.png)
A meeting with three elephants
One of our founding principles was that we wanted to enable non-technical users and smaller companies to benefit from AI automation. This is because there is a huge distortion in the market: There are companies that capitalize on AI and there are companies that don't.
There are a few reasons we found to be true in almost every customer interaction we had so far. Three of them stand out, which is why we call them our elephants in the room: "We cannot afford such people", "those projects take forever", and "AI is mostly hype".
In the pre-Levity time, all of them were true but as our product and company vision matured, we consciously thought about these factors and made sure they would be addressed.
We believe that it is a good idea to go into more depth on all three and tell you how these findings have inspired what we do.
🐘 Data scientists and engineers are expensive
People related to Data Science are in demand and employers are fishing in the same pond as some of the largest tech companies. To add a number to it, the online platform DataJobs found a salary range of USD 85-170k for data scientists.
Needless to say that it does not end with the base salary and to cut a long story short: Employing someone who can automate a process with Machine Learning is expensive and in most cases, the savings do not justify such investments – and certainly not for "seeing what can be done with AI". As a consequence, workflows remain manual, come whatever may.
To leave a personal anecdote: Before joining Gero and Thilo at Levity, I witnessed how this very cost aspect took away all potential of the technology even at a large company. Making a budget request for engineering capacity over several months was a big stretch without a crystal-clear savings commitment. This, in turn, made me choose to work on other issues instead, knowing all too well what inefficiencies we would be carrying along and what could be done.
Our philosophy around this at Levity is that cost should not be the roadblock when attempting to improve an existing process with AI. Our platform costs a fraction of that since we are building it for scale: Adding a feature means adding it for all. This allows virtually unlimited freedom to experiment with some powerful technology, without having to consult one of these people entirely.
🐘 Projects take months to complete
If we assume you had access to the right people, projects involving Machine Learning tend to run longer than others.
Contrary to "normal" projects around improving performance, machine-learning-related projects introduce additional complexity; they are not about process changes or "normal" software development, but both of these plus data. As such, the time-to-value of such projects is impacted by three components: Process analysis, data collection & engineering time.
With all three usually being highly intertwined and each team member's time allocated across five different projects, fast iterations are simply not a thing. And that takes all dynamics out of the initiative.
While we cannot speed up the data collection (except when we have an existing model already) and process analysis on your behalf, what we wanted to provide was a self-paced user experience that would not require additional engineers on your end. Instead, you – as a non-technical user – can control the whole process yourself until you know that the system works. And even if you cannot use one of our pre-built integrations, any entry-level programmer can integrate our API with the rest of your systems landscape.
🐘 The AI hype does not live up to its promise
We partly agree with this. However, most people who dedicate a bit more time to the topic will quickly realize three things:
- There is a discrepancy between those who talk about AI and those who make use of it.
- The technology has become incredibly powerful in some areas.
- Those companies which use AI productively in those areas can reap substantial benefits from it.
Unfortunately, many people get stuck at the first point and don't understand the technology well enough to make it work as desired. This is not meant in a diminishing way, quite the opposite: If more people knew the inner workings of AI and how it can be applied to their processes, the world would be a better place.
The reality, however, is that not everybody knows how to transfer a manual, time-consuming task into „that can be solved with image recognition“ – this simply requires some thought and/or a bit of practice. We want to make that practice as playful and intuitive as possible.
All the above can be answered differently, either by consulting a book or an expert. But most humans learn the best by doing and in a visual way.
Simply getting used to the typical workflow of teaching a machine is a valuable experience on its own and answers questions like "How much data is enough data?", "What happens if I have 4,700 data points for one category and only 350 for another?", "What happens if the AI is not sure?", and most importantly "Can I solve this issue I am encountering each day in a better way?"
Our approach to this was and is to focus our attention on a few things in particular:
- Great UX: Fewest possible touchpoints with AI technicalities, guided experience, lots of content on our website.
- No code: Lower the technical barriers to entry.
- Workflow first: Lots of pre-built integrations so that users can use the software as part of their existing processes.
You can judge if we are successful at all of the above.
Picking the right tools
Not every process is cut out for Machine Learning, let alone what we offer around handling unstructured data like documents, images, text, emails, and so on. We often advise our customers not to use our software when there is a better way to solve a problem.
We aim to offer something very specific for people who have an intuition for how a problem can be solved using Machine Learning but lack the skills, time, or both to go about it but instead seek to take action quickly and in an affordable way. This is (still) a small fraction of the population but we are just at the beginning.
