Connect with us

Thought Leaders

Navigating the AI Gold Rush: Unveiling the Hidden Costs of Technical Debt in Enterprise Ventures




Over the past year, artificial intelligence has captured the attention of enterprise leaders, prompting them to hasten their investments in AI companies or expedite the introduction of their own products in order to catch up. However, in the rush to join this new era of technological advancement, organizations who are new to AI may not be considering one important factor that should be top of mind when investing or creating new AI products: technical debt.

Though the idea of technical debt isn't new, AI technology brings about a different kind of technical debt compared to regular software services. And as AI continues to rapidly improve, it's causing this important issue to grow along with it.

What Is Technical Debt?

Technical debt, in the simplest definition, is the accrual of poor quality code during the creation of a piece of software. This typically stems from either an accelerated go-to-market timeline to meet business needs, or to get something out there in order to get customer feedback faster. When considering technical debt, it’s important to focus on the deliberate aspect of it, as decision-makers are often aware of the risks with software and the impacts of taking shortcuts for speed. The emergence of AI has brought on a different and unique challenge when it comes to technical debt, and with it significant risks and repercussions that could result.

As AI systems begin to age and their training data becomes inaccurate and outdated, the cost of investing in AI now outweighs the time and investment required to maintain high quality training data, otherwise known as data hygiene.

Let’s explore how technical debt is accrued, the impact it has on the bottom line, and how organizations can remedy it.

How Do Organizations Acquire Technical Debt?

 There are two ways software can accrue technical debt. One is through plain old bad code. Organizations may purchase products or inherit them through M&A activity, only to later discover quality issues on top of slow rates of change and innovation. The other is when leaders deliberately choose to take on technical debt.

When it comes to AI, just over 72% of leaders want to adopt AI to improve employee productivity, yet the top concern around implementing AI is data quality and control. It seems counterproductive for an organization to use a product promoted to increase productivity, while simultaneously detracting time away from the vital work to continuously address any and all quality issues caused by technical debt that may jeopardize productivity. But the promise of the eventual payoff for increased productivity outweighs these roadblocks in the immediate future, that will come back to eventually haunt the software in the long run.

Model Drift: A New Type of Technical Debt

With the emergence of increased investments in AI, organizations have rushed go-to-market strategies to cash in on the generative AI gold mine. While this may work as a short-term revenue driver, organizations are overlooking what could amount to a large amount of technical debt down the road, known as model drift.

Model drift occurs when an AI system’s performance begins to decrease and outputs become less accurate as training data ages out. Looking at the AI life cycle, it’s obvious that the training data will need to be continually maintained and updated to ensure the responses the machine provides are as accurate as possible—this is where the breakdown begins. When rushing to get solutions out, decision-makers often deprioritize issues such as obtaining additional training data, maintaining the system’s data hygiene, and ensuring there is a workforce that has enough people to support these tasks.

As training data continues to age and the gaps between reality and outputs widen, organizations will be left with increased costs and time spent on addressing these lapses that could have been avoided with proper planning procedures and protocols. In short: skipping the next step when planning a go-to-market strategy may allow for faster delivery, but it’s not worth the inevitable fall out that will cost in several ways in the long term.

Technical Debt’s Impact on the Bottom Line

Technical debt can also deeply impact organizational efficiencies — for example, consider sales teams. When technical debt starts to build and the rate of change slows, it becomes increasingly more difficult for sales reps to entice customers, which slows close rates and inevitably revenue streams as a result.

Beyond sales, technical debt also greatly impacts developer teams. Not only will it require more time spent focused on updating code, that averted attention effectively backburners innovation. By shifting attention and time to maintenance, the product roadmap then becomes delayed or abandoned, creating a ripple effect that could ultimately result in mistrust between the engineering and commercial side of the business. Without a product roadmap to follow, sales teams are left with either broken promises or nothing to show prospects, again greatly impacting revenue.

How to Address Technical Debt

As the predictability of delivery decreases, organizations will begin to see the breakdown of organizational efficiencies, leading to conversations about how to address the challenges at hand. There are two ways that decision-makers can leverage to combat technical debt. The first is throwing away the platform and code entirely and replatforming, or embedding small incremental changes, similar to slowly cleaning a bedroom one item at a time, to eventually get the systems up to speed.

The first method, re-platformization, requires a complete overhaul of your systems, and is a huge and costly risk to take. Similar to a large-scale construction process, any delays in scheduling can throw off product timelines and could cause the whole effort to fail. This method can work sometimes though. Take LinkedIn for example – after their 2011 IPO, the company replatformed the site and is now a huge player in the market.

The safer bet, making small changes that will eventually add up to major improvements, is another use case to argue for. With developers already interacting with data on a daily basis, going in to make tweaks here and there can shape up systems to be rid of their technical debt. It also benefits developers' skill sets, as it requires them to stay up to date with the latest code and technology standards, which in turn sets an organization up for technical success as they have fewer skill gaps. Implementing an engineer-driven initiative, where they are allocated 20% of their time to schedule for product updates, is a great way to get started. While this process is much slower than replatforming, it is less risky and still produces value to the business model.

Leave Your Technical Debt Behind in the Age of AI

As the AI space continues to rapidly develop, we’ll continue to see more solutions arising touting productivity gains and organizational efficiencies. While this is true, decision-makers must prioritize embedding techniques like continual data maintenance and think of the big picture when it comes to your solution’s life cycle. Investing in AI doesn’t have to be costly and overwhelming, and with a few small changes in planning and go-to-market strategy, you can avoid the next mound of technical debt.

Tony Lee, CTO at Hyperscience, leads the Product, Design and Engineering teams. He has held senior leadership roles at Yahoo, Box, Zendesk, and Dropbox. Tony began his 25-year engineering career at NASA, where he worked on automation software for air traffic control, and later continued his research into computer network optimization. He holds a PhD in Engineering, and a combined degree in Engineering and Political Science from Brown University.