Machine learning will leave no stone unturned in its relentless progression of every industry.
Below you'll find an expose of some of the great work delivered by Machine Commons' Suppliers.
All of these projects can be replicated, so consider them thought starters for your own business!
With access to all sales and marketing data, we were able to deliver intelligent machine learning solutions to improve all key performance metrics. It begins with understanding customers from input signals, profiling and segmenting consumers at a more granular level than would be possible with traditional statistical or human teams. The result is the presentation of far more accurate product recommendations, also delivered at an incredibly granular (essentially 1 to 1) level, thereby improving customer retention and satisfaction scores, further resulting in increased uptake of both special offers and the loyalty program. All results are A|B tested against traditonal methods and performance regularly improves revenue by 10% across all digital sales channels.
ARA - automated review assitant
For a large medical device regulatory certifcation company, we incorporated machine learning technology (Neural Networks, NLP and logistic regression) into their processes to optimize the time spent by domain experts. This increased their productivity and reducing the cycle time for the certification process.
Information extraction and RPA
Invoice management can be one of the most time consuming manual prcoesses that human accounting teams have to manage. We worked closely with the client teams to understand their processes, their systems and to understand the huge variety of nuances their invoice teams came across. The result was an information extraction system that isolated key information from invoices, filing and actioning administrative tasks accordingly.
Medical facility management
This governmental medical facility struggled with resource like any other. An accurate prediction of staffing requirements directly translates to better medical care. Predictions included occupancy prediction and no-show prediction so staff managers could use this foresight to better allocate human resource.
Automated speech recognition based solution
Covid has resulted in vast numbers of stay at home kids and the education sector has sorely lacked the infrastrucure for effective home learning. Language are especially tough as they require hands on learnin and in person communication. We developed a module for home assignements to faciliate the use of speech in language learning, powered by a speech recognition tool built using Machine learning.
There's a huge amount that goes into city managment and traffic management is no exception - it arguably influences the greatest number of people on any given day and the increased efficiency of traffic management results in a direct impact on economic efficiency. We were able to deliver an visual analysis system that worked down to the granularity of a specific car detection (i.e. the make and model classification).
Automate client agreement procedures
Legal time is expensive. We created a cloud based document agreement tool with electronic signature functionality to reduce the number of administrative hours.
We built a computer vision based system for SKU (stock keeping units) classification. The result is a camera that feeds into the data management system, which can identify and count stocked items. This allows for a fully automated SKU register to reduce unnecessary inventory and to place orders at the right time to ensure there is always just enough stock ready to go out on the shelves. The number of manual stock checking hours saved rendered this a highly effective cost cutting exercise.
We created a bespoke content recommendation tool (across multiple types of media), based on consumption pattern and content similarity, in order to provide viewers better content recommendations and as a result increase engagement time.
CRM cleaning, normalization and enrichment
SIMON holding (Switches, light, sockets)
CRM cleaning, validation and geolocation of all data (+100k entries) to optimize postal marketing (15k detected duplicates) and improve the efficiency of workforce in-place visits. Our model analysed errors for patterns and was able to suggest other line items as potential errors with a high degree of accuracy, also making error correction suggestions to save time (in many high-confidence cases, without need for human oversight).
Analytics engine (IoT Analytics)
Designed and developed an analytics engine for this smart home automation company.
Clinical trial optimisation
Develop a technical infrastructure and AI to predict human behaviour for the purpose of clinical trial optimisation. Sometimes trials can fail simply because the trial participants were not optimally chosen; selecting people via machine learning reduces this risk.
Prostate cancer diagnostics
Broader statistics show 1 in 5 scans reviewed by a human have some kind of condition missed. We developed an AI computer vision solution to automatically read MRI scans and determine presence of cancer. In conjuction with human oversight, huge numbers of hours are saved and medical treatment is hastened and improved.
Health and fitness
Weight loss app
Customer Success Consulting LLC, US
We developed an app for monitoring the exercise habits of consumers and thereby help their progress towards achieving health and fitness goals. The functionality of the app also allowed their progress to be shared via social media. For the client, this translated to increased consumer engagement and improved top of mind brand awareness - so patients knew where to go when they needed professional guidance.
Identify class of reflective sheets (IoT Edge Analytics)
Developed a mobile application so workers could use their mobile device to predict the class of relective sheets manufactured by the company. The ML model is designed to run on mobile phones and turned every such device into a point of quallity assurance.
Using sensor data for status and maintenance prediction.
Distributed acoustic sensing for real-time observation and prediction of the state of atrack, the location of rolling stock and other assets, and general event observation. The interpretation of these massive and incredible noisy datasets presents a major challenge, but using convolutional neural networks we were able to cut through the noise and provice meaningful information for decision makers. The client was able to use this information to understand both the status of the track and goods, as well as make predictive assessments about future maintenance requirements.
Evolutionary algorithms to power a dynamic web enviornment to improve conversion rates.
Econometric market mix and propensity modelling with impressive, award winning results in UK and Germany. When a customer walks into a typical bricks and mortar store there are salespeople to assist towards making a purchase; the same is clearly not possible in an online setting, which the pandemic has made the most important focus and competitive environment for retailers. We were tasked to create a machine learning solution that could identify types of customers to the website in real time based on their predicted purchase behaviour. We used data signals from web and sales and explored trillions of possible model and data variants - using "genetic algorithms" - to find the best models. The result was an online web environment that dynamically presented consumers information that was most likely to lead to a sale - much like a real salesperson would provide.
Extracting skills from the text, we created a skills multi dimensional skill matrix and could therefore assign both role requirements and human resource a profile against this matrix. Using this information, we were able to identify both present day and future skill gaps so hiring and training efforts could be adjusted accordingly.
Player segmentation to predict revenues and churn rates, based on captured and lookalike behaviour.
Smart reconcilliation tool
Qument SaaS product for delivery services
A SaaS based solution which allows customers (restaurants, cloud kitchens) to do smart matching of payment against orders. Helps the customers in reducing manual efforts thereby reducing lost revenue.
Employee messaging platform
A secure employee instant messaging platform with interactive dashboard
Patient waiting list automation
Automate the clean up of patient waiting lists to save hospitals 1000s of man hours of administrative work.
Reliable training data
Machine learning models are limited by the quality of their data. Image annotations services are an intensely manual yet necessary component in creating reliable training data. This is true of essentially every industry application and this service would be used in conjuction with another data engineering consultancy.
Recommend the right insurance policies to customers.
A recommendation engine specific to the client's consumer data was created to recommend the insurance policy that was most relevant to a given customer (thus increasing conversion rates and satisfaction). A market basket analysis technique uncovers meaninful correlations between product and consumer segments, used to match insurance policies to customers profile information.
Demand forecasting engine for fertiliser
A statistical engine to forecast demand for various fetilisers and pesticide products sold in France. There are several environnmental and countless seasonal, cyclical and business factors that influence demand. Our forecasting system cut through the noise to provide this organisation the ability to manufacture quantities more accurately.
Wheat crop yield prediction
Government of India, Agriculture Agency
A deep learning powered model to predict wheat crop yield for a particular district in India. We used a combination of remote sensing data and crop cutting experiment data to provide the government a reliable means of yield prediction and therefore allocated supporting resources optimally.
Now especially, candidate applications are through the roof. Companies are overwhelmed with the number of applications and are therefore unable to screen all CVs manually. Training advanced machine learning - convolutional neural networks, we scored future candidates based on captured (/past) lookalike profiles. We used a number of informational sources (such as online profiles and written applications) to analyse the appropriateness of a given applicant for a specific role. This candidate-to-company matching provided the online recruitment agency a degree of automation so their staff were more efficient in delivering their clients effective human resource.