Why data strategy needs to include operations
Your operations need to be proactive, not reactive. Data is the key to unlocking this opportunity.
In the spirit of Marketing and Analytics collaborating. Clint Dunn, my former Data Lead at Hairstory and I co-wrote this piece about how data can be a more integrated part of any marketing strategy.
Let’s start with some good news. E-commerce sales were up over 40% this year, which can be heavily attributed to stores closing and reduced in-store capacities.
However, this isn’t great news for logistics operators, who have borne the burden of sky-rocketing orders. Particularly for last mile delivery, an elusive yet crucial step in delivery from distribution centers to end users.
The tandem rise of e-commerce and digital marketplaces, broadly coined as “The Amazon Effect”, “have forever changed the composition of consumer buying behavior and expectations. We now all expect fast, free shipping, or highly competitive pricing. The Cambrian-like explosion of e-commerce demand has significantly outpaced that of evolving supply chain and logistic models, and operationalists have been forced to bend their strategies to the low-cost and on-demand delivery models that consumers relentlessly demand.
This shift has left operations and logistic managers engorged in data but starving for insights. It’s logical that data driven decisions are more likely to succeed, yet many teams are unable to transform raw data into actionable insights. However, it’s not by lack of trying. As of now, the talent and tools this would take are scarce because the data inputs are disparate and internal stakeholders are disjointed.
It’s daunting to imagine extracting actionable insights from large, complex data sets manually, and is a heavy lift for even the sharpest analytical mind. Consequently, business decisions are still heavily biased by human judgment, guesswork, and intuition.
The Data Delta
To improve upon this, data teams, adept at organizing, structuring, and interpreting data, should sit at the intersection of marketing and operations and assuage the friction of collaborations between teams.
Data teams already have experience speaking other teams’ language, making them perfectly placed to service each team’s needs; marketing is focused on campaign, engagement, and channel data, whereas operations might consider revenue, expenses, and inventory. The first step in achieving this is to sync demand forecasting and supply models together — there’s clearly no reason to spend marketing dollars on products that won’t be available for sale.
Many companies manage forecasting through Google Sheets or Microsoft Excel, which aren’t optimized to clean and extract data into a standard structure. In an ideal world, sales forecasts would auto-generate with little manpower; yet, for a variety of reasons, this isn’t realistic for most. Many lack enough accurate, or interpretable data. Advanced statistical forecasting models need years of data with predictable trends and growth rates. High-growth companies are often experimenting with new products and marketing that make long-term statistical modeling near-impossible. Furthermore, it’s difficult to proactively aggregate historical, contemporary, and forecastable data to create the full picture.
For mature organizations with steady growth, machine learning models can forecast sales, and even incorporate seasonality factors. Yet this is more challenging for newer, high-growth companies who don’t have years of data that they can use to create a holistic composition of trends and variables.
In defense of the models, they’re only as valuable as the quality and amount of inputted data, and there’s no statistical model that would have forecasted the impact of COVID. With a nuanced understanding of how volatile our economy and way of life is, it would be naive to say that consumer demand can be accurately predicted using historical data points alone.
Humans are often better at understanding the world around them, more so than a model, particularly at the earlier stages of a company’s growth. The challenge then is for data teams to enable marketers, operations, and finance experts to interact with models in a hands-on way to generate meaningful forecasts while simultaneously cutting out manual, error-prone workflows.
Enabling business users to independently manipulate data is not only essential for ongoing model adjustments, but it also invites diverse considerations for multiple scenarios. This flexibility requires forecasting models, and their underlying data structure, to be extremely pliable and resilient to changes by non-technical users.
For example, operations teams often face urgent predicaments with unclear outcomes. If a component of a product becomes unavailable and the company needs to scramble for suppliers, the forecasting model should be able to weigh how different suppliers will impact short and long-term growth. Simultaneously, it must dynamically inform the marketing team’s forecast so that they aren’t pushing unavailable products, which might cause brand damage in addition to supply chain chaos.
This calls for a new generation of tooling, which enables processes and workflows to be streamlined and automated. An emergence of these tools will likely include:
- CRM x PM: Currently, there isn’t a standard platform that supply chain operators use, making data abstraction and use difficult. Having a centralized tool where a team could manage existing vendor relationships, automate purchase orders, and visualize, track and share the status of the supply chain funnel would also make the data more seamless to use, particularly if the tool has native integrations with something like Segment. This would also lend to enhanced collaboration, with visibility into factors like late shipments and material availability dates.
- Alerting: Notifications in real time when stock levels are at risk, inventory is expiring, or other issues with funnel management. This can save lead time between production and consecutive delivery stages.
- Forecasting and Analysis: Tools that use AI to better understand the impact of demand spikes or different business scenarios, like seasonality spikes, slowdowns in customer acquisition, and shifts in customer behavior from one SKU to another.
Outside of modeling, the data team should help create environments where all stakeholders have the information necessary to make forecasting decisions and to catch changes to assumptions early. Whether through periodic reports, automated alerts, or BI dashboards, stakeholders should be able to easily access data that informs their forecasts, and reviews should be done regularly to ensure that historical forecasts are directionally correct in predicting the future. The challenge is that forecasts often encapsulate dozens of assumptions about how a company functions and customers interact with its products. If a data team can’t provide context on the entire ecosystem, there should be ruthless prioritization to report on the pillars of the business.
It will likely be a long time until a true panacea is created for flexible forecasting across multiple industries. Challenges that stand in the way might include:
- Varying forecast and analysis variables between industries, such as average order value, frequency, and seasonal fluctuations.
- Product retail value (i.e., average order value) as a variable, which is highly dependent upon industry. This will be a key factor for high-velocity businesses, like fast fashion, with large, trend-driven catalogues.
- Fragmented inventory and supply chain data. Often suppliers, wholesalers, manufacturers, distributors, etc. have their own inventory and tracking platforms, making it difficult to generate a single source of truth.
- Stakeholder management and collaboration. Gaining clarity on who drives demand forecasting, finance, marketing, etc.? Either way, finance needs to be involved at the bottom line and systems need to be transparent and collaborative.
There are a few companies, like Well Principled and Fountain 9, that are leading this charge and building software that automates traditional management consulting projects, like supply chain and forecasting, costing and resourcing, and marketing spend. These tools take large data sets, leverage artificial intelligence and machine learning to evaluate decision drivers and build complex decision-making models. Those following along should also keep an eye out for lightweight ERP tooling for production management, including workflows, collaborations, and manufacturing.
If you consider one of the fastest-growing e-commerce giants, Shopify, it’s exciting to imagine a universe where tooling like this can be easily rolled out via plugins into Shopify storefronts. Plus, with sales, marketing, and inventory data integrated, companies can unlock sophisticated ways to manage their cash and assets. This could include discovering optimal times for lending (inventory financing, for example), and negotiating smarter deals.
The Bottom Line
For revenue/SKU forecasting, both from bottom-up and top-down perspectives, a new perspective may merge that’s less finance-led, and more integration with inputs from data operations, supply chain, and marketing. Scenario planning for inventory would be more streamlined and inclusive of more model-shifting factors.
E-commerce products that are commoditized are more scalable and accessible for this change. Upon adoption, as new products are created that support the integration of finance, marketing, and operations data at scale, companies can continue to improve efficiencies as they mature. Alternatives need to be at least 10x better in terms of efficiency to drive adoption. Once this can be achieved, the start of a broader cultural shift to integrated workflows will be ushered in shortly after. Embedding data into these work streams makes the outcomes even better.
Or at least, that’s what my integrated data indicates to me.