Health startups are saying that unclear regulations are stifling AI innovation in the sector. Of course, such precautions are necessary in the healthcare industry, where it’s literally a case of life or death. But what makes less sense is the sluggish adoption of AI across enterprise SaaS – a space that isn’t being held back by red tape like other sectors are.
So what’s stopping enterprises from adopting AI to streamline and optimize their processes? The primary culprit is the hoards of messy data that accumulates as companies grow and add new tools and products. In this article, I’ll delve into how messy data is a blocker to AI innovation in enterprise, and explore the solutions.
Welcome to the data jungle
Let’s start by looking at a common data challenge that many modern businesses face. Initially, when businesses offer a limited range of products, they typically have clean revenue data that’s all housed within a single system. However, as they expand their offerings and adopt a range of revenue models, things quickly get messy.
For example, a business might initially employ a one-time purchase model, but later introduce additional options such as subscriptions or consumption-based pricing. As they expand, they’ll likely diversify their sales channels, too. A company that starts with 100% product-led self-serve sales may realize over time that they need the help of sales teams to up-sell, cross-sell, and land larger clients.
During rapid growth stages, many businesses simply stack new sales systems onto existing ones. They’ll procure a different SaaS tool to manage each different motion, pricing model, purchasing process, and so on. It’s not uncommon for a company’s marketing department alone to have 20 different SaaS tools with 20 different data silos.
So while companies generally start with clean, integrated data, growth causes data to quickly spiral out of control, often well before businesses recognize it as an issue. Data becomes siloed off between billing, fulfillment, customer success, and other systems, meaning companies lose global visibility into their inner workings. And unfortunately, manually reconciling data is often so labor-intensive and time-consuming that insights can be outdated by the time they’re ready to use.
AI can’t fix your messy data for you
Several prospective clients have asked us – “well if AI’s so great, can’t it just solve this messy data problem for us?” Alas, AI models are not the panacea for this data problem.
Current AI models require clean datasets to work properly. Companies relying on diverse sales motions, SaaS platforms and revenue processes inevitably accumulate disparate and fragmented datasets. When a business’s revenue data is scattered across incompatible systems that can’t communicate with each other, AI can’t make sense of it. For example, what’s labeled as “Product” in one system could be very different from “Product” in another system. This subtle semantic difference is difficult for AI to identify and would inevitably lead to inaccuracies.
Data needs to be properly cleansed, contextualized and integrated before AI comes into the picture. There’s a longstanding misconception that data warehousing offers a one-size-fits-all solution. In reality, even with a data warehouse, data still needs to be manually refined, labeled, and contextualized, before businesses can use it to produce meaningful analytics. So in this way, there are parallels between data warehousing and AI, in that businesses need to get to the root of messy data before they can reap the benefits of either of these tools.
Even when data has been contextualized, AI systems are still estimated to hallucinate at least 3% of the time. But a company’s financials — where even a decimal point in the wrong place could have a domino effect disrupting multiple processes — require 100% accuracy. This means human intervention is still essential to validate data accuracy and coherence. Integrating AI prematurely may even create more work for human analysts, who have to allocate additional time and resources to correcting these hallucinations.
A data catch-22
Nevertheless, the proliferation of SaaS solutions and resulting messy data does have several solutions.
First, companies should regularly assess their tech stack to ensure that each tool is strictly necessary to their business processes, and not just contributing to the data tangle. You may find that there are 10 or even 20+ tools that your teams are using daily. If they’re truly bringing value to departments and the overall business, don’t get rid of them. But if messy, siloed data is disrupting processes and intelligence gathering, you need to weigh its benefits against switching to a lean, unified solution where all data is housed in the same tool and language.
At this point, businesses face a dilemma when choosing software: all-in-one tools can offer data coherence, but possibly less precision in specific areas. A middle ground involves businesses seeking out software that offers a universal object model that is flexible, adaptable, and seamlessly integrated with the general ecosystem. Take Atlassian’s Jira, for example. This project management tool operates on an easy-to-understand and highly extensible object model, which makes it easy to adapt to different types of project management, including Agile Software Development, IT/Helpdesk, Marketing, Education, and so on.
To navigate this trade-off, it’s crucial to map out the metrics that matter most to your business and work back from there. Identifying your company’s North Star and aligning your systems towards it ensures that you’re architecting your data infrastructure to deliver the insights you need. Instead of focusing solely on operational workflows or user convenience, consider whether a system contributes to non-negotiable metrics, such as those crucial to strategic decision-making.
Ultimately, it’s the companies that invest time and resources into unjumbling the data mess they’ve gotten themselves into who will be the first to unlock the true potential of AI.
Credit: Source link
