the other one can be refered to as item code. A demand planner is a key contact person between the commercial and production team, so a demand planner has to be comfortable speaking both languages. This means mesmerizing a lot of unique names for newcomers. Can we combine customer item and item code to create a single name for cross-functional use? Again, SKUs are intensively complex. It is possible but actually hard to systematize them in one way that serve all teams and purposes.

A great challenge I’ve got to handle in the demand planning position is low engagement and inefficient meetings to drive the results. As explained earlier, this is a B2B manufacturer with a MTO manufacturing strategy. To make good predictions of sales, commercial teams must have a deep engagement with customers to understand their demand. And to ensure smooth supply and operations, they must proactively share a follow-up on customer order plans with operations teams. Or else, poor commitment to customer service and cross-functional communication will lead to missing predictive insights and inconsistent data. Then, a cycle of inefficient meetings is repeated to drive some alignments but ironically they turn out to be wasteful and demotivating for everyone. Some major causes of inefficient meetings are unclear objectives, unstructured agenda, lack of PICs, unprepared participants, etc. But actually, running such stakeholder and strategic alignment meetings are difficult, even for experienced supervisors and managers. It takes time to take control of these meetings and drive super-productive outcomes.

Limited access to planning systems and resources is another uneasy constraint in demand planning. Being the starting component of the entire supply chain and business planning process, it must aggregate different types of information in different dimensions as inputs. On one hand, it requires an open friendly DBMS for many sales PICs to view and adjust forecast inputs at their conveninence. On another hand, data analysts need a much more advanced platform to organize, retrieve, and explore data. Lack of access to effective tools and softwares is a great difficulty for planners. Consequently, they have to spend a lot of time manually consolidating, cleaning and mapping data. But it is also important that planners are well trained for using these systems. For new businesses and team, the lack of standard structures is also a problem and planners hardly have time for projects due to repetitive daily tasks. This not only causes loss of productiveness but also impacts the planners themselves emotionally negatively. What’s important, they should be empowered to focus on areas of work that can be productively streamlined and automized.

Final thoughts

During my time at ADVN, I was able to learn about the company’s business, sharpen my data skills, and learn to tackle problems that are not on paper. It was an enlightening journey, technically and emotionally. I like that I become a little better now at collaborating with others, coping with challenges, and managing stress thanks to that journey. Grateful and looking forward to what’s next.