There's time and place for SMBs to adopt SaaS-based ERP
SMBs face many issues when implementing SaaS ERP operations versus on-premises, plus the fact that one size cloud-based ERP platform doesn't necessarily fit all companies.
Artificial intelligence has finally arrived -- at least in some applications. While AI's potential benefits to business, the economy and services of all kinds are incalculable, those benefits may not be equally available to all.
For most of the major SaaS ERP platforms, it's not so much a matter of cost. These platforms have at least some AI built in, and most are available to businesses of all sizes. It's more a matter of range and customizability. The canned analytics in SaaS-based ERP is limited and will only take users so far. And fine-tuning the output of these features against your own company's outcomes, let alone expanding the analytics, is cumbersome and resource-intensive.
Put simply, cloud-based AI can't be entirely relied upon. In-house analytics talent is necessary, and that's just really expensive.
Dangers of cloud-based AI
SaaS AI and analytics inherently suffer from the same weaknesses we encounter with IBM's Watson and Salesforce's Einstein. These cloud-based AI systems proffer largesse -- for example, "AI for everyone!" "It's like having your own data scientist!" -- while touting drive-thru convenience for smaller companies -- "Just drop off your data and we'll serve up some numbers!" Of course, it isn't that simple, and there are reasons to be wary of SaaS ERP.
First, we don't know what algorithms are at work inside the black box. Different types of models yield different results. There are countless regression methodologies, for instance, and each has its own strengths and weaknesses. Second, there's no real context for evaluating the quality of the results: Was the data clean enough going in? How can we know? How many passes will it take to generate a trustworthy result? It's anybody's guess.
Can't afford a data scientist
SaaS AI can't help us unless we can help ourselves. And that means having qualified people in-house who can set up an analytics problem correctly, stage the data appropriately and evaluate the results meaningfully.
But people with those skills are expensive and hard to find -- there are far more data science positions available in business today than there are data scientists to fill them, and SMBs have an even harder time hiring them. This problem will eventually self-correct as more professionals are trained for these roles and the tools they use improve. But that's no help to companies today.
Perhaps the greatest danger to cloud AI -- namely, Einstein and Watson -- is that vendors promulgate the notion that the days are numbered for true data scientists. That's about as real as the myth that all app development will soon be code-less. Reality dictates that analytics and applications aren't just coding; they need to be designed, which requires a deep understanding to bridge the problem to be solved with the technology that will solve it. We're a long way from machines with that depth of understanding.
And it's not just about securing skilled data scientists; it's also about infrastructure. SMBs are forced to accept the limitations of cloud-based AI because the build-out for in-house big data is cumbersome and expensive, requiring massive server resources, terabytes of storage and complex interfaces hither and yon.
One step at a time
I'm loath to simply plug Microsoft at this point, but it has the right idea -- at least in principle -- with its Azure Data Lake. The message here is similar to what we heard in 2005, when SQL Server started bundling data warehousing capabilities for free: Learn how to do it yourself, and take it one step at a time.
This concept is akin to Watson and Einstein taking us beyond the more canned features of SaaS-based ERP -- basically pay-per-view analytics. Infrastructure is rented instead of purchased, with terabytes available for big jobs, easily mastered implementation of your own code -- R, Python or whatever -- and plenty of canned AI features available to handle what you can't, including facial recognition, text analysis and sentiment analysis.
SaaS ERP is preferred when the data comes from many places and the AI and analytics solve real-time problems.
Then there's the pitching and catching of the data itself, which is also easily accommodated on both ends. Most SaaS-based ERP platforms are making big data import and export simpler with each release.
Microsoft aside, this general framework is the cloud AI and analytics future for companies too small to do it all: Rent big data space, take on the tool set you're able to competently master and farm out the rest -- and take it one job at a time.
Still, to say that SaaS-based ERP is practical because all the pieces, parts and expertise are correctly distributed doesn't necessarily mean it's always the best. There's a time and place for SaaS ERP and a time to stay on premises.
SaaS ERP is preferred when the data flowing into the enterprise comes from many places and the AI and analytics solve real-time problems. But go with on-premises ERP deployments, when the primary uses of AI and analytics are to work with historical data, which can be more cheaply staged on the ground. Not only is this path less expensive, but it's also more flexible, more adaptable and easier to integrate with existing processes -- and allows your in-house people to grow as they go.
