So you’ve put your best foot forward to optimize your inventory, but it’s still anything but perfect? To ensure continued success and boost your bottom line, check out these five easy action points below.
1. Opt for the right approach.
When deciding on optimal inventory levels, companies want to steer clear of risky assumptions and avoid relying on “gut feelings” and personal experience. This is why best practices in inventory management tend to rely on data science. However, even if your optimization initiatives do involve data science, this doesn’t guarantee that you are using the most efficient, objective approach.
To illustrate this, let’s consider three different inventory optimization techniques:
- Approach #1 — You use advanced data science methods to calculate demand, but then you add an unoptimized hunch-based safety buffer to the predicted value to form inventory. This doesn’t sound like a scientific approach to inventory optimization, does it?
- Approach #2 — You calculate demand probability distribution and apply a formula that weighs holding costs against shortage costs. Although this is pure data science, it’s again based on expectations, as we don’t really know the nature of demand probability distribution and can only assume what it may be.
- Approach #3 — You use a deep neural network (DNN) that considers multiple demand-influencing factors. The network’s complex architecture and its “intelligence” are designed with your specific situation in mind. Also, the network is tailored to directly predict an optimal inventory level, without producing demand forecasts in between.
While each of these three approaches has something to do with data science, the differences between them are striking. The third approach is the only one that can guarantee a higher degree of objectivity.
2. Build trust in deep neural networks among your team.
It’s highly unlikely that your team will accept the shift to DNNs without any resentment. So far, they have exerted total control over their categories, and now some strange new tool is going to dictate how much to order. So, be prepared to address your team’s skepticism.
Deep neural networks are technically complex, which makes them difficult to understand for non-techies (and this is exactly who your category or account managers are). If they don’t understand these networks, they won’t trust them — especially when such an important issue as an optimal inventory level is at stake.
So, here are some ideas of what you can do to help build trust:
- Reassure your managers that they are still the ones in charge. They still control their categories, but now managers just have to switch from making assumptions to handling more meaningful tasks. DNNs are designed to support these managers, not replace them.
- At the very earliest, announce that your managers’ support is critical. Who else can share such invaluable expertise to increase the accuracy of DNN’s forecasting?
- Organize a meetup, and involve tech professionals to explain in detail how DNNs work and how they produce forecasts. Make sure that your team fully understands the concept.
3. Prepare the ground for predictions.
First, consider what factors to take into account. For example, would promotion influence and seasonality be enough for you, or would you additionally consider more sophisticated factors, such as weather conditions, supplier risks, and store locations? You should make this decision before designing DNN architecture, as DNNs are able to see only what you tell them to see. So, if you don’t include weather among the factors, don’t expect a DNN to be able to determine if the weather will have any influence on your stock level.
Also, you should distinguish between perishable and nonperishable products. These two large groups will definitely require different approaches — at least when it comes to holding costs vs. shortage costs balance.
4. Prepare your data.
- Check data availability. — To become super-intelligent and produce accurate predictions, DNNs need to learn from your most detailed data. So, before going for DNNs, check your data. Do you have a detailed split of sales figures per SKU, at least for the last two years? Does your data set reflect any product attributes that can influence the stock level? If not, then it makes sense to consult professional data scientists to check whether you have sufficient data available to make a DNN work.
- Ensure data quality. — Erroneous data records can ruin your predictions. Take a thorough look at your data to identify which of the sales records are misleading or straight-up incorrect, and clean your data from such noise before you feed it to a DNN.
- Think about external data sources. — You can also go beyond internal data. If you believe that predicting your inventory level requires some additional external data — gleaned from social media, weather forecasts, or official statistics detailing industry trends, for example — you can make good use of this as well.
5. Bring data scientists on board.
If you need assistance while taking any of the above actions — for example, training your team on DNNs, assessing and managing the quality of your data, or defining demand-influencing factors — then this action point should be #1 on your list. I made it #5 having just the technical aspects of the future solution in mind.
Your DNN project is doomed if you don’t have professionals to build its architecture, set the network’s hyperparameters, choose proper activation functions, and assess the network’s performance. Whether these professionals are internal team members or external consultants will depend on your company’s unique needs and capabilities.
Putting Inventory Optimization Into Action
When it comes to inventory optimization, data science will be your key to success; however, you must know what door to use it on. Of the three approaches described — adding safety buffers to forecasted demand, assuming demand probability distribution, or using deep neural networks — I recommend paying special attention to DNNs.
As for the rest of the action points on the list, each of them requires a certain level of expertise and focused efforts, such as:
- Mastering efficient communication as well as change management for Action #2.
- Managing data and its quality for Action #3.
- Making serious analytics efforts for Action #4.
- Hiring talent or choosing the best partner among consultancies for Action #5.
As you work to optimize your inventory, keep these five key action points in mind; they’ll serve as useful guides while navigating the more complex aspects involved, making it easier to get your whole team on board as you implement these important changes.