Eddie Elizondo | Last updated September 2024 | No Hype Analytics blog homepage
Can AI and advanced analytics be commodified? I’m defining commodified as “plug and play” software with minimum customization required - the AI algorithm works with small amounts of configuration. Businesses face this question through AI solutions sold by software and consulting firms, with claims that it can be deployed quickly. With the proliferation of GenAI-related software, we’ll likely see this asked more in late 2024 through 2025.
If you or your business uses AI / advanced analytics software, you may already have had a suboptimal experience with an AI deployment and familiar with those sequence of events. It starts with impressive sales demos and claims of the value its cutting-edge AI features will have on your business. People arrive to implement. Your team takes new trainings. You start seeing the solution. A feeling of anxiety sets in: it seems like your business is force-fit into software: the interface is clunky, and its workflow and terminology don’t align with your business. The “AI intelligence” of the solution seems suspect. You raise these issues and you either hear of limitations or are faced with extended timelines to customize. You start looking for workarounds. You many even start missing your old ways – it was imperfect, but it worked – while you ask yourself, did the AI do anything?
This is a worst-case, extreme scenario for most. Still, perhaps you have faced some of these issues. What went wrong?
There are prerequisites that could be at fault, like issues with the underlying business process failures, data availability, or data quality. All valid reasons, but we also don’t talk enough about whether the AI is working well.
Almost every AI developer, from startup to large software or consulting provider, faces a push to commodify their AI or analytics products. It’s inevitable: a commodified product is easier to scale and monetize. From the perspective of a customer or client, does this work?
Often, it does not. AI and advanced analytics are sensitive technologies: they require significant tailoring, specialization, and fine-tuning to extract the additional value they can provide. The best question to ask yourself when faced with a provider selling a plug-and-play solution is: are there hundreds of identical deployments solving my exact (not similar, exact) problem? If so, a commodified solution likely works. If not, proceed with caution.
Here are a few examples:
-
Large businesses with many locations (retail stores, factories, warehouses, etc.) solving a common problem can often make any AI software provider work, because any significant customization that may be required will still yield a good ROI when rolled out to each outlet.
-
Businesses of any size with a unique, specific problem will struggle to find commodified AI software that works. They should look to customized / bespoke AI firms and/or their internal data science teams.
-
Small or medium businesses wanting to use a commodified solution must ensure the problem they are trying to solve is the same as hundreds of other deployments before looking to commodified AI. Even then, they must ensure their underlying businesses processes work and parameters in the AI software are well-customized. For example, for even commodified supply chain planning software to work, businesses must ensure their supply chain has been segmented correctly, has ideal decoupling points, etc., then adjust safety stock parameters in the software.
My best advice, in addition to asking the question are there hundreds of identical deployments solving my exact problem, is to be honest to yourself about your needs, add a healthy dose of skepticism to commodified AI solutions, and turn to a trusted AI advisor for an opinion, if needed. (Sidenote: I am happy to give an opinion – feel free to reach out.)
There’s one important distinction for data science teams. Above, I’ve been talking about entire solutions purchased, not solutions built by data science teams. Custom analytics solutions can almost always use commodified components of software. In fact, data scientists should be leveraging commodified components as much as possible.
These components are LP/MIP optimization solvers like Gurobi/CPEX/COIN-OR, machine learning packages like XGBoost, reporting interfaces like PowerBI, logging tools, etc. Alone, these are computation or software engines and don’t deliver standalone value. The data scientist should know the art and science of utilizing them effectively, and using them well is a part of a DS toolkit. Those components are almost always better developed and maintained, and they run faster.
Do you have questions, thoughts, feedback, comments? Please get in touch - I would love to hear from you: eddie@betteroptima.com