11 Feb 2020|by Rocio Wu
Although adoption of artificial intelligence (AI) and machine learning (ML) for the enterprise is still in the early days, the technology has matured enough for entrepreneurs to start gathering inspiration and evaluating opportunities for potential applications.
Every day, processing capabilities for neural networks increase, as does accessibility to off-the-shelf APIs from big tech and academic institutions that help speed up innovation. Entrepreneurs have also learned the wisdom of targeting AI applications toward specific, well-defined business problems, rather than trying to sell generalized toolkits to business users or get bogged down in custom software consulting engagements that solve nonrepeatable use cases.
There are more opportunities today for entrepreneurs to focus on solving vertical problems than creating horizontal solutions. It is in the ethos of established tech companies to build generic solutions for customers across industries. But for challengers, the more they can focus on solving core business problems, the more successful they will be. They can understand customer needs deeply and customize product features based on their client’s specific pain points. As a result, the customer will see more business value in the solution and be more likely to convert to a paid customer. (An added bonus for the startup: lower customer-acquisition cost.)
Still, it’s proven difficult for entrepreneurs to write successful AI strategies. After all, the technology is a moving target, potential customers are wary of costs and implementation complexities, and use cases, while powerful, are still lacking in many areas.
Take advantage of uncertainty
All this uncertainty is a fertile breeding ground for entrepreneurs ready to make their mark at the start of the digital enterprise era. Here are 10 rules of thumb to consider as you develop an AI strategy in an established company or sinking roots as an AI-first venture. Some are unique to AI-first startups, while others are generally applicable to enterprise Software as a Service (SaaS) companies.
- Understanding the business problem you are solving is at least as important as your algorithms. Even though the technology and data science behind AI is what makes these applications different from conventional software, your business customers are not looking for technology per se. They want solutions to solve problems. Positioning your service or product as “AI for health care” or “AI for sales” is not nearly specific enough. While you can sell AI tools to data science teams or IT departments, business leaders want to know you understand their problems and opportunities intimately, and that your solution is tailored for their situation. Artificial intelligence should enable better solutions.
- Is the market ready to support your solution? As more of the world gets digitized, “datafication” (our ability to capture data from many aspects of the world have never been quantified before) continues to accelerate. The problem: Especially in legacy industries such as health care, manufacturing, and agriculture, much of the data is not digital and unstructured, increasing significantly the effort required to extract, clean, normalize, and wrangle. Before starting an AI strategy, determine the digital maturity of the industry in terms of adopting AI. Is the industry mature, with infrastructure already in place to collect data and ready to implement? Are industry stakeholders willing to adopt AI? Do they agree with you on the value of potential benefits? Is the product intuitive enough to speed users through the learning curve? Are there regulatory hurdles to overcome?
- Develop your data strategy from day one. Training machine-learning models to improve based on experience often requires large amounts of high-quality data, so it’s extremely important to lay out your data strategy from day one—including how things like data sourcing, volume, diversity, privacy and security will be handled. Data can be acquired in a number of ways including crawling public data, finding data-rich partners, gathering it from customers, or generating it internally. Each has its own pros and cons and their application may be best suitable at different stages. Data strategy is a strategic business decision that entrepreneurs need to define from the start.
- Even if your AI is brilliant, your product still needs great user experience (UX), the right workflow, and robust reporting. You will win not because your AI is superior, but because the end-to-end product is better. Focusing on serving your customers’ end users should be baked into the team’s DNA. In most cases, you are building more than the ML product, so collaboration and coordination is required across functions and among both frontend and backend software engineers and UX designers.
- AI can be magical, but sales still win. Fred Wilson, venture capitalist and popular blogger, holds the view that “marketing is for companies who have sucky products.” Similarly, many AI founders believe that if the product is amazing, it should sell itself. However, that’s not the reality of the enterprise world. Big enterprises by default are averse to change unless they are convinced the alternative is worth their business development effort and the time involved of the legal-finance team to negotiate contracts and switch to a new vendor. Value experience in your sales and marketing team, especially if they come from the industry or companies on your target list. Nowadays functional leaders have more power than in the past to decide which software to use, so founders need to figure out an entry point to the enterprise.
- Be careful about “AI-first” messaging in marketing. Given the hype around AI that has raised everyone’s curiosity, an AI-forward positioning might be an effective strategy to get a first meeting. However, when it comes to actual buying decisions, customers do not really care whether it has AI inside or not. Some startups have actually removed AI from their marketing and sales messages. While it might not make sense to lead with AI, there’s value in weaving it through the product presentation, especially when it comes to transparency and explaining the underlying machine-learning algorithms.
- Avoid the “science project” trap. What’s the MVA (Minimum Viable Algorithm)? As the saying goes, “perfect” is often the enemy of the “good.” Business strategy has come to embrace the power of injecting a minimally viable product into the market quickly and iterating based on real-world feedback. Likewise, AI projects should similarly strive to develop and quickly market a minimum viable algorithm. This approach may take some convincing; the DNA of founding technical teams is often around solving technical puzzles and increasing accuracy from 90 percent to 95 percent. Many customers are not going to see the difference, but they will notice as the product improves from release to release.
- Manage customers’ over- and under-expectations. When it comes to successfully deploying AI in the real world, half of the battle is over expectation management and communication. Customers often overestimate your AI’s effect, thinking of it as superhuman—especially if AI is solving complicated problems such as 100 percent accuracy in self-driving cars and medical diagnosis. You have to help them understand that the performance of ML products improves over time (it’s machine learning after all) and unlikely to deliver perfect results in the beginning. Underestimates can happen as well if AI is solving a constrained problem like back-office automation or insurance claims. Helping customers understand which problems can and cannot be solved with AI is vital, just as Lemonade Insurance, which uses AI and other technologies to determine coverages and set rates, explained very clearly to potential customers how their product worked and what was within coverage and what was not. AI is still a very imperfect technology that often fails. There should not be any surprises for customers on that score.
- Hire both tech and domain experts from industry. Your team needs both ML engineers (often PhD level) and top software engineers who can productize and deploy AI. (Ideally you want talent who can do both, but good luck finding them!) There’s a limited supply of ML engineers and big tech companies will pay dearly for a brand new PhD in deep learning. It’s hard for startups to attract top AI talent, but even harder for Fortune 1000 companies. However, attracting domain experts from the traditional industries you are trying to disrupt is even more important. They are critical in helping target customers, deploy technology, and understand the input needed and the internal workflow used by businesses in order to trust AI’s judgments and validate results.
- Organizational shift: Towards a more open and experimental culture. ML/AI engineers are still a novelty in the enterprise world. Managing an AI-first startup requires fundamental organizational changes: an experimental culture, data analytics-driven mindset, and more openness towards uncertainties. As a founder, you should help cross-functional teams understand how ML products are different from conventional software products, address potential conflicts, and promote a more open and experimental culture.
Just the beginning
Artificial intelligence continues to evolve at a breakneck pace, creating plenty of ground-floor opportunities for entrepreneurs who are disciplined in their approach, identify best markets for vertical solutions, hire talented and experienced teams, and who are successful at selling AI not as a technology but rather as a means to the best solutions.
Special thanks to Brian Ascher at Venrock for valuable contributions.
Rocio Wu is a second-year MBA candidate at Harvard Business School and a venture capitalist.