While building our own platform, we have been keeping a close eye on the no-code AI space. We realized how difficult it was for non-technical people to build custom AI solutions and AI-powered process automation. That is why we wanted to share this knowledge with you.
While the no-code market is maturing as a whole (Dreamweaver and MS Frontpage, the first WYSIWYG (what you see is what you get) solutions, both launched in 1997), certain sub-segments are just emerging, making this space more powerful. No-code AI is one of them. As we are constantly observing the field, we thought that sharing these insights would be useful for you as well.
We are mapping out the intersection of no-code, SaaS, and AI: AI tools that don't require any coding or infrastructure to be set up for building powerful applications that can make decisions that previously required human judgment.
No-code in a nutshell
For as long as there have been computers to program, there have been attempts to make programming easier, faster, less technical, and available to a much broader audience. Essentially, any end-user programming signals that even though most computer users lack coding skills, they would welcome the application potential of various tools – as long as the effort to obtain these skills is low.
No-code stands for a family of tools that allow people to build applications and systems without having to program them in a conventional way. Instead, the core functionality is accessible through visual interfaces and guided user actions, as well as pre-built integrations with other tools to exchange information as needed.
While these self-imposed restrictions can lead to issues for very large or complex applications, the whole family of no-code tools is handing a big chunk of power to their users. As Alex Nichols from Alphabet's growth fund CapitalG said:
"No code is empowering business users to take over functionality previously owned by technical users by abstracting complexity and centering around a visual workflow. This profound generational shift has the power to touch every software market and every user across the enterprise."
To give you a few examples, here are some common things that can be built entirely with said no-code tools (check out Nocodelist for more examples):
- Websites and landing pages with Webflow (ours is built with it!)
- Web or mobile applications with Bubble, Adalo, Mendix or Thunkable
- Chatbots or virtual assistants through Octane AI, Kore.ai, Landbot or mindsay
- Databases through Airtable
- Connecting your tool stack with Zapier, tray.io, Integromat, Parabola or Paragon
- E-commerce through Shopify or Weebly
- Manage memberships with Memberstack
It is fair to believe that the no-code space is here to stay. AI tools built on these principles are showing that the field not only grows in width but also depth when it comes to the job to be done and technology in place.
Before we move to no-code AI, we will quickly touch on one fundamental question first: When does it even make sense to use AI?
When to use artificial intelligence
Note that AI can be used for a variety of applications but we intentionally limit our discussion to business applications.
Broadly speaking, AI is particularly helpful when there is some sort of intelligent judgement to be made by humans and when there are many of these on an ongoing basis. We often use the phrase "AI starts where rule-based automation ends" – which makes sense from our viewpoint but should not be generalized (there are tools that go beyond pure automation, e.g. Obviously AI for analyzing tabular data at scale).
More practically, whether AI should be used or not is a question of whether there are other solutions that can do the job at the same (or higher) level of quality, cost, or speed. If so, they are generally better suited to do the job. AI is (still) inherently fuzzy due to not being explicitly programmed to do x.
At the same time, explicit programming often leads to problems when there are simply too many rules or exceptions to be considered. In that case, AI often works better. For example, it is certainly possible to set up rule-based automation for processing text by using a long chain of words and phrases but in many situations, this wouldn't be efficient due to high costs or poor performance.
The promise of no-code AI
A vast amount of AI and Machine Learning companies claim that they democratize AI and this is probably true for their respective target users, which often times are still regular engineers. Out of all these companies, those that are building no-code tools get the closest to the ideal of "any person without prior training".
This increased level of democratization seems overdue: It has been proven time and time again, the majority of businesses struggle to implement AI is its full potential and scale, making the ease of this trade-off even more crucial.
Easy-to-use ML platforms leverage the time/value/knowledge trade-off in a genuinely attractive way and allow users with no AI coding skills to optimize day-to-day operations and to solve business issues.
Visual, often drag-and-drop, no-code AI tools make AI less intimidating and more comprehensible to non-technical people or those who lack the time or resources to build such systems from the ground up.
Besides this, there are some additional advantages to no-code AI:
- Accessibility: No-code AI enables organizations to make use of AI in the first place and can act as the stepping-stone towards intensified use of data science or AI in the future. The comparably low investment paired with people building up hands-on knowledge of AI tools mitigate the biggest obstacles to AI adoption at small and mid-sized companies.
- Usability: Plug-and-play allows for anyone in the organization to find an AI solution to a problem, and more often than not, in a budget-friendly way. These tools are built with non-technical users and non-developers in mind.
- Speed: The best no-code AI platforms allow users to iterate through the whole value chain of machine learning quickly. This allows for more rapid experimentation to see what can be done using one's own data – and getting back to business right afterwards. There is no better way to convince someone than to show them the process in a simple, intuitive way.
- Quality: No-code tools are built for people who may not possess a technical degree or even deep expertise in the subject to begin with. This requires extensive work going into the product as sane defaults and safety measures need to be carefully chosen on behalf of the user. To further mitigate such risks, some AI platforms have human review built-in and ask for input when needed. This combination reduces the human error when setting up such systems in the first place and allows direct interaction with the platform during daily operations.
- Scalability: AI itself doesn't care whether it performs a task for a single or a hundred users and neither do servers that are automatically scaled up or down, depending on the load.
Mapping the no-code AI landscape
Besides providing a current snapshot of the industry, it might also help better understand subtle differences between seemingly similar tools. For seasoned ML practitioners, this may be obvious but no-code tools are addressing a less technical audience by definition, so there's that.
While observing the field, we noticed that two dimensions stood out:
- Use-case specific versus agnostic generalists: Companies either build their business models around a specific industry and use case (e.g. Accern) or leverage the fact that companies across industries have a similar problem and lack similar AI development resources (e.g. MonkeyLearn, Levity).
- What data types can be processed: AI isn't to be confused with stew – just throwing a bunch of data into it won't give you what you want. Therefore, a key question is what data a company focuses on in the first place – the most important types being images, text, documents, or structured (tabular) data.
No-code AI is still a rather growing market – and most companies who operate in this space tend to have positioned themselves in technologies (NLP, Voice Recognition, Computer Vision) vs specific use case management (classification problems, CRM, web-builders, business apps). It is often hard to draw the line where one application ends and the other starts – especially when we look at AI applications. To get a clearer picture, we decided to take a deeper look into no-code AI players, and what they offer. The list below is in no way exhaustive, nor in any particular order (well... alphabetical), and we will keep on adding new players as they come – but bringing some structure into the landscape was a necessity.
What made the most sense to us was grouping based on core value proposition – we know that many of these companies are active in more than one scene. Leveraging the no-code movement to become a maker is fantastic – but we need to know what we want to create in the first place.
In a nutshell, we took the following criteria to qualify as no-code AI to heart:
- Tools that enable users to build solutions from scratch, which would have previously required one or more (ML) engineers to build.
- Creates value on its own for users and companies of all sizes – and is not just an enterprise-level developer tool (think Uber's Ludwig).
- Usable by non-technical people – this is essentially the core of the no-code movement. More importantly, this is one of the criteria we had the longest debate on. The level of knowledge plays a key role - and while there are tools like MS Azure, or C3 AI Suite, or even deepCognition - they are not built for the average knowledge worker, but for people who already know what they are doing in the developer stage.
We'll take a brief look into a selection of these tools.
Clarifai is an NLP and Computer Vision tool founded in 2013 that offers an end-to-end solution for modeling unstructured data for the entire AI lifecycle. Image, video, and text recognition solutions are built on an advanced machine learning platform and made easily accessible via API, device SDK, and on-premise. Boasts with accurate and detailed results with a fast API, they have some neat pre-trained models on offer (people, vehicles, and general detectors).
Since 2018, Apple has let developers deploy custom ML models by using transfer learning. The independent app allows users to build models for object and activity detection, image, video, sound, text, and tabular models. Create ML leverages the machine learning infrastructure built into Apple products like Photos and Siri but also allows training with custom data, train multiple models using different datasets simultaneously.
AutoML is the Google package star, and the tool works much the same way as CreateML – just on the cloud. The model package currently includes Sight (Vision and Video Intelligence, the latter in beta) and Language (NLP & Translation) as well as structured data (Tables) functions. AutoML overall manages to cover a lot of ground already in no-code – but once again, if you’re not a developer it’s hard to operationalize.
The DataRobot enterprise AI platform democratizes data science and automates the end-to-end process for building, deploying, and maintaining AI. Founded in 2012, it’s core focus is predictive models and is powered by open-source algorithms and available in the cloud, on-premise, or as a fully-managed AI service.
Levity focuses on image, text, and document classification and enables users to train custom models on their use-case-specific data – and is meant for businesses of any size. Custom models and flows also include a human-in-the-loop option, so users have full control, as the model asks for input where it is unsure – and will automatically learn from interactions. Levity focuses on providing an end-to-end solution and integrates with all the tools people use on a daily basis.
Lobe, a product by Microsoft, offers image classification, with object detection, and data classification coming soon. Lobe is a free, private desktop application with a fair amount of pre-trained solutions (e.g. Emotional Reactions which allow your app to react to different expressions allowing people to send emoji reactions using just their faces).
Founded in 2018, MakeML is a developer tool used for creating object detection and segmentation models. They have some great tutorials out on detecting a number of touch-points for sports games e.g. tennis and football and segmenting potatoes for something different - like chips production. MakeML has a dataset store, with a ton of free options, but also allowing platform users to sell and buy datasets from each other.
MonkeyLearn offers an all-in-one text analysis and data visualization studio, for unstructured text-based data to get topic, sentiment, intent, keywords, etc. Features include automatically tagging business data, visualizing actionable insights and trends, and simplification processes for both text classification and extraction. Integrates with Zendesk, RapidMinder, and Google products, with a whole bunch more coming soon. Also - in our humble opinion - one of the best blog resources out there when it comes to text analysis.
Nanonets falls in the Computer Vision domain - they have ready to use solutions for most common document types, but offer a set up for custom models as well. One of their cooler solutions offers to build an ID card verification model for any country, format, or language – including perspective transformation, meaning models that can work with tilted or angled images.
Obviously AI, founded in 2019 uses NLP processing to perform tasks on user-specific text data. Drag and drop your data as CSV or integrate with HubSpot, Salesforce, or MySQL (among others), pick your prediction column, and it'll auto-build a custom ML algorithm and you’ll end up with a prediction report. The platform is especially useful for SMEs, who are looking for a tool that chooses the right algorithm for their needs.
RunwayML is a tool specifically designed for creators and focuses on creative work involving interacting with images, videos, text, latent spaces, and segmentation masks and supports motion capture, object detection background removal, and style transfer. They offer a Generative Engine – a storytelling machine that automatically generates images as you write.
Teachable Machine is another Google tool – and a quick-and-dirty one. A web-based tool that offers creating models to classify images, sound, and body postures, and has an easy drag-and-drop UI. You can teach your machine by creating a data set with your webcam, live in your browser, and run a decent model with 30 images per class. They run pretty cool projects with TM - check out Project Euphoria here. The only downside, really – you can also export the model file and work with it someplace else, which can be a bit painful for no-code building.
Use cases for no-code AI
“What can I do with it?” is arguably the most common question in this space and there is a good reason for it: By definition, the primary user group of no-code AI consists of non-technical people. They may know a thing or two about AI but they are certainly not dealing with the subject on a daily basis, let alone code neural networks for a living.
As it turns out, the quickest way to grasping the usefulness of AI as part of business operations lays in studying a few use cases. That’s when the “aha” moment usually happens. You can find a bunch here.
Note that some tools imply the use case by way of how they have been set up (e.g. for a specific industry or process) while others are meant to be trained by the users with their specific purposes. A few platforms offer both. And naturally, there are different application layers at play – classification, tagging, detection, data extraction... the list goes on and on – and so do the possibilities.
Nevertheless, there are things to consider...
One of the myths in the no-code space is that if you want to get to the stage of any solution implementation, you have to lower your expectations. The days of where we had to choose two-out-of-three between fast/cheap/good are numbered, but expectations do have to be managed.
The current no-code AI space shows that each solution is intrinsically bounded to the design of the tool. Some practitioners point out that in some cases, it is important to remember that once you have developed an application on a platform, you are linked to that platform for as long as the application is running. In the context of a PoC, this is not a problem, but in the context of an application that is expected to last, things can be different.
And even though no-code platforms mitigate engineering and coding complexities, it is not a magic tool that can be used for everything. Instead, you should consider (as a process-owner) some of the following questions:
- What problem am I trying to solve? What tasks make up this problem?
- What is the level of project management that we need?
- What is the role of the tool/platform in the company architecture?
- Does the platform fit the problem needs?
- Is using a no-code AI tool a strategic choice that will drive value in the long run?
What will the future bring?
Businesses are steadily moving towards no-code platforms for a number of reasons. Partially due to the ripple effect on workforce management, access to developers and software engineers slows down project delivery – and this is where technology can add real value. Not only enabling your workforce to deliver solutions but also staying relevant and competitive in the current landscape is the unicorn we all want to catch.
Research shows that it is estimated that nearly 65% of application development will be done through low-code and no-code platforms as soon as 2024 – and no-code AI will play a significant part in this. It is hard to see the logic of doing things the traditional way when disruption of current process management is possible and widely available to everyone.
Nevertheless, useful AI applications require a good use case, to begin with. Just having an AI model is worth relatively little, regardless of how powerful it is. But just as people have found a new love for databases (thanks, Airtable!) and Wikis (Notion), people are going to pick up on the potential of AI. Just as no-code AI tools will mature, so will their users.
This article has been written in collaboration with our amazing team members Arne Wolfewicz, Hanna Kleinings & Thilo Hüllmann.
Note: We aim for a complete overview of the field. If you are missing a tool, please let us know and we add it for the next iteration!