As a rough guideline, we can process up to 15,000 data points daily. We can double this within days, if necessary, and have done so in the past.
Integration is actually very simple. What usually takes the most time is correctly defining the input (images, fields, etc) and output (schema) as well as writing detailed instructions (see our documentation how to write instructions). Once this is done, implementation is quick: connect to our API or download processed data in batches from the dashboard.
Most of our clients need their data processed within 24 hours. For some clients, we can return data within 2 hours, and, once our AI is trained, it's instant. The exact SLA will depend on your desired levels of quality, cost, and speed.
Yes. We have had clients in similar situations (see our LogMeIn case study). Get in touch with us so we can get to work!
Integration is actually very simple. What usually takes the most time is correctly defining the input (images, fields, etc) and output (schema) as well as writing detailed instructions (see our documentation on how to write instructions). Once this is done, implementation is quick: connect to our API or download processed data in batches from the dashboard. If you need to work with your internal IT team and request resources, our guidance is that they should allocate 3-20h in total, depending on the complexity of your own system.
Typically we charge between $0.15 and $1, depending on use case, volume, and performance metrics. The more data you submit, the lower the costs. As we process data, we train our AI and decrease the dependency on humans. We pass these savings on to you. Get in touch with us for a quote that is specific to your use case. We optimize this for you, according to your needs for quality, cost and speed. Need super high accuracy? We will make sure you get it, but it may cost more than lower accuracy and take longer. Need results instantly? We can train our AI, but at the start we won't be able to provide you with 99% accuracy, although we expect quality to go up as the AI improves. Additionally, as our ML models get trained and more and more tasks are handled by the AI, we can pass these cost savings on to you.
On average, we have save achieved cost savings of about 62% for our customers. Results vary per use case. To understand what we can do for your business, think about any tedious, repetitive work you have your employees perform either on a permanent or on a temporary basis. We can probably automate 70% in the near term and up to 99% in the long term
As a default, you can cancel anytime.
As our ML models get trained and more and more tasks are handled by the AI, we can pass these cost savings on to you.
Google Cloud AI and MTurk are like their VM/servers: extremely powerful and require a lot of set up and configuration with little support. Using our solution, you get a guarantee of quality (our labelers pass a qualifying exam and are monitored throughout the labeling process and disqualified if their performance drops), a customer success agent to help you write and iterate on your instructions and analyze your results, and built-in machine-learning-based automation of labels where applicable. It’s the difference between a fully managed assembly line that works immediately and an empty factory with a lot of potential. Of course, we build on top of those powerful building blocks. Set up a call to discuss more.
Get in touch with us to discuss your requirements. We will make this work for you. You need to have data or tasks you need automated. And you need to agree with our terms and privacy policy. We will also charge you a nominal fee.
Please drop us a quick note. We can send you case studies and one pagers, as well as ROI analysis for your needs.
We use a set of ground truth data to measure the quality of each method of labeling. We do this by applying 12 different kinds of checks that we apply against each source and each method of labelling, be it machine or humans. We are then calibrating our router according to the quality score of each source of labeling and will route your tasks to the best worker. This allows us to 'weed out' poor performance and optimize the process. You can set performance targets for the quality (as well as cost and latency) you need, and we will guarantee the performance accordingly. Make sure to set the performance target according to your business needs: better quality may lead to higher initial costs. To guarantee high performance from the start, we are also training our crowd analysts. For a group of tasks, e.g., OCR/image transcription, crowd analysts have to pass qualifiers/exams, and only qualified analysts are allowed to perform certain tasks. For larger projects, we work with our clients to write custom qualifiers before we even perform the first tasks.
It's a bit of our secret sauce, but we can hint at a few things: we use a set of ground truth data to measure the quality of each method of labeling. We do this by applying up to 12 different kinds of checks that we apply against each source and each method of labelling, be it machine or humans. The more ground truth data you supply, the better we can measure the quality. We are then calibrating our router according to the quality score of each source of labeling and will route your tasks to the best worker. This allows us to 'weed out' poor performance and optimize the process for you.
Yes, we measure each worker or AI against the performance metric you specified. We are then calibrating our router according to the quality score of each source of labeling and will route your tasks to the best worker. This allows us to 'weed out' poor performance and optimize the process for you
We will try and re-process your data, free of charge. We will work with you on improving the quality. In the very unlikely scenario that we are unable to fix it, we will issue a full refund.
Yes. You can monitor project performance in our dashboard.
We put A LOT of effort into our crowd and into the agreements with any of our suppliers. To guarantee high performance, we are continously training our crowd analysts. For a group of tasks, e.g. OCR/image transcription, crowd analysts have to pass qualifiers/exams, and only qualified analysts are allowed to perform certain tasks. For larger projects, we work with our clients to write custom qualifiers before we even perform the first tasks. External suppliers have to pass the same training. They will also be disqualified from working with super.AI should they not meet our quality standard. Occasionally, we need to employ specialists (e.g., doctors, non-standard languages etc). We also require these specialists to pass qualifiers before they can perform any work.
We cover all major Western languages. For some non-standard languages, we may need to "activate" some analysts that have not been needed. This may take 2–3 days and we advice you to speak to us beforehand.
When you upload your data and give us instuctions (your input), we may break it up into smaller tasks and route these small tasks to the best humans or machines to make predictions/label this data. We combine these predictions and assemble them into a single label. The labeled output is used to train the AI algorithm and rates the performance of each human or machine, so we better understand who is best qualified to make predictions on your data. A QA system measures label quality, relables data where needed, and updates the routing. During this process, each machine/human only sees a small, broken-down part of your data. Because we break up each task into smaller tasks, labellers (machine or human) cannot reverse engineer the overall intent of the labelling. We do not allow our crowd to copy any of the data and we require them to sign strong confidentiality agreements. As a default, we delete all data one month (30 days) after processing. We may retain some anonymous, aggregated statistics to improve our services.
As a default, we delete all data one month (30 days) after processing. We may retain some anonymous, aggregated statistics to improve our services.
All of our crowd sign very strict confidentiality agreements. Additionally, we make sure each analyst only sees a small, fragmented part of your data. Analysts are not allowed to copy any of your data. Digitally, we have comprehensive data security protocols in place.
If we send your data to humans, those analysts will see a small, fragmented part of your data when they make the prediction/labeling that is required. They will also see the instructions you have written. Our crowd is based all over the world. If we send data to machines, some fragmented parts of your data may be processed by third parties such as Google or Amazon and are governed by their respective terms. super.AI may also process your data. Our servers are in the U.S. and Europe. Additionally, super.AI staff has restricted access to your data, which is strictly on a need-to-know basis, and reserved for customer service and bug fixing. We are headquartered in Berlin, Germany. We also have a presence in Jakarta, Indonesia.
When humans perform the tasks, each machine/human only sees a small, broken-down part of your data. Because we break up each task into smaller tasks, labellers (machine or human) cannot reverse engineer the overall intent of the labelling. We do not allow our crowd to copy any of the data and we require them to sign up to strong confidentiality agreements. When we send your data to machines, the same process applies. We don't send the entire dataset to a single provider. Equally, we vet 3rd parties to make sure their confidentiality agreements meet our standards.
Data sharing with 3rd parties is restricted to processing your data according to your instructions. We don't and will never sell your data to 3rd parties. Any 3rd party that may receive your data for processing is bound by confidentiality and data privacy requirements that meet our standards. These 3rd parties may provide AutoML solutions such as Google or Amazon. They may also be suppliers to super.AI.
While we carefully review terms and conditions of each API / 3rd party we work with, we are unable to monitor potential changes in the API terms of all our suppliers. However, suppliers that manage a human crowd are bound by our terms and conditions, which correspond to the terms of service and privacy policy.
Of course! Simply choose a data program, write instructions, and upload your data. Alternatively, get in touch for a more comprehensive trial. We can help you with instructions, give you a quote for your use case, and guide you through the process.
.svg)
