Build or Buy, the Value of Critical AI Partnerships
If you’ve been keeping up with our AI Build or Buy series, you’ll notice that finding the right partnerships is at the core of each decision. The right AI partner serves an essential role of any finance technology strategy. They bring a level of expertise and dedication to the company’s data science needs that is often very expensive to grow natively and can help activate new resources quickly and efficiently. That said, these experts can also become an unnecessary escalating expense when organizations fail to communicate and utilize the vendor’s function effectively.
An experience in bad AI partnerships
A previous contact of mine felt the burden of a misdirected AI strategy. Their company had ambitions for a broad portfolio of AI projects, but lacked the expertise to hire their own data scientists and were hesitant to invest long term in the practice. Instead, they brought in an AI vendor to take point on their AI projects.
After two years, the vendor deployed three models, including one core to the business’s daily functions, but at almost three times the cost of in-house development. Recognizing the unnecessary expense, the company hired their own Chief Data Scientist to take over the projects. Unfortunately, the internal CDS couldn’t reverse engineer the vendor’s model and created tension with the vendor.
After losing three Chief Data Scientists and an ugly a dispute over the ownership of the model’s IP, the company and the vendor severed ties, and the organization was forced to scrap all three AI models and start again from scratch with a new internal data science team.
For my contact, what they saw as an easy solution quickly created unnecessary and expensive drama.
So, what can we learn from this?
- If you sign a contract with an AI vendor who will be writing custom code for you, ensure that you are aware of who will own the IP both during and after the project. If you believe that this model will be important for your organization to own and build on, you’ll want all assets and artifacts that result from it. However, if you don’t plan to take the models further in-house, assigning that IP to the vendor may not be a big issue.
- Always enforce strong documentation for every piece of code written specific to your data.
- Never rely on vendors as a replacement for creating critical in-house workloads.
There’s always risk that comes with bringing on new partners, but while my contact had a negative experience, a majority of vendors provide reliable, quality, and honest work.
The power of positive AI partnerships
A separate colleague of mine recently finished an AI partnership that saw their company execute an image identification model wonderfully. Initially, the company was collecting tens of thousands of images to train its model to analyze image data and identify objects in the images. Unfortunately, the images weren’t generalized enough to work effectively. They lacked the compute power, thousands of additional images, and appropriate sources for their use case. After three months and over $100,000 dollars, the company turned to Microsoft’s Computer Vision API and found a perfect solution for its model, delivering a completed deployment in an afternoon.
In this case, my colleague’s company was trying to solve a problem that Microsoft had already cracked. There are dozens of challenges that have been solved by AI in recent years—image analytics, sentiment scoring, optical character recognition, anomaly detection—and companies need not invest their efforts to reinvent the wheel.
These and other off-the-shelf solutions, along with the consultants that champion them, thrive as complimentary resources to your existing teams and fold nicely into more complex models. In a worst-case scenario, if the accuracy of these solutions isn’t good enough, try a transfer learning approach and refine, instead of starting again at zero.
What are our takeaways?
- Most AI challenges have already been solved. The right vendor can deliver what you need.
- Off-the-shelf AI solutions are a great compliment to an existing AI strategy.
- AI solutions should always be adaptable and usable, regardless of the engineer.
The right AI Partnership empowers your team with complimentary expertise, without using their expertise to hold your data and AI processes to their demands.
Whether you build a solution yourself, or grab one off the shelf, the right AI tools equip today’s financial services companies to expand their capabilities and services while drawing a richer understanding of the individuals and organizations we serve. Throughout this series, we’ve explored the benefits and risks and the highs and lows of both experiences, but the final solution will always hinge on context. There’s no one-size-fits-all approach for AI adoption and implementation. Every organization brings its own needs, experiences, and expertise, but the right AI partner can help you make the right decision.
At Microsoft, we are dedicated to serving as a stable, reliable AI partner for financial service companies of all sizes. If your struggling with your own AI implementation, or simply need consultation on your next project, feel free to contact me directly at [email protected]. Let’s make AI work for you.
