AI & the Enterprise: Quo Vadis?

By Geoffrey Moore

Author – The Infinite Staircase: What the Universe Tells Us About Life, Ethics, and Mortality

Everybody gets that AI is going to change the world, but nobody is really clear as to how, which would be OK except that CIOs everywhere are under pressure to invest now, at a time when things are still forming, norming, and storming.? We need to step back for a moment and survey the landscape so we can prioritize where we should engage first and why.

Let me suggest the following framework as a point of departure:

I submit that most of the value of enterprise IT today is delivered by the two systems highlighted in bold—the systems of record, which include finance, HR, supply chain, purchasing, and the like, and the systems of engagement, which include sales, service, marketing, commerce, and the like.? These are the mission-critical stakeholder systems that define every enterprise’s relationship with its customers, partners, investors, regulators, and employees.? They are quite simply indispensable, and as such, they are the anchor tenants of the Stack, with all the other elements ultimately justifying their existence by being in service to one or the other of the two.?

Both systems rose to prominence in the 1990s as client-server applications running atop the Internet, and in this century, they have been advanced dramatically by two sets of adjacent sets of systems, the systems of infrastructure, which include cloud computing, mobile computing, data management, cybersecurity, and hyperscale outsourcing on demand, and the systems of collaboration, which include video conferencing, file sharing, messaging, threaded discussions, e-signatures, and the like.? These have enabled companies to expand both the external reach of their systems of engagement and the internal productivity of their systems of record.

Now, we are seeing the emergence of a third set of systems, still nascent, consisting of systems of intelligence, which include predictive AI and generative AI, and systems of autonomy, which include agents.? Both are based on advanced statistical software that can learn, and both are driven by machine learning feasting on all the data it can hoover up, including massive troves of log files that were never before examined except forensically.?

The question before us is how will systems of intelligence and systems of autonomy impact the operation of the Stack as we have known it to date.

The first claim of this framework is that the AI-enabled systems will operate in the middle of the Stack, not at the top, and not at the bottom.? That is, they will not displace any of the other systems but rather, will enhance them dramatically by operating behind the scenes.? To get to more specific claims, let’s look at each layer of the stack one at a time.?

Systems of Infrastructure

With respect to our systems of infrastructure, the most astounding impact of AI to date is the realization of just how much compute and storage it can consume.? To generate a competitive Large Language Model (LLM) requires a hyperscale compute footprint that only a handful of companies have the resources to deploy.? When we hear that Microsoft has invested $13 billion in OpenAI, we can rest assured that the bulk of that will come in the form of compute service.? Ditto for Amazon, Meta, and Google.

Everyone else will need to license one or more of these LLMs and then adapt it for use in their own Stack.? These adaptations, in turn, will rely heavily on first-party data from the enterprise’s own systems of record and engagement, supplemented by data from their systems of collaboration, as well as licensed second-party data, and publicly available third-party data.? Extracting all that data, normalizing the metadata, filtering out the information that is protected by data sovereignty regulations, and staging the rest in a data lake for real-time use cases represents the most immediate challenge for CIOs today.

For many real-time applications in the physical world, latency issues will dictate that some processing needs to be done at the edge rather than in the core.? This calls for a new kind of PC server armed with a GPU as well as a CPU, running a real-time operating system, connected under next-generation cyber-security protection.? ?By contrast, digital-only applications for the virtual world, including most use cases for systems of record and systems of engagement, require modest infrastructure changes, if any.? The current generation of co-pilots all run on the current end-user platforms, with the heavy lifting all being done by either in the core.

Systems of Record

Systems of Record, as we have already noted, are the foundation of enterprise operations.? They are intentionally conservative by design.? This ensures their integrity, but at the expense of ease of use and adaptability to circumstance.? AI changes this game dramatically.

Generative AI augments the click-based UI of the core system, which users need to be trained on, with a natural language interface that is much more forgiving and can be learned through trial and error.? Advanced use cases can be created through prompt engineering, allowing a general-purpose LLM to be used in tandem with proprietary enterprise data to ensure privacy, integrity, and relevance.? These prompts can be reused and are likely to become an important reservoir of trade secrets.

Predictive AI is an even bigger game-changer.? We have been doing this long before the current AI wave but via software that is preprogrammed and does not learn.? Machine learning allows for continuous discovery of next-best actions, be they for predictive maintenance, fraud detection, energy optimization, demand forecasting, or product sourcing.? It is like having a six-sigma black belt on duty 24/7.

Agents are a bit more dicey, particularly for regulated applications or ones that pose liability issues.? Here a human-in-the-loop co-pilot model is likely to prevail for a long time to come, even after it has been shown conclusively that agents can do the job as well as, or even better than, humans.? Think X-ray diagnosis or self-driving cars.

Systems of Engagement

We are skipping up to the top because systems of record and systems of engagement have much in common.? That said, they are less conservative because they must continuously adapt to unpredictable workflows based on prospect, customer, transaction type, market, and use-case.

Generative AI has a much bigger role to play in these market-facing applications because, in addition to all the UI benefits mentioned earlier, it can actually substitute for human beings in Level One interactions, including creating and running email marketing campaigns, fielding customer service requests through both email and chat, supplying field engineering professionals with on-demand technical support information, alerting sales professionals to next-best actions, and the like.?

Predictive AI, on the other hand, is more of a stretch because human factors are less predictable than system behaviors.? Nonetheless, data-driven decision-making trumps intuition in the long run, and advanced statistical software that learns outperforms even the best humans eventually, as our friends at DeepMind taught us with respect to the game of Go.? Sales forecasting and market campaign attribution are two areas in particular where there is low-hanging fruit to pick.? It just takes more patience and, given the reputational risks involved in market-facing interactions, a more prudent approach to rolling anything out at scale.

Agents fall somewhere in between in that, for highly routine interactions, they can shine and actually provide better customer service than humans, as we all learned many years ago from ATM machines who trumped in-bank tellers from day one.? One thing to guard against here is to measure the success of such programs by cost savings instead of friction reduction, the latter creating a much bigger payoff in customer loyalty, active use, and churn reduction.

Finally, co-pilots for all customer-facing functions are really a no-brainer.? They do not lose context, they do not lose track of time, and they do not get bored with routine.? In addition, they are great at prompting and taking a first cut at any natural language task.? The critical issue here is focus.? The more tuned the co-pilot is to your specific business, the more impactful its contributions will be.

Systems of Collaboration

First of all, we need to appreciate the fact that systems of collaboration are a differentiating source of information for any enterprise.? That is, more than any other source, they represent the quality and texture of all your relationships in flight, not just their progress toward achieving your targeted outcomes.? They embody who you are and what you care about.? By including such data in data lakes that feed your AI models, not only will you improve your systems of collaboration themselves but also make better recommendations to your systems of record and engagement.

Specific to improving collaboration itself, AI excels at summarizations that cut through the clutter of long communication threads, timely reminders that keep workflows from bogging down, and sentiment analyses that can detect concerns and improve tone.? The more your enterprise depends on nuanced knowledge work, the higher such improvements need to be on your priority list.

Systems of Intelligence

As you can see from the foregoing, in my view, we should not think of systems of intelligence as a separate layer in the Stack but rather as a technology infusion that changes its very nature.? That is, our core business systems are poised to become more intelligent.? Indeed, a fruitful metric might be something like a “system IQ test,” which CIOs could use both to assess their current state and to target their future state, all to be done by overlaying and integrating a layer of advanced statistical software that learns.? As the BASF slogan used to say, “We don’t make the product, we make it better.”

What we should not endorse is the notion that Systems of Intelligence, or indeed any form of AI, is going to “take over.”? Yes, they could have unintended consequences, but no they could not have Game of Thrones ambitions.

With one possible exception.

Systems of Autonomy

Systems of autonomy are the natural extension of systems of intelligence wherever the task in question can be done better by a machine than a person.? We use them to place digital advertising, to get astronauts to space stations, to orchestrate the Internet, to run the GPS applications that help us navigate the world, to detect military threats, to prevent spam.? They are an indispensable part of the digital transformation we are still in the midst of, and their role will only increase going forward.

The question is when you take the human out of the loop, who or what is in charge?? This is not an intractable problem, but it is also not one that will get solved theoretically.? This one will require lived experience, a history of successes and blunders, some pleasant surprises, and some very unpleasant unintended consequences.? We should not be shocked.? Life never stops evolving, and natural selection never stops operating.? We just need to step up.

That’s what I think.? What do you think?

Follow Geoff on LinkedIn | Geoffrey Moore Main Mailing List | Infinite Staircase Mailing List

_____________________________________________________________________________________________________

Geoffrey Moore | Infinite Staircase Site | Geoffrey Moore Twitter | Infinite Staircase Twitter | Facebook | YouTube