I first heard the TCO (Total Cost of Ownership) acronym when I joined Gartner in 1995. Gartner had formalized the term for IT usage in the late 80s. However, its specificity applied much more to PCs and other personal productivity items. The term as applied to enterprise tech turned out to be more elusive. Trying to estimate TCO for enterprise projects and assets which deliver value over 20+ years and which have a wide range of hardware, software, systems integration, application management, upgrades and other cost elements is not for the faint of heart. Yet, without reliable TCO numbers, the ROI calcs are by definition soft.
So, one of the nice nuggets from the Workday innovation Summit a couple of weeks ago was to hear co-founder, Aneel Bhusri take a walk down memory lane and talk about “measurable business value."
“Going back to the early days of cloud, we didn't sell cloud and say, “Hey, this is cool new technology with the new user interface. Buy it.” It was basically you can make your business more agile, better data, and you can do it at half the cost, and we grew 50% during the financial collapse of 2008-2009. Today it's no different. All technologies need to have a really, really strong TCO and ROI model.”
“You have to go to a customer and prove that we're going to save you money. You're going to get all these capabilities, but it's also going to be cost beneficial. If you can't have that second piece, it goes away. I mean, AWS took off because it's, frankly, cheaper than running your own data centers. It's a great platform. There's 100 reasons why it's a better platform, but if it was more expensive than what's out there, it'd be, it'd be a hard one to sell. I mean that. I think that almost every piece of technology comes down to a payback, TCO or and when you say about TCO, it can be productivity, but that productivity has to be translated into cost, or you're never going to get those things through the CFO”
I would argue that one of the main reasons legacy ERP, CRM and other vendors have seen only 20 to 25% of their customer bases migrate to their newer architectures is because they do not fully developed TCO and ROI models. Most vendors have “value realization” teams who can help a customer develop nice slideware for board presentations but they often lack rigor, especially about migration and on-going maintenance costs for even 5 years, forget 20 - especially when it comes to validating numbers their partners present. And when it comes to outcomes to anchor ROI around, the metrics are often too generic, not industry or geography specific.
We may face the same dilemma as we try to sell agentic solutions. There are new cost elements – GPUs, hyperscalers, data acquisition costs, data science talent, new testing and validation processes among them that few customers know how to measure with much specificity.
The ROI may in some ways be easier to correlate to outcomes if agents mimic current job roles. If agents allow you to scale up volumes without new human effort, if they allow you to reduce some of the tasks human employees currently perform that productivity is relatively easy to quantify. As I found in research for my book Silicon Collar a few years ago automation tends to target “dull, dirty and dangerous” tasks. You can qualify savings from the reduced human time for the tasks. There is, however, a nuance as I described in the book. Automation transforms roles. Bank tellers did not completely disappear as ATMs targeted their cash handling tasks. Their jobs became more of salesperson and customer service.
In some scenarios, the payback is from increased revenue. More from the book - UPC scanners and related self-checkout did not completely eliminate grocery clerks. CPG companies found UPC codes allowed them to come up with a lot more mass customization and SKU codes and grocery volumes grew significantly. There was positive revenue payback for grocery chains from their investment.
There is however one big constraint when it comes to today’s enterprise software. Our definition of agent is not ambitious as that of Jensen Huang of NVIDIA. He is thinking beyond agentic to physical AI.
Actually, even Jensen is not as ambitious as Barry Roman. Barry as in the CEO of Polestar in our recent fiction thriller, The AI Analyst. Polestar’s “agents” include robots, drones, UAVs, sensors and other “physical AI” which allows them to target firemen, shop floor workers, pilots, doctors and many of the 800 occupations the BLS tracks via its SOC methodology. And it has plenty of worker skills, attributes etc. they use to identify which automation – machine vision, robotic limbs, conveyor belts can be applied to each task.
Today’s agents that enterprise software vendors are showcasing mostly focus on white collar, back office workers. Clearly useful but when it comes to payback and ROI that only delivers to a tiny portion of modern economies. And they will deliver even less to the re-industrialized workforce coming to US and other factories, farms, construction and other sectors as a result of global and vertical re-alignment coming with Trump's "tariff turmoil"
We are still in the “soft” phase of calculating TCO and ROI numbers for AI. We are focused on “trust”, “ethics” and often on short lived, “ephemeral” agents. Aneel is right – we need to become much more quantitative in our enterprise AI pitches. But I would also like us to sign up for a stretch goal for our product development -and become much more ambitious in our definition and reach of “agents”.
