Will website owners pay for a platform that can automatically improve conversion? A platform that itself proposes hypotheses, makes changes to the site, and conducts testing of thousands of variants. Of course, they will pay for such a service! And instead of inventing such a platform from scratch, one can take an already ready-made example for copying along with its development plan 😉
Project Essence
Evolv AI calls itself “the first AI platform for website optimization, which suggests what can be changed on them, allows these changes to be made quickly, and tests the results of the changes swiftly.”
Or, in other words, the startup offers a “self-executing hypothesis testing process.” The platform operates on its own, and the website owner sees the results – how many variants of changes can be made, how many have been tested, how much the optimized metrics have improved, and what financial performance improvements this has led to.
The testing process consists of three stages.
First, Evolv AI compiles a list of changes it recommends testing – for example, “increase the font size of subtitles by 20% and change their color to #0099eb” or “reduce the size of the image on the first screen by 40% and shift it 35% to the left in the mobile version of the site.”
The necessary changes on the site can be made by the site owner using the AI assistant of the platform, without even needing to seek help from programmers. This is because the AI assistant has already analyzed the site’s code and knows where and what needs to be changed to achieve the desired effect.
In some cases, the AI assistant can generate several variants of changes, from which the site owner can choose the one they like with a single click – for example, which specific “happy dog” picture they want to see at the top of the site.
Next comes the experimentation stage, where Evolv AI presents different visitors with various new design options for the site and compares the resulting performance indicators with the control version of the old design.
In doing so, Evolv AI segments different visitor groups – at least mobile users are compared with mobile, and desktop with desktop. Probably nothing prevents segmenting even by traffic sources – because visitors coming to the site from, say, TikTok might behave very differently than visitors from Google search results.
Evolv AI has ready-made integration modules with popular statistics collection systems and website builders, allowing modification of sites created on these builders and gathering experiment statistics using the counters already installed on the sites.
One online store, using Evolv AI, tested 26 variants of 19 key website elements, resulting in three optimal designs that were supposed to ensure a 95% probability of increasing the size of an order by at least 5%. The client switched to one of the proposed design options and is now on track to gain an additional 1 million in revenue “out of nowhere.”
On the left of the picture above is the control (old) design variant of the order placement page, and on the right is the new optimized one.
A hotel conducted tests with Evolv AI on 4,608 variants of new designs, resulting in a 127% increase in booking rates on the site. The process of conducting experiments with the platform was 12 times faster compared to the traditional A/B testing process.
The startup released the beta version of its platform only this summer. Nevertheless, it managed to attract well-known brands for pilot testing, including NBCUniversal, Rakuten, Verizon, and others.
Following successful beta testing, Evolv AI raised $13.3 million in investments, adding to the $10 million it had received earlier.
What’s interesting
Minor design changes can indeed lead to significant conversion changes on a website. However, this only applies if there is already a conversion 😉 Struggling with design changes to sell an unwanted product is completely pointless 🙁
If the site is working and selling, there will always be many design change options – if you combine possible changes of all important elements, the number of combinations can indeed be in the thousands. Manually coding all these changes, manually conducting testing, manually comparing results, even with auxiliary tools, will take a lot of time and effort. But for AI, which is actually only good at quickly iterating and comparing options, this is a very suitable task 😉
Therefore, today’s startup has both a very good task and a very relevant solution method. So either it or someone else, or everyone at once, is sure to take off with this theme.
However, one should not be deluded about the current functionality of Evolv AI. After viewing the site, one might get the impression that they have automated the entire process of hypothesis generation and experimentation. However, if you read articles about the startup, it seems that in the current version, they only have three main functions working:
- “Text into design.” The platform can be given a single text instruction, for example,
- “change the font and color of subtitles in such a way” – and it will make the corresponding changes in all the necessary places in the website code. “Text paraphrasing.” The platform can rephrase texts on buttons, calls to action, headings in the required style – the owner will receive options, from which they can choose the one they like, and then the platform will insert it into the website code.
- “Changing images.” The platform can be asked to create new images in the required style, change the sizes and styles of images on the site, and it will offer image options and make the corresponding changes in the site code.
Nevertheless, the startup’s co-founder claims that “Evolv AI is currently working on automating the entire design thinking process – from understanding user needs and their navigation problems on the site to the process of proposing new hypotheses, breaking them down into a set of necessary design changes, making changes to the site, and evaluating the effectiveness of these changes.”
Thus, the impression is that the startup’s founders have outlined three stages of development for their platform:
- An AI tool that allows changes to be made to a website’s design without the involvement of programmers.
- An AI tool that can automatically test different design variants and produce a comparative table of their effectiveness. Simultaneously, it will collect and analyze data about changes made by website owners and the impact of these changes on conversions.
- An AI tool capable of suggesting changes to website owners based on the statistics gathered from all the platform’s clients, which could potentially improve conversion rates. Depending on the platform’s settings, these recommendations can be either manually approved by the website owner or automatically launched for testing. These are three logical stages that bring the product closer to its ultimate goal.
Moreover, at each stage, users receive a useful product that can greatly assist them even in its current form.
This step-by-step product development plan reminds me of another startup – Siro. Siro is creating an application that will use AI to train salespeople. Their product development plan is also divided into three stages:
- An AI tool analyzes the conversations of all salespeople, then identifies weak spots in individual conversations and sends them to the sales department manager, who in the application gives salespeople advice on improving communication effectiveness.
- The AI tool identifies weak spots but sends them directly to the salesperson along with similar examples from conversations of strong salespeople.
- The AI tool identifies weak spots but sends generalized recommendations for improving communication effectiveness to the salesperson, which are created based on general advice from sales department managers and examples of successful communication.
The need to move from the second to the third stage is explained by the fact that most people have poorly developed abstract thinking. Therefore, they cannot first derive general rules from others’ examples and then understand how to apply them in their specific case. Such people need to be given both general rules and others’ examples, as well as recommendations for applying these rules and examples in their own situation.
And these two plans remind us of the well-known “Tesla master plan,” which Elon Musk formulated back in 2006. It also consists of three points:
- Build a sports car.
- Use the money from its sales to build an affordable car.
- Use the money from its sales to build an even more affordable car.
Where to Run
I really like the idea of a “master plan,” where the process of creating the final product is broken down into several stages – but at each of which a standalone product is created that is valuable to users and therefore brings money to its creators. Moreover, these products smoothly transition into each other during the plan’s execution, gradually bringing the creators closer to the more distant final goal.
This contrasts sharply with the plans of most startup founders, which usually consist of two points:
First, we immediately (meaning for a long time) create a complex final product. Then we immediately (meaning find a lot of money) scale it. Implementing such plans requires a lot of time and money. And in the process, product creators do not receive any new information and learn nothing – because there are no intermediate products with which they can test the direction of movement and make changes in their functionality. The entire bet is made only on the fact that “at the very beginning, I was initially right about everything” 😉
So the first general conclusion – whatever product you have in mind, try to envision the process of its creation in the form of a master plan, consisting of the sequential release of intermediate products that are independently valuable. Whether this master plan consists of 3, 5, or 7 points is not so important. The main thing is that the principle itself is preserved.
Returning to today’s startup topic, automating the proposal and testing of product hypotheses (including design, as in the case of Evolv AI) is a very promising and relevant topic.
“Promising” – because all application developers, online store owners, and cloud service creators are constantly proposing and testing hypotheses, conducting various experiments. If we can significantly speed up, simplify, and improve the efficiency of this process – people will definitely want to pay for it.
“Relevant” – because such automation at a high quality level has become possible only due to the application of modern AI technologies.
So now the whole question is only about:
finding an area where such experiments are already being conducted, developing a master plan for creating the final product in this area, sequentially creating AI tools for implementing each point of the master plan. But if you don’t want to bother with searching for areas and drawing up master plans, then today’s Evolv AI can be taken as a ready-made example for copying. Since the potential market is huge – and far more than one such platform can work and make money on it at the same time.
About the Company
Evolv
AI Website: evolv.ai
Latest round: $13.3M, 09.11.2023
Total investments: $23.3M, rounds: 2