AI & Business Impact | It's the outcome, Stupid!
Part II: For business impact, just 'AI adoption' isn’t the holy grail. Tangible, meaningful, and responsible business outcomes are.
In the first part of this two-piece post, I emphasized how the impacts of AI should be viewed and dealt with in a broader sense to achieve meaningful and productive social outcomes. In this part, let’s discuss how the same lesson applies to business transformations using AI.
Let’s start by distinguishing between potential and value of AI. The hype on business transformation, valuation of companies, and the upcoming revolution is hinged on perceived potential. Now, the first thing that there is a broad consensus on is that AI has significant potential — evidenced by the successes of the past, as a result of which AI has permeated many aspects of our day-to-day lives, along with the ongoing novel developments. Even though the developments in the GenAI space are highly skewed and reliant on the questionable premise of scaling unfortunately, rather than robust science, there are benefits from these novel developments that need to be harnessed productively and responsibly. However, the real-world value realization (read actual business impact, scalable products or services for enterprises, net positive contributors on meaningful metrics) esp. for GenAI, is lagging far behind, with most companies still at the experimentation phase more than two years after ChatGPT was released. For many non-native digital companies, this holds true for non-GenAI AI too. Moreover, the picture of realized value in many businesses, when put in perspective against the costs incurred, is even less pretty.
Now, this is not news! We’ve been here before and will continue to revisit these junctures.
In a light vein, you can certainly survive (even thrive in the short-term) in such “uncertain” (or fast-moving, or, choose a euphemism of your liking) times by following a simple, oft-adopted, “playbook”:
Industry learns of the potential (of a new technology, say “Big data”).
Gets super-excited with the fantasized results, generating hype disproportional to the reality.
Kicks-up significant investments without requisite homework or due-diligence in understanding the nuances. The investments are often skewed and ignore business readiness.
Doubles down on these investments when hitting roadblocks, at the cost of reasonable, meaningful (but less sexy) initiatives.
Realizes the lack of readiness or understanding, and eventually (but mind you, only tacitly and internally) accepts the reality that meaningful results can’t be achieved.
Then winds down these highly-visible initiatives quietly, or rebrands them to accommodate a new wave of technology.
Goes back to step 1 with a new “in-fashion advancement”.
Rinse and repeat. :) until this leads to either unsustainable costs or existential questions for business.
But (semi-) jokes aside — in between these cycles, we also witness real progress for certain businesses that take the effort to properly understand the potential, educate themselves, empower the right people with the right resources, understand the entire value-chain, understand complete requirements, understand investments needed along with cost-sensitive ROI, appreciate the required efforts and long-term perspective, have a systematic and disciplined process of sizing the initiatives with the requisite gating decisions, and most importantly own up to the failures and take the right takeaways leading to meaningful course-correction.
And this difference decides the winners and losers in any technology adoption race, with potentially massive, even existential, consequences to the future of the business. There is one key difference between the above two groups, a key determinant of successful business impact. We will get to that shortly.
But first, we should understand that the current phase of AI is no different when it comes to being a participant technology in this ongoing cycles of business transformations. In the latest iteration, we are undergoing the exuberance phase with GenAI for most businesses and are bound to see some winners and many “also-ran’s” in the process. Here’s how the GenAI driven transformation has been shaping up until relatively recently. A direct quote from the article:
According to a recent IDC study (Future Enterprise Resiliency and Spending Survey, Wave 4, IDC, April 2024), companies are conducting an average of 37 GenAI proofs of concept (POCs), with only five advancing to production. Of those, just three are considered successful.
This Hitachi Vantara report, in fact, can serve as Exhibit 1 in the study of what’s missing and why most such initiatives are destined to fail. It details how the majority of companies (97%) believe in the potential of GenAI, but most executives feel that they are not “ready”. Among the stated reasons for this inability to effect transformation with GenAI: lack of infrastructure, lack of data ecosystem readiness, trying to limit reliance on proprietary algorithms (LLMs), data security, and to some extent, cost. Interestingly, not once, do any executives mention that there exist an informed basis and roadmap to undertake the “GenAI transformation” journey in their business. Other than vague promises, there is limited, if any, effort devoted to understanding to what end GenAI will be adopted. Rarely, do we see empirical-evidence-based strategies that can provide a principled cost-benefit analysis to guide the efforts.
And none of this is new (we seem remarkably consistent and adamant about not taking the right lessons from repeated cycles of failure). Here is a 2017 study on trying to understand why most “Big Data” initiatives were failing. Here’s another 2015 Gartner estimate on the failure rate in excess of 60% (arguably a conservative estimate based on analysts) for big data projects. An October 2020 Forbes article revisited the big data transformation predictions and found that:
2021 will represent a decade since Big Data came into prominence, and based on industry findings, the promise remains largely unfulfilled.
We had initial exuberance and a hard realization subsequently that most of the digital transformation initiatives were failing too as noted in this 2018 Harvard Business Review article.
Most of these analyses were focused on the so-called non-native digital companies undertaking transformation initiatives, or claiming to do so at least — driven by big data, digital transformation, analytics, and now GenAI (which is the most visible cycle, but it can be said about broad AI too, of course). In fact, while we ride the enthusiasm around GenAI, there are studies pointing to over 80% failure rate on “data science” projects (remember data science?). And when it comes to GenAI, we also have some of the major big tech players in the waiting queue to demonstrate tangible value. What gives?
Despite the “branding” across various cycles, the common theme in the current context is the quest to realize business impact through data-driven technologies. And despite the technology “wave” in question (big data, data mining, machine learning, data science, AI, or GenAI), after every failure analysis, the most commonly identified core issues that are recommended to be fixed have been more-or-less the same — data issues, infrastructure issues, skills, user studies, organizational maturity, and some hand-wavy issues around change management, resistance from legacy teams/processes, analytical readiness, etc. There have been other “waves” (e.g., IoT/Industrial IoT) with similar challenges and similar-looking failure analysis findings (customized to intended audiences, e.g., telecom players, hardware manufacturers, and so on).
So, in a nutshell, we have kept running into the same issues, by industries’ own admission, for over a decade and a half now and there is not much demonstrable progress (commensurate with the original expectations) — at least, that’s what study after study suggests.
Now, one can always rebut this in many ways:
General arguments take the form of one or more of these: since the technology is advancing, the past challenges aren’t the same as the new ones; there are so many examples of developments; industries have made so much progress over the past years even if we haven’t hit the tangible outcomes; there have been macroeconomic, geopolitical or other unexpected developments that disrupted the progress.
More specific arguments can focus on specific industries. For instance, the cost of the IoT devices ended up being prohibitive for low-margin use-cases (on failings of the telecom industry in IoT space); data availability ended up being much more difficult due to, say, evolving regulatory challenges or non-employment of connectible devices by users; and so on and so forth.
To which I’d say I agree, but only to some extent. It’s not as if most of these challenges just showed up surprisingly and surreptitiously at the end of the transformation journeys. They would have been clear in a simple, principled and informed SWOT analysis, if nothing else. On examples of developments, yes, I concede that there have been isolated examples of success but if we have to provide isolated examples (and even those are anecdotal at times) to make a point, we can agree that these exceptions prove the norm.
Almost 7 years ago, I was fretting about these same issues! If I were to rewrite that article, I don’t think I’d need to change much (ironically, very much like templates after templates that have conveniently been re-used around transformation analyses pointed to above).
So, what’s going on? We keep getting distracted by questions around how to achieve something without really understanding and articulating what is it that needs to be achieved. A classic organizational challenge. And after a period of initiating new transformation initiatives with much fanfare, the objectives (the what) take a backseat even if they were ever established. The means become the goals. Just as I pointed out in the context of society in the previous post, the same holds true for AI and business impact:
Businesses can’t get excited just about AI advancements. They need to assess and work towards mutual readiness — of AI and business for each other— and anchor these efforts in a well-defined set of meaningful business objectives.
For businesses, AI adoption isn’t the holy grail. Tangible, meaningful, and responsible business outcomes are.
Just looking at new AI advancements (which are often incremental in most cases) and assuming that they’d somehow automatically translate into business outcomes only if you could “adopt” them is naïve at best. There is a steep journey from any technological, let alone scientific, development to its sensible productization. And this journey is both business-specific and non-trivial. The success of businesses depends on mutual readiness — of technology and businesses — to co-work. That is, the technology needs to be mature enough for the set of use cases that the industry targets, and vice versa — the businesses need to build readiness for the adoption of technology (e.g., organizational, and operational readiness, migration plans, mitigation planning, and so on).
And that brings us back to the key determinant of success — between businesses that can achieve meaningful AI impact and the ones that keep getting stuck repeatedly.
Understanding “What” is the determinant of success
Understanding, and candidly and meticulously working towards, meaningful business outcomes is the ultimate test to know if a transformation can be successful. This is where top leadership across the spectrum has apparently struggled, and struggled spectacularly! Instead of getting into deep nitty-gritties of the technology (which is most often counter-productive, not to mention frustrating to the operational teams), leadership needs to devote efforts in articulating the what — whether you call it a business vision or transformation objectives. What outcomes are you striving for? How should your business leverage AI and to what end? Are you targeting AI adoption for new products, new services, expanded markets, recurring revenue channels? Would AI be leveraged to allow the business to move up the value chain? Would it help to build product- or operational-differentiators? Every business will need to assess the alignment with their business model and build context-sensitive strategies (see, for instance, this story on Britannica’s journey). There is no uniform template for achieving business impact with AI. You’ll certainly need to partner with the AI teams and even learn about the technology along with its capabilities and limitations. But that isn’t enough. Once there is an evidence-based (not hype-based) consensus on realistic objectives, they need to be translated into effective strategy and efficient roadmaps. Leadership also needs to empower the operational teams. One of the reasons why most AI efforts are stuck in the experimentation phase is the lack of clearly defined business-relevant objectives. Consequently, neither an effective strategy nor any informed analysis and readiness can be developed. Needless to say, the entire value and development chain — from opportunity delineation, to investments required, to efforts needed, to personnel and skillset decisions, to cost-adjusted ROI estimation — suffers. The AI experiments then become window-dressing destined to be sunset once their PR utility is diminished. These are lost opportunities.
Instead of getting caught up in the hype, and associated “pressures”, it is much more productive to invest efforts towards defining what it is that you’d like to achieve with AI. Is AI a tool for automating tasks that are routine (or even value-added)? Is it something that you’d like to exploit for business transformation? Or is it something in between these two extremes?
If you consider AI just as an enabler for automation, that can be a valid practical goal. Even if it is not a transformative goal, it can give a tactical advantage. Automation of feasible tasks would be a bare minimum that will need to happen to make sure you don’t outright lose out to the competition. And even then, you need to develop a robust understanding of what, the objectives, that can drive informed initiatives with meaningful timelines and milestones.
The goals along the spectrum of limited automation to all-out transformation can all be valid given the context of any business but confounding them isn’t helpful. One can argue that automation in the extreme is in fact transformation but that argument runs into its own challenges. For instance, if automation can be achieved at the entire enterprise scale, chances are that the barriers to entry will go significantly down for the business. Consequently, the business may not just have not been transformed, it may have been endangered making the ecosystem ripe for newer, faster, and more effective competition. The disruption of the advertising industry over past two decades is a good case study of this effect - the industry has fundamentally changed with the advent of big tech offerings.
I have called out the importance of an informed business vision as a top priority, along with a comprehensive guide to achieving successful AI transformation in the past. The article is just as valid and relevant in the current context as I see the same challenges manifest across the board repeatedly.
Here are some other key practical, often ignored, lessons on outcomes that businesses would be well advised to pay attention to for once:
The biggest challenge has always been a lack of clearly defined set of envisioned business outcomes, let alone locking them in to anchor AI or data-driven transformation initiatives.
The investments typically precede the vision and strategy. It’s no surprise then that most of the oxygen is taken up in running through the made-up metrics of progress (I’ve witnessed entire departments in companies mandated with building big data infrastructure without any business-relevant metrics; The KPIs of success? In one instance, just being able to generate $1 of revenue - no kidding)
Be careful! If done carelessly, even short-term successes can be counter-productive. There have been examples of “revenue-generating” data businesses that end up hurting the company — by focusing on building a myopic data business at the expense of much larger opportunity and compromising the business’ differentiation. Hence, focusing on the right outcomes is also critical.
Another common challenge across the legacy industry is the abused phrase - “we want doers, not thinkers”. I emphatically disagree. You (should) desperately want thinkers and there aren’t many around. So much “doing” without “thinking” has only resulted in massive sunk costs and about 15 years later we are in no better position than before. Thinking and doing don’t need to, and should not, be exclusive.
There are many additional aspects that need to be included and worked on as businesses look to realize impact and transformation with AI. The main objective of this post was to sensitize the leadership and operational teams to appreciate the necessity of building a robust understanding and plan for the desired outcomes. If not, we will inevitably revisit the retrospectives and try to re-learn the lessons. It’s not just endless cycles of transformations that seem to recur, it’s also those of retrospectives.