AI, Old and New — Trends from an Enterprise ML Platform (Zehra Cataltepe from Tazi AI)
Zehra shared her experiences and insights into the AI Solution industry and her start-up building journey. She mainly talked about interesting trends in the industry and the challenges she’s facing in 2022. We provide an extended version of the conversation for 2024 in the midst of the GenAI boom.
About the company:
TAZI AI, established in 2017 in San Francisco, is an AI, Generative AI and Machine Learning (ML) Platform that enables business experts (and data scientists) to easily and rapidly create, update and put into production AI based solutions to make smart business decisions in continuously changing business environments. Thanks to TAZI’s patented continuous learning, explainable AI, and human-in-the-loop technology, TAZI has been chosen by Gartner as a Cool Vendor in Core AI Technologies, an honorable mention in Cloud AI Developer Services MQ and is cited in more than 30 analyst reports for its Responsible and Adaptive AI. Currently 14 enterprises mainly in financial services and retail industries are using TAZI to create huge value.
About the Founder
Zehra, CEO and Co-founder of TAZI AI, got her doctoral degree from Caltech with a specialization in computer science. Having worked as a professor and researcher at several renowned institutions, Zehra is an expert in AI and machine learning. Now she aims to help transform AI into a technology that can be used easily and useful for everyone in the enterprise.
Why we invested in TAZI:
TAZI has a strong product with demonstrated results in a growing industry — it has consistently achieved large increases in customer retention in banking, financial services and insurance clients. The founding team is also incredible: both PhDs, both with deep technical knowledge and strong sales orientation. The company has been built deliberately, with a relatively low burn for the amount of product and customer development.
What is TAZI AI? What was your inspiration?
My co-founder Tanju and I decided to establish TAZI because we realized that AI solutions were being created, but they were not going into production easily, and it was very difficult to maintain and update them while the business world changed. So we created an AI platform that could learn continuously and automatically from data as the world changes. We also wanted to enable business users — non-data scientists, non-coders — to be able to understand how machine learning works, and also control the AI solutions to monitor and update them without having a PhD expertise. So we created easy to use interfaces, and hence transparency, for business analysts and business experts to understand how the particular AI model worked and also how the whole solution was created. Finally, we made the updating a particular model and also the way the solution was created, by business users to be easy. AI solutions that learn continuously automatically to cope with new business conditions, by means of learning easily from new data and human knowledge, AI solutions that are understood, controlled and owned by business users is what we are better at and getting better at every day.
What is the most interesting trend in the AI Solutions Space right now?
- Responsible AI. AI models can not be responsible, people can. SaaS products like ours, have to be no code and easy to use, but also code and easy to deep dive into so that both teams on both business and data side can take responsibility. We are shifting a lot of the model creation, update, verification and deployment processes to business analysts and users, so that data scientists and IT teams can spend time on other tasks and also business users are always aligned with what AI does and they can take responsibility and drive AI solutions to get to their goals. In addition to that, our continuous learning allows models to consume less energy because when in continuous learning, instead of recreating models from scratch, we are updating them. This requires a lot less computation and also a lot less data movement, which means you can save a lot in terms of the costs of computation, storage and energy consumption, and that’s going to be a big benefit for large enterprises and the planet.
- Adaptive AI is another trend. Think of the world after COVID, everything keeps changing. Usually, machine learning needs lots of data to learn well, but if there is change happening, then there isn’t really a lot of relevant and useful data that is at your fingertips. As a result, having models that learn while changes happen is important. At TAZI, the models can be trained automatically with the new data, so that data and business teams don’t have to create new models but they can edit the updated model to the current needs in the new world. One of our observations is, customer facing teams, outreach, support, sales, marketing, fraud teams, they are always always in changing environments due to market, competition, product and they need continuously learning adaptive AI, but responsible AI that they can trust all the time. You need business eyes, legal eyes, and people with lots of business expertise. This is a responsibility in the sense of democratization of AI, which means non-data scientists, non-coders, and business users can be included in those processes. I am proud to mention that, in 2023, we were referenced in 15 Gartner Hype Cycle reports as a Responsible and Adaptive AI platform (https://tazi.ai/awards/).
- Composite AI and Generative: With the advancement of Generative AI and our ability to compose GenAI, AI, ML and domain experts together, we are seeing very important changes in the way AI solutions are created. First of all, especially when the business tasks solutions need to provide are divided into steps and then those steps are handled using AI agents that are controlled by business users, we are seeing an increase in accuracy of models and also a decrease in size of data needed and model complexity needed. Sometimes, an open source model, like llama3 or phi3, can perform as good or better than GPT-4 when the task is divided like this, LLMs have a chance to improve the initial results they provided with agentic flows and human domain experts are able to provide feedback. Another benefit we are observing is the reduced resource requirements. When the task is divided up and each part is well defined, sometimes, you may not even need an LLM, a deep learning model could be just great. This reduction in resource requirements enables us to perform some of the computation utilizing the CPUs as opposed to GPUs. Finally, there is a huge revolution going on in terms of the labeled training data requirements. From the day we created TAZI, we have already been utilizing Human in the Loop, particularly, active learning algorithms and co-training algorithms, to enable the AI models to ask for labels, from domain experts or other AI models, where it would help them to learn the most, as opposed to asking for labels for tons of random or just available data. We were able to reduce training data requirements by 75 to 90% for different tasks. With the LLMs, we are able to create the initial solutions with zero training data and just a description of the task with examples. This is amazing! One of the reasons AI solutions take a long time to create is, you have to ask for training data from the past. Especially for transactional systems in financial services or banking, sometimes the data may not even be saved in the best format for AI to learn and it might be very difficult, if not impossible, to transform the past data. If you are in a dynamic market, the world and data keep changing anyway, so not all the training data you prepared is even relevant. AI solution teams lost a lot of time and patience with training data preparation when GenAI wasn’t in the picture. With GenAI, the initial solution uses the domain expert’s description and examples and an LLM for the initial solution. Then the solution is tuned based on the feedback of the domain expert or another LLM on specific selected examples. This is propelling our already rapid solution development cycles. For solutions such as customer complaints, topics or sentiment prediction, we are able to deliver the initial solutions in days and now we are aiming to reduce this to hours or minutes.
What is the role of investors?
I believe that investors are not just investors, but accelerators of learning in a startup’s journey. Foothill and our other investors have provided us with opportunities to be in the same room with all the partners and other entrepreneurs and learn from each other so that we can accelerate the growth of our company and learn from our and each other’s mistakes.
What are your main challenges?
Our main challenge right now is to continuously update our product-market-fit in evolving AI and GenAI landscapes. For example, currently we are targeting the customer facing teams in banking, financial services and insurance industries. We chose these industries because the business users in these industries are great potential users of no code SaaS platform providing AI for them, because they already have experience with automation, analytics and even some AI modeling. We are able to keep our product expansion, especially with GenAI, while the market changes, because we have constant communication between our customers and GTM and product teams. I can not believe how much we have learned and evolved. An efficient product market fit process is the number one challenge and also the number one opportunity for a startup like ours. Because the vendors that are already established or without the AI/GenAI platform and the expertise that we have or without the customers that we have, simply can not keep up with us.
The other challenge is really networking with partners and customers. There are critical partners in every vertical and in every solution we provide in that vertical, so finding partners who are a good fit for our technology and our company culture is very important. Warm and trusted introductions to potential companies who can benefit from our AI and GenAI is another challenge and request from our investors as well as anyone else who is in TAZI’s ecosystem.
What has been the most difficult or contrarian decision you’ve made?
The most difficult decision in my opinion was moving to the US. We are a husband and wife co-founded company, and initially, I thought I could keep traveling between Istanbul and San Francisco. In one year, I had 12 entries to Istanbul. I can’t imagine my carbon footprint in that year, and really feel bad about it. Also, I couldn’t keep up with the time zones. Consequently, I decided to move totally to San Francisco, to really create the company and network. In order to learn how to build a company in SF, I had to be here personally, and that really paid off.
Admittedly, it wasn’t easy. Having a family on two different continents of course is not easy, but I think it was the best decision we have ever made. So my advice would be, first have a complimentary and working co-founder team as opposed to a single founder. Second, if you have multiple teams in multiple geographies, make sure that everybody can communicate and contribute constantly and can collaborate openly. People are working in a startup like ours, because they see that they are contributing to solutions of very important and difficult problems, their efforts are making huge changes and are also being acknowledged by the rest of the team.
To watch our full interview with Zehra from last year, check out the video on our Foothill Ventures Youtube channel: https://www.youtube.com/watch?v=DpfvOuQC3oc
To get updates on this series, please follow our publication on Medium.
Follow us on LinkedIn: https://www.linkedin.com/company/foothillventures
Foothill Ventures is a $150M seed-stage technology firm located in the Silicon Valley. We back technical founders across software, life sciences, and frontier technologies.
Questions, thoughts, reflections? Let us know in the comments below. We’re always looking for great entrepreneurs and early-stage ideas, and we’re always interested in having a discussion about venture, technology, and anything related. To see more about Foothill Ventures, please visit our website: foothill.ventures
