Enterprise hits and misses - H-1B Visa news hits - how should enterprises respond? Supply chains need to transform, and the AI data sovereignty era is here

By Jon Reed September 22, 2025
Dyslexia mode
Excerpt:
This week - the H-1B Visa news hits enterprises (and tech vendors), but what should they do next? Is SaaS dead? Not if you examine the pros and cons of LLMs. Oracle gets TikTok, and lessons galore from Workday Rising. As always, your whiffs.

Lead story - Getting supply chains to match the speed and volatility of global business - a progress report

Supply chain management software seems like it's aged more than any software category, despite vendors' concerted modernization efforts. Nothing puts modern software to the test quite like the economic stressors that supply chain professionals take on the chin, from tariff uncertainty to sourcing regulations - with unpleasant downstream impacts for customers. 

So what are we learning? Start with this: real-time visibility matters, and we need industrial machines to 'talk' to us. Then we can plan, manage, and predict their issues. Two diginomica stories delve in. Start with Alyx's From blind spots to digital twins – how Dexory aims to transform warehouse visibility: 

Dexory’s story sits at the crossroads of two industrial shifts. First, the drop in sensor costs has enabled data capture at a speed that would have been unimaginable a decade ago. Second, logistics providers are under mounting pressure to do more with fewer people, less space, and tighter environmental constraints.

The company’s approach is telling because it sidesteps the heavy engineering of conveyors or robotic arms and focuses on intelligence. That lowers barriers for adoption while promising fast payback through error reduction and transparency.

For example? 

In one warehouse, Dexory’s robots uncovered more than $1 million of stock that managers thought was lost. In another, they flagged safety hazards before pallets collapsed. 

The obstacles? Yep, we're back to human change: 

The risk, as with any data-driven solution, may be information overload. Customers suddenly confronted with their own inefficiencies may struggle to prioritise action. Dexory’s success will depend as much on guiding change management as on technical accuracy.

Still: 

The shift from manual checks to autonomous visibility feels less like an incremental step and more like a reset of expectations.

Mark Samuels brings another angle in How Joseph Joseph is designing a next-generation approach to supply chain services: 

Vaughan says transforming supply chain operations involves a significant shift. A crucial element of her organization’s successful transition to XPO Logistics has been a change management process with support from senior stakeholders.

Hmm... I think we've seen that before. Recall my barb about downstream customers? Mark writes: 

APIs and effective data links between Joseph Joseph’s head office in London and XPO’s facility means the partnership has already had an impact on customer service. 

With today's technology in mind, I like our chances - if (and only if) we keep that downstream customer impact front and center of our supply chain pursuits.

Diginomica picks - my top stories on diginomica this week

  • Lessons in professional AI - how to adopt it safely in manufacturing, legal services, and architecture - Instant AI growth and productivity? It's not nearly that simple, cautions Chris, as he looks at the AI adoption lessons in three different sectors.
  • What can enterprise AI learn from 700 million ChatGPT users? - I'm not taking the MIT AI study cheese again, though Derek is tempting me... Instead, I'll call out another crucial theme in Derek's piece: why decision support could be an under-utilized enterprise use case: "The data reveals something unexpected: rather than primarily using AI for task automation, users are getting satisfaction from what the researchers call "decision support" - seeking information and advice to inform better choices. This pattern, whilst happening amongst consumer usage, may offer some clues to how enterprises could deliver more value." Indeed - though I'll snarkily point out that the reason consumers aren't primarily using automation via ChatGPT et al is because right now, it doesn't work. Enterprise AI vendors are getting far better results, via compound AI architectures and workflow/task specific automation.
  • Data sovereignty emerges as universal business risk just as billions flow to US clouds - Derek again, on one of the key AI adoption issues ahead: "Pure Storage released findings this week showing that 100 percent of industry leaders surveyed across nine countries now view data sovereignty risks as forcing organizations to reconsider where their data is located."

Vendor analysis, diginomica style. Here's my three top choices from our vendor coverage:

Workday Rising - wall to wall coverage: I was on the ground in San Francisco for Workday Rising US 2025, but our team coverage didn't stop there. A stellar cadre of virtual diginomica contributors chipped in, culminating in our complete Workday Rising coverage. I can't do that coverage justice here - especially given that Workday issued several major announcements, from AI flex credit pricing to Workday's Data Cloud and Build solution for AI agents. Plus an unprecedented acquisition binge to make sense of, including Sana. And, I'd add - an important shift in tone re: AI and automation versus humans, via "human-centric" AI. Plus: noteworthy customer use cases. Here's a few top picks: 

A few more vendor picks, without the quotables:

Jon's grab bag - Chris issued a fascinating series of reports from Shenzhen, via Chinese multi-national Tencent's annual summit, including this important missive: Tencent Summit – why general Large Language Models and chatbots "no longer meet business needs" for enterprise Artificial Intelligence.

George looks at the implications of Trump's UK visit in Shiny new AI data centers and duller questions – what could possibly go wrong? Cath always seems to issue a new tech-for-good story just when we need it most: Flo Health CTO reveals how data lakes boost women's health app performance. Finally, Stuart couldn't resist taking a bit of air out of Zuckerberg's balloons in The God of Demos is an angry god! There's a whole lot of smiting going on to strike down the unwary, isn't there Zuck? (Zuckerberg blamed his demo snafu on bad wifis, but Stuart isn't having it).

Best of the enterprise web

My top seven

Is SaaS Really Dead? - implications for evaluating LLMs. Dr. Michael Wu is, amongst many other things, my gracious AI debate sparring partner when we make special guest appearances on CRMKonvos. He's one of my go-to thinkers on enterprise AI, and he's written one of a handful of must-read posts on enterprise AI this year. 

Yes, you've probably seen your share of "Is SaaS Dead" posts. What makes this one different? An incredibly useful checklist of what LLM applications are good for, and what they are not. 

This informs decisions on when LLMs can enhance SaaS, replace SaaS, or be better off on the sidelines. I also like Wu's ethical candor. Under his list of applications LLMs can't replace yet: 

  1. Fairness and non-discrimination: LLM agents would not be suitable for any use case that has strict regulatory compliance for fairness (e.g. HR platforms for hiring, recruitment, and interviews; judicial risk assessment, etc.).

This is a notable stance, and I know some vendors would disagree. But: customers should look hard at these riskier areas and think through their design framework carefully. My only addition for now, to Wu's solid lists: it's also worth noting LLMs lack effective physical/spatial world models, and while they can help robots interact with humans, to this point, the (over)-hyped impact of LLMs on robotics is explained by this deficiency. The LLM-robotics use cases are there, but it's not a revenue festival yet.

And, on accuracy: there are different levels of accuracy required for different companies and use cases. For 100 percent accuracy, as Dr. Wu notes, LLMs are not appropriate - other forms of deterministic automation are better (e.g. we would not want an LLM to replace a calculator). 

From my research, enterprises have to work hard to get to 80 percent accuracy with out of the box LLMs. But what Wu hasn't ventured into here is "compound" architectures where LLMs would be informed/constrained by a variety of third party verifier tools, retrieval databases, and smaller models for subtasks and auditing. Designed well for specialized workflows, I've seen a move with these compound architectures to 90 percent, and then in some cases to high 90s. The last mile gets harder, and then the move from 97/99 to 100 is not yet possible with these technologies. Will it be possible? I have strong views on that - it's a topic suitable for our next CRMKonvos debate...

Update: these accuracy levels provide a useful framework for enterprise project design. Under 80 percent? I couldn't recommend anything but AI-enabled brainstorming. Getting closer to 90 percent? Could be useful for a range of internal search/quality assurance/anomaly detection scenarios, meeting summaries etc. Mid-90s? Now you may be in first-tier customer service and "document intelligence" use case territories perhaps, depending on industry etc - but with human-in-loop as a real factor. Higher 90s? Now you may have more autonomous use cases, at least within agreed-upon thresholds - and industrial use cases. And remember: accuracy isn't a constant. Models are known to drift, and shift output based on subtle context changes.

This doesn't rule out LLMs, but it makes the case for what SaaS systems do best as well. That includes bulletproof workflow automations in compliant processes, which can of course be invoked by agents, but that's different than the fantasy (at this point) that AI agents can extract and somehow replace your systems of record. (My Workday AI jobs piece gets into the fusion between SaaS and agents, something I did not attempt here). 


Whiffs

 

 

In other news, the audio-optimized version of my Enterprise Month in Review on services disruption with co-host Brian Sommer and guest Frank Scavo is out:

 

See you next time... If you find an #ensw piece that qualifies for hits and misses - in a good or bad way - let me know in the comments as Clive (almost) always does. Most Enterprise hits and misses articles are selected from my curated @jonerpnewsfeed.

Image credit -

Waiter Suggesting Bottle © Minerva Studiom, Overworked Businessman © Bloomua, Worker made a mistake in a factory © Photographee.eu - all from Adobe Stock.

Disclosure - Oracle, Workday, SAP, Pure Storage and Salesforce are diginomica premier partners as of this writing.

Category:
Disqus Comments Loading...