Ari Jacoby is the CEO and cofounder of Deduce, a leading provider of cybersecurity solutions powered by real-time customer identity data.
The AI boom is transforming industries in remarkable ways, with startups working to revolutionize sectors like healthcare, finance and beyond. But while the opportunities seem boundless, launching an AI company comes with its own unique set of challenges.
I’m on my third venture-backed startup, with two successful exits. Each of the businesses I have founded has required the assembly of a unique data asset. As the old saying goes, “data is the new oil,” and AI companies are trying to turn that “oil” into useful and profitable businesses.
Building a unique data asset is not for the faint of heart, and I’ve certainly stubbed all 10 toes figuring this out. Here are just a few things I’ve learned from my own AI startup journey.
1. Data Is King
Data is the core of every AI company. Without high-quality, vast datasets, it’s impossible to build effective models. Founders must determine whether the necessary data exists or needs to be generated. Who controls the data? Is it scalable?
Accessing a diverse, high-quality dataset is crucial. Relying on a single source is risky, and pulling data from multiple providers can result in inconsistency. A startup’s choice of dataset can determine the bias of its models in the future. AI companies need strong data strategies to ensure a steady, clean and scalable data flow to feed their models.
2. Copyright Compliance Is Key
Training AI on copyrighted data without authorization can lead to lawsuits. Companies like OpenAI have secured licensing deals for content, but smaller startups might not have the resources for such agreements.
It’s essential to have a legal strategy in place to navigate copyright issues. If training data includes copyrighted material, obtaining permission or finding public domain data is critical to avoiding legal headaches that can derail a startup early on.
3. Strengthen Data Privacy Policies
Compliance with data privacy regulations like GDPR and CCPA is a crucial challenge for AI companies. These laws impose strict requirements for collecting and handling personal data, and violations can lead to hefty fines.
AI companies must implement strong data privacy policies, ensuring user consent, anonymizing sensitive data and adherence to privacy laws. Hiring a legal expert or consultant early on can help small startups navigate the ever-changing regulatory landscape.
4. Govern Compute To Build Infrastructure
AI companies need robust computational infrastructure to handle large-scale data processing. This typically requires significant investment in cloud computing services like AWS, Azure or GCS. Building scalable infrastructure that can process increasingly large datasets becomes a complex, time-consuming effort.
Many startups begin with cloud services to reduce initial costs, but as the company grows, the computing side of data processing gets expensive quickly. Access to specialist hardware, such as GPUs or in-memory processing, adds complexity and costs. Governing the resources of a company’s computing environment is essential in starting an AI company.
5. Ensure Data Redundancy And Monitoring
Data loss is catastrophic for any AI company. Ensuring data redundancy—storing backup copies in different locations—helps protect against system failures, cyberattacks or accidental deletions.
AI companies must also implement continuous monitoring to ensure data pipelines are running smoothly. A robust data pipeline automates the flow of data from its source to its final destination, whether a storage database or a machine learning model.
6. Develop Models That Continue Developing
The success of AI hinges on the quality of its predictive models. These models rely on historical data to forecast trends or behaviors. Developing effective models requires a balance between complexity and interpretability.
One key challenge is overfitting, where a model performs well on training data but fails to generalize to new data.
7. Manage Model Risk
Model risk management in regulated industries provides oversight of the provenance of every data source. In financial services, for example, it identifies, assesses and mitigates risks associated with using models in lending decisions.
Designed to ensure the data sources and the resulting model do not introduce bias—where the model may discriminate one demographic or socio-economic group over another in credit scoring, loan approval or pricing—MRM maintains compliance with regulations supporting fair lending practices, shielding institutions from losses due to poor lending decisions.
8. Form Partnerships For Data Insights
Collaborating with data partners can give AI companies access to unique datasets and insights, enriching their models.
Trust and clear expectations are key to forming successful data partnerships. Misaligned goals or legal complications can quickly sour these relationships.
9. Secure Valuable Data And Data-Sharing Deals
Access to valuable, real-time customer data gives AI companies a competitive edge—if and when they prove their business impact to potential partners or customers. Having developed a model to solve a specific business problem, a company must involve sales and marketing to sell the solution to a business and also foster customer success. If a solution is built on the consortium give-to-get model, an AI company must convince the customer that the solution is safe, secure and beneficial to secure the customer’s crown jewels of data.
Securing data-sharing agreements requires long-term relationship-building, trust and legal negotiations. Companies must act responsibly to avoid privacy violations. Mishandling customer data can destroy trust and lead to regulatory consequences.
10. Retrain And Refine Over Time
AI models are never truly finished. Data evolves constantly, and models must be continuously refined to stay effective. This means regularly retraining models, fine-tuning algorithms and testing for bias and inaccuracies.
Refinement is an ongoing, resource-intensive process for which founders must account. Without continuous improvement, models risk becoming outdated and less effective in the face of new data.
A Journey Like No Other
Despite these hurdles, the potential to revolutionize industries and create groundbreaking opportunities is immense.
Through my own experience, I’ve learned that founders who grasp the importance of data, infrastructure and constant model refinement, without losing sight of the business impact value they are trying to deliver to their customers, are the ones best positioned to succeed in today’s competitive AI landscape. It’s a tough path, but with the right approach, the rewards can be transformative.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Read the full article here
