Inventory optimisation using AI made simple

Inventory optimisation is a classic use case for AI. Machine learning techniques can analyse your existing sales data to forecast demand on a SKU by SKU basis. This forecast information can then help you deploy inventory optimisation techniques to reduce stock outs whilst minimising working capital.

Step 1: Gather Your Data Start by consolidating your sales and inventory data. Most SMEs maintain this in spreadsheets or basic accounting software. Export this data into a CSV file. This file should ideally have columns like ‘Product Name’, ‘Date Sold’, ‘Quantity Sold’, and ‘Stock Remaining’.

Step 2: Choose an Easy-to-use AI Tool such as Google’s AutoML Tables – see below for instructions. It allows you to upload CSV files and automatically builds machine learning models for you.

Step 3: Upload Your CSV File Go to the chosen platform (e.g., AutoML Tables) and follow the prompts to upload your CSV file. Typically, this involves:

  • Creating a new project.
  • Clicking on an ‘Upload’ button.
  • Selecting your CSV file.

Step 4: Generate Insights Once uploaded, the platform will guide you through analyzing the data:

  • Forecasting: The tool can predict future sales based on past trends. For instance, if ‘Nature’s Delight’ tea sales spike every December, it will forecast a similar spike this coming December.
  • Anomaly Detection: It identifies unusual patterns. If there’s a sudden drop in sales of a particular product, it will highlight it, prompting you to investigate further.

Step 5: Interpret Results Most of these platforms provide easy-to-understand graphical results:

  • Charts & Graphs: Visual representations of sales trends, stock levels, and predictions.
  • Recommendations: Some tools also offer actionable recommendations. For instance, if a particular tea flavor’s stock is predicted to run out faster than usual, the tool might suggest an earlier reorder.

Step 6: Act on Insights Based on the insights:

  • Adjust your inventory orders.
  • Consider promotions or discounts on overstocked items.
  • Investigate any anomalies in sales patterns.

Step 7: Regular Updates To keep insights fresh and relevant:

  • Update your CSV file monthly or quarterly.
  • Re-upload to the AI tool for new predictions and recommendations.

Using AutoML Tables for inventory optimisation: A Step-by-Step Guide

1. Setup & Access:

  • Google Cloud Account: Before you can use AutoML Tables, you’ll need a Google Cloud account. If you don’t have one, you can sign up and often get free credits to explore their services.
  • Access AutoML Tables: Once logged in to the Google Cloud Console, navigate to the Navigation Menu (hamburger icon in the top-left corner) > Artificial Intelligence > Tables.

2. Create a New Dataset:

  • Click on the + NEW DATASET button.
  • Name your dataset, something relevant to your data, like “Tea Sales Forecast.”

3. Import Your Data:

  • You’ll be prompted to import data. Here, you can upload your CSV file. This is the file that has your sales and inventory data.
  • Once uploaded, AutoML Tables will automatically detect column types (numeric, categorical, date/time).

4. Identify the Target Column:

  • After importing, choose the column you want the model to predict. For inventory forecasting, this might be something like ‘Quantity Sold’ for the next month.

5. Train the Model:

  • Once the target is set, click on the TRAIN button. This will instruct AutoML Tables to start analyzing your data and build a model.
  • Training can take a few hours, depending on the size of your dataset.

6. Evaluate the Model:

  • After training is complete, you’ll be taken to the “Evaluate” tab.
  • Here, you’ll see various metrics that indicate how well the model performed. For non-tech users, focus on the “Mean Absolute Error” (lower is better). It gives an average of how much the model’s predictions deviate from actual values.

7. Deploy & Predict:

  • Once satisfied with the model’s performance, navigate to the “Deploy” tab and deploy the model.
  • After deploying, go to the “Predict” tab. Here, you can input new data (in the form of a CSV) to get future predictions.

8. Act on Insights:

  • Based on the predictions and insights from the model, make informed decisions regarding inventory ordering, promotions, etc.

Tips for Simplifying the Process:

  • Data Cleanliness: Ensure your CSV file is clean. Remove any blank rows, ensure consistent data formats, and avoid special characters in column names.
  • Regularly Re-train: As you gather more data over months, periodically re-train your model. This ensures it stays accurate and relevant.
  • Use Google’s Resources: Google provides tutorials, documentation, and support to help you navigate AutoML Tables.

While this is a simplified guide, remember that machine learning and prediction are complex fields. The initial setup and understanding might take some time, but once you get a hang of it, the process becomes more straightforward. With tools like AutoML Tables, the goal is to make advanced analytics more accessible to businesses of all sizes.

What is the half-life of your App?

An immediate result of the rapid rollout of LLM tools like ChatGPT, ChatGPT-4 and Bard is that application development has accelerated. Acceleration leads to faster execution and, just as significantly, it leads to lower-cost execution. Anyone with an idea for an app can generate an app. Anyone with an idea for an app but with no money can also generate an app. Furthermore, you don’t need to know how to code. There’s a marketing arms race for attention and we are already at DefCon No-Code. But this is a problem for app half lives. What is the half life of your app going to be?

Against this backdrop, app churn is going to go exponential. It’s too easy to create newer, better, faster, cheaper, shinier apps in an already crowded space and it’s getting harder to build moats: factors that protect you from competitors. The half-life of new app-based businesses is dropping, fast. ChatGPT and Bard have accelerated this trend.

If anyone can create an app then apps themselves, which are already fighting for attention-space in the app stores, are going to be impossible to defend from a marketing point of view unless they are supported by either extraordinary, existing brand power or by existing channels to market. Preferably both.

What this means for developers is that, even if you create an app that is outstanding in a given niche, it will be days or less before a competitor app does the same thing albeit with a little more magic dust sprinkled on it in terms of features. And there’s an app to spot new apps.

Two unstoppable forces are at play. The cost of app dev has plummeted. It’s effectively zero now and as apps are used to develop apps and apps develop apps that are themselves used to develop apps etc. there are parallels with the AI singularity. As a result, the number of functional apps vying for consumer attention is rocketing. But there is still a relatively fixed number of human attention minutes (HAMs) available. Consequence: an almost impossible to win, Pareto distribution, free-for-all fight. The cost of consumer attention, defined as the number of apps needed to get a single HAM is subject to almost Zimbabwean inflation.

Apps will need sophisticated, clever, fast-learning and intelligent marketing algorithms to get any significant attention at all. But there will be an app for that too, available to other apps. So developers are already in an attention-seeking arms race.

The second force is that app stores, already impossible to navigate easily, are crowded spaces and most apps, even excellent ones, are being crushed by sheer weight of competitor numbers.

With the cost of app development plummeting, app stuffing will be a growing problem. App stuffing, like keyword stuffing, is a brute-force method of gaining HAMs by developing large numbers of apps that operate in the same functional area. A consumer searching for a dieting app is already overwhelmed with options – I counted well over 100 diet apps before I got bored of counting – and if your 50 apps appear alongside 20 others from different developers, you stand the greatest chance of a download. App store curation does help of course but algorithms are busily analysing these too in order to game them.

A measure of how this is going is the half-life of an app as defined by the average number of days that pass before usage drops by 50%. There are, of course, many apps that never get used at all whilst others seem to dominate the market although they too are subject to decay.

So how to deal with this as an app marketer? You need to establish and build a brand. Successful companies know this and invest $millions in brand awareness etc. For individual developers, one method is to build a personal brand on Twitter.

If you are an individual and need to know how to do this, just ask ChatGPT. It’s a crowded space already.

Key concepts: App stuffing, App half life.