In 2014, MobiDev became involved in a small modernization project of a Point of Sale system for our US-based client, SmartTab. What started as a simple transfer of code base to a newer platform, grew into a leading POS system with top-notch functionality, ML-based demand forecasting and a recommendation engine. The result of a more than 10-year partnership with a bunch of other projects gave our team a deep understanding of market dynamics and all-around expertise in POS software.
A cornerstone solution we came up with, is the importance of innovations as a means for our clients to maintain their competitive advantage. This entails the use of AI and ML to modernize well-established features by new approaches. So here, we’ve put together five of the most effective and realistic examples of how you can upgrade your POS software to adapt to changing market conditions and customer expectations.
1. AI-Driven Demand Forecasting
Demand forecasting gets as complicated as the number of products and venues you want to cover with inventory planning increases. Applying machine learning to perform demand forecasting on a regular basis resolves a lot of challenges related to collecting, analyzing, and making use of the available data in your organization. Besides that, you get a competitive advantage to oversee demand fluctuations and respond to market needs in a timely manner.
Sales data, in this case, is your most available asset that naturally comes through the point of sale system. Although, to analyze market conditions properly, we would also need to source data related to geography, global economics, marketing activities, seasonality changes, etc. Machine learning modules integrated with your POS and data storages can aggregate all the available data and provide you with precise reporting automatically.
Value for business summarized:
- Support in strategic pricing planning
- Improved demand prediction for inventory planning
- Efficient resource allocation and financial planning
- Single source of truth for distributed management teams
How to implement demand forecasting
Demand forecasting relies on machine learning forecasting models that aggregate historical data to find patterns and predict future demand. This is done using sales and a bunch of other data from the past, and could be translated into an automated pipeline that reports demand forecasts on a weekly, monthly, or quarterly basis.
While it’s a common scenario for businesses to have tons of sales data stored for years, a 3-month period can be enough to start with forecasting for a new venue or product type. However, the minimum time required to start developing a model for your specific business is set by forecasting accuracy standards and specific business goals.
Process-wise, the work starts with data collection and analysis, followed by a data understanding stage. During this phase, we provide our clients with preliminary estimates and project scale, because AI engineers can understand the specifics of available data and what it takes to build a functioning demand forecasting system. After this, the model training starts, which further proceeds to developing customer-facing logic.
The rawest form of presenting forecasting results are spreadsheets, which allow the users to quickly hop on a new system and dig into valuable data. However, we commonly develop an explicit user-interface that consists of customizable dashboards, data visualization tools, different query mechanisms and user access policies. We’ve described the process of developing demand forecasting in detail in a separate article, so check it if you need a more in-depth explanation.
What you need to get started:
The amount of data you have is usually a determining factor with regard to whether we can use this data for the model training. However, the lifespan of a product is a determining factor for what period we need to use to collect data to have a starting minimum. Usually, a three-month period of historical data for a single item is enough to begin with. But if your task is to predict the weekly demand for a product like coffee, a three period of data will do.
Each product has different seasonality cycles, which will also impact the period of data collection. For example, if certain products fluctuate significantly in demand during the year, we need a sample of yearly data to train the model that produces sufficient forecasts.
Success Story: SmartTab
SmartTab is a US-based premium POS hardware and software provider for high-volume bars and restaurants. The MobiDev team has covered a full-cycle of SmartTab ecosystem development, including POS, customer application and AI analytics with a demand forecasting module, since 2014. Now, SmartTab offers its customers a customizable analytical dashboard that presents business insights for each venue, as well as total chain statistics.
The data is constantly updated through an automated machine learning pipeline that connects numerous sources of data and refreshes once new information is available. This provides users with 365 days of prediction periods, but the time period of forecasting can be set for a custom period or chosen from the available options. Data-wise, SmartTab processes data incoming information in real-time, which is another useful feature for operating high-volume venues and quick monitoring of venue performance.
2. AI-Driven Sales Forecasting and Advanced Sales Analytics
In recent years, artificial intelligence data analytics has become one of the most popular POS technology trends. Similar to demand forecasting, sales data can be used to provide sales forecasts based on historical numbers generated through actual transactions. While both demand and sales forecasting often come together, other smart algorithms can be used to implement advanced analytics. For example, we single out the most and least selling products, the amount of earnings for each product, or a category of products. You can also opt for the tracking of selling trends by time periods, calculating optimal margin or comparing forecasts to actual sales.
Value for business summarized:
- up-to-date insights thanks to operational data analysis (this can be done both in real time or as fixed periods)
- minimize stockouts, and improve overall inventory efficiency
- optimize pricing strategies at the point of sale
- effective tracking and analysis of sales performance metrics
- optimize sales strategies
How to Implement Sales Forecasting
The process of gathering and analyzing data for sales forecasting as a standalone module has a lot in common with demand forecasting projects. The difference will appear depending on what type of metrics you want to see in the final system. These peculiarities become clear during the business analysis stage and data exploration procedure.
Since sales data comes at you rapidly, a machine learning model needs to get constant updates to provide up-to-date analytics. Our common approach is to set up an automated pipeline to source new data and launch the retraining procedure. Generally, this process is done by running a script once a day in the quiet hours as a scheduled job. However, time periodization of the model updates depends on the reporting frequency. So time granularity can range from days, to months and quarters, or any other custom option suitable to your needs.
A separate point of discussion is implementing a clear user interface. This is something that our team has worked through thoroughly with the UX and AI departments. This is because analytical dashboards show all the data, and you need to understand which data a user might want to see in the first place. For example, a user interface has to show the correlation between the forecast and the actual sales numbers, as well as automatically highlighting anomalies and deviations in the prediction ranges. The deviation can be caused by marketing campaigns or economic fluctuations on the market, so the interface has to navigate the user to analyze such points.
What you need to get started:
As we already mentioned, sales forecasting is similar to implementing demand forecasting except for the fact that you only need sales data. The more data you have, the better the future model will be able at creating forecasts. The only thing worth noting is that there is always a list of factors that affect sales, so these indicators should also be taken into account during the development process.
3. Intelligent Inventory Management
Effect inventory management is all about keeping the balance between overstocking and understocking your products to meet your customers’ demand without losing money. AI can act as a great tool to simplify your day-to-day inventory planning operations by automating manual calculations and refilling your stocks based on the data you have. Combined with hardware to keep track of your physical storages, inventory management can save you the cost of hiring on personnel while helping to keep everything organized.
Value for business summarized:
- reducing the need for manual intervention and repetitive tasks
- improved inventory tracking with minimized stockouts and overstocking
- better decision making through valuable insights into inventory performance, sales trends, and key metrics
- enhanced visibility of inventory across different venues/locations
How to Implement Inventory Management With AI
Demand forecasting systems can be naturally extended to hop over your inventory management, because it provides the information based on which you can regulate your stock levels more flexibly. For this purpose, we need to set up an inventory tracking system that will digitize the numbers of stock levels, so we can use it with AI algorithms. This may include optical sensors, as well as thermometers and weight measuring devices that gather data about the remaining stocks and pass further.
What you need to get started:
The preparation for such a project will involve AI engineers analyzing the available historical sales data, as well as your inventory data, to develop a technical vision. Depending on your business goals, the final product may vary in complexity. So here we can look at these concepts applied in practice on the example of our project BarTrack.
Success Story: BarTrack
The founders of BarTrack developed an innovative IoT solution designed to automate inventory management for serving bars, breweries, and restaurants. Our team was in charge of developing integrations with several of the most common POS systems, as well as devising a complex ETL process to supply BarTrack with data for downstream tasks and analysis.
The resulting platform combined a proprietary hardware beverage sensor that measures product volume and various system conditions, such as temperature and pressure. All of this data is used by software components to make data-driven decisions on the beverage stock to automate inventory management, or simply track the bar capacity.
Through seamless integration with the customer’s point of sale system, detailed analytics come to the forefront, providing an hourly breakdown of losses — pinpointing where, when, and under what conditions they occurred. The result is an invaluable tool for businesses aiming to optimize their operations and enhance overall efficiency.
4. Automated Identity and Age Verification
Automated ID verification can both enhance customer experience and help your business comply with regulations. For example, for businesses that require age verification, such as liquor stores or establishments selling age-restricted products, automated ID card recognition allows users to quickly verify the customer’s age by scanning their ID card. It can also streamline the checkout process, integrate with loyalty programs, prevent identity fraud and unauthorized transactions.
Value for the business summarized:
- reduce a risk of legal penalties or license revocation due to unauthorized sales to underage individuals
- prevent potential financial losses due to identity fraud and the use of fake or altered IDs
- serve as dispute evidence that verifies the identity of the customer in case of a disagreement over payment
How to Implement Identity and Age Verification
Optical Character Recognition (OCR)-based solutions are commonly used to scan and extract data from IDs. The OCR module captures the identity document and extracts relevant information from the document, such as name, date of birth, or ID number. After that, AI algorithms can check the document for certain conditions, such as identity document format, authenticity of MRZ, expiration date of ID documents, verifying hologram/rainbow prints, etc.
The developed AI algorithm can also analyze and match identity information, such as ID cards, passports, driver’s licenses, or facial features, against existing databases or reference data if the data is stored. In this way, you can form full-fledged customer profiles based on ID for more efficient service for each customer.
In most cases, to achieve these goals, you can rely on ready-made solutions and adapt and integrate them into your business. It is easier and faster than developing such a system from scratch. But then again, you might want to get more advanced functionality based on what you already have. For example, the case described above still does not solve the risk that the client will provide you with someone else’s documents. To minimize this risk, you can add a face match to compare a person’s face and a scanned photo.
We did a similar internal project to verify employees at the entrance to the office. We used a computer vision model to identify a person’s face, compare it with the previously obtained photo and unlock the door.
What you need to get started:
The scope of such projects depends on exactly what problems you want to solve and what features to implement. For example, a solution for age verification does not require any special input from you, since the main task of engineers will be to choose an effective third-party product and competently integrate it into your system.
5. Smart Personalized Recommendations
Recommendation modules in POS systems can help increase sales because they allow employees or customers at the point of sale kiosk to receive personalized product suggestions. By consistently providing tailored recommendations that align with customers’ preferences, businesses can strengthen customer relationships and encourage repeat purchases. Smart recommendations can also apply to pricing and discounts based on demand and sales forecasting.
Value for the business summarized:
- Enhance customer experience
- Increase sales and revenue thanks to upselling and cross-selling opportunities
- Optimize financial planning
How to Implement a Recommendation System
Recommendations are based on the client’s previous data, or sourced via a database of products considering the most popular choices paired with other products. AI engineers build a model that solves a specific problem, such as understanding the patterns of purchases from your menu, so that the waiter can offer suitable dishes or beverages to customers.
Recommendations can be built according to various logic patterns and based not only on the internal data of the organization. For example, for a restaurant chain, the system can gather ratings, reviews, and other relevant attributes from third-party resources to improve recommendations. The system may also consider contextual information, such as the time of day, occasion, or current trends to provide more tailored suggestions. The recommendations can also be personalized to the user’s preferences, dietary restrictions, location, and other relevant factors if you have user profiles and can analyze the data of their previous interactions with your business.
If we are talking about pricing recommendations for managers, price predictions based on various factors affecting price growth, such as cost, changes in regulations, seasonality, etc., come into play.
What you need to get started:
A dataset with your products. Available information about the pricing features of your business if we’re talking about price recommendations.
Success Story: Comcash
ComCash is a US-based ERP software company we’ve been working with on integrating AI modules for demand forecasting, providing statistical reports and recommending products to the user. Our long-term partnership with ComCash began in 2013, and since then we’ve built a robust ecosystem of tools that led to the acquisition of ComCash by POS Nation in 2022.
The recommendation system in ComCash uses associative rules, and an Apriori algorithm to recommend goods that customers buy together. This model was a part of bigger data exploration we did on this project, so that the whole AI backend is bound in a single system that communicates available data from the ERP system and helps the user to make more informed decisions.
Innovate Your POS Software with MobiDev
All of the above are just a few options for how we can help you innovate with your POS software with the help of artificial intelligence. As part of our AI consulting services, we will be able to study the specific requirements of your business, sync them with the needs of the market, and offer a roadmap for the technical implementation of the most effective solution.
To get started, all we need is an understanding of exactly what data your business collects and what tasks you see in front of you. You can provide a small snapshot of data that has undergone sanitization, i.e. which involves removing or modifying sensitive or personal information you don’t want to share. From this point, our AI engineers can brainstorm with you to help you get real value out of your data.
Book a call with our experts or contact us through the form below to start a conversation.
