All-in On AI by Davenport and Mittal

AI is the buzzword today and many companies are getting onto the AI bandwagon, either developing it or adopting it.

And the authors start the book by opining this will accelerate because the “AI fuelled” companies they cover are doing well and “… have effective business models, make good decisions, have close relationships with customers, offer desirable products and services, and charge profitable prices. They have become learning machines, and their people are accelerated by AI. They are typically able to do these things because they have more and better data than other companies that is analyzed and acted upon by AI …” – wait a minute.

From what the authors are saying these companies sound like they already knew what they were doing, and AI was just a booster.

And I think I’m right, it aligns with a common understanding that technology, including AI, often magnifies existing competencies and strategies rather than being a standalone solution for fundamental business problems.

So, any firm touting the use of AI to lure customers does not get a free pass on the fundamentals – its foundations, such as its product-market fit, channel-market fit, price ladder and so on, must be robust. In fact, if it uses Predictive AI as a central feature of its product then there is additional cause to worry.

Going all-in on AI requires strong leadership commitment and may not be everyone’s cup of tea, but those who do (and have their house in order) will likely see disproportionate gains. Though in the future using AI for business may become table stakes.

The degree of AI adoption by companies places them on a spectrum with ones aggressively applying AI called “transformers” (nothing to do with the Transformer model of Generative AI), and the most aggressive adopters being labelled “AI fuelled” by the authors, “… some have hundreds of deployed systems and business outcomes too numerous to count.“.

The authors further drive home the point that AI is not a silver bullet by emphasising that executives desirous of applying AI to their departments need to do their homework before making any major changes or investments, they need not become experts but do learn the foundational, Pareto concepts of the kind of AI they are interesting in benefitting from – and if I may add, also need to look at real world evidence of benefit from the AI instead of relying on the seller’s claim of performance. Naturally, employees down the line will also need to be unskilled in the effective use of AI.

And even if your company has a strong foundation to build on, its leaders knowledgeable about AI and its pitfalls, it is likely to face issues in deployment – many companies excitedly get on the AI bandwagon but do not stop to consider the systemic changes that will need to be made to actually integrate the AI – deployment is often an afterthought and the team building the AI may consider it “someone else’s job“.

“Many [companies] never get to the only step that can add economic value—deploying a model into production.“.

A company that is all-in on AI is basically one that embeds “… AI into products and services, and to conduct many tasks and even entire business processes in a more automated and intelligent fashion.” – okay, very good, no surprises there.

Per the book, if a company aims to become AI powered it can do so by giving the AI either of the following two things (or both):

1. Data: Give data to train AI so that it can learn patterns.
2. Rules: Give rules to the AI for it to know how to behave.

With the above two basic building blocks, AI will be able to do four tasks for the company:

1. Identify: The AI can identify hidden patterns, help in classification, recognise text or images (and the intent behind them) and so on.
2. Predict: The AI can make predictions aid in decision making.
3. Enforce: The AI can enforce rules and policies previously manually implemented.
4. Generate: The AI can create text, images, video and audio to assist in business areas such as training, customer support, sales and so on.

Having the AI do these four things the “AI Fuelled” company gains the following six benefits:

1. Faster execution: Tasks that previously took humans much longer can be accomplished much more quickly, leading to faster project completion, quicker responses to market changes, and more agile decision-making.
2. Lower costs: Primarily by automating human labour required for various tasks.

3. Deeper understanding: This benefit stems from AI’s ability to work with vast amounts of Data. When AI is tasked to identify subtle patterns, correlations, anomalies, or customer segments that might be invisible to human analysts
4. Improved engagement: Creating better, more personalised, and effective interactions with stakeholders, especially customers but also employees.

5. Better innovation: Using AI to identify unmet customer needs or emerging market trends from diverse data sources.
6. Increased trust: AI helps identify and resolve issues like security breaches or fraud customer problems more rapidly and effectively.

Per the authors, the key raw material that will make or break a company’s benefit from AI is data (esp. proprietary data) – this tells me that the foundational use of AI (for key business advantage at least) relies on machine learning.

Without data you cannot “do AI”.

“Companies that are serious about AI must be serious about data—collecting it, integrating it, storing it, and making it broadly accessible.“.

So, leaders looking for benefit from AI must ensure that their data stacks are solid and also think about new uses of existing data as well as additional sources of data.

Before any AI magic can happen, there’s a fundamental prerequisite the authors likely stress: the data itself must be machine-readable. This sounds basic, but it’s a crucial first step.

For AI systems, especially machine learning models, to learn patterns and make predictions, the information needs to be in a format they can computationally process.

This means data should ideally be structured and clean, either captured in a machine-readable format right at the source or meticulously prepared and transformed to be so. Without this, even the most advanced AI is working with garbled signals.

Once we have usable data, the next step for an ‘AI Fuelled’ company, as the authors seem to suggest, is often to create a centralised data lake. I understand this as a vast, accessible repository where all sorts of internal data from across different parts of the company are consolidated.

This breaks down the traditional data silos that often prevent a holistic view of the business. Crucially, these companies also emphasise the importance of integrating and analysing external data.

This allows the AI to learn from broader trends, market conditions, or other relevant information outside the company’s own walls, leading to richer insights and more robust predictions.

This is where it gets really interesting, and it shows the depth of commitment ‘AI Fuelled’ companies have to data.

The authors highlight that these companies will consciously change their business models or design entirely new ones specifically to gain access to greater quality and quantity of data.

The example of Airbus’ Skywise platform is relevant – by creating a platform for airlines, Airbus not only provides a service but also gains access to a wealth of operational data, which in turn fuels its AI capabilities and service improvements.

Furthermore, going beyond their own boundaries, these companies often spearhead the creation of a ‘data universe’ or ecosystem, as exemplified by Shell’s Open Subsurface Data Universe (OSDU).

I find this fascinating because it involves consolidating data not just within the company, but then building a collaborative environment where other stakeholders – partners, vendors, researchers, even competitors – can contribute to and benefit from a shared data lake. This requires standardisation and a new level of industry collaboration, but the payoff is an incredibly rich data pool that no single company could create alone.

Finally, even with vast amounts of well-organised, machine-readable data, the authors emphasise that latency is critical.

Low latency ensures that decisions are based on the freshest possible information, enabling real-time responsiveness, operational agility, and a significant competitive edge.

What I take from this is that “AI Fuelled” companies strive to minimise the delay at every step: the time between raw data being generated or recorded and it reaching the AI for processing, and then, crucially, the time between the AI deriving an insight or prediction and a decision being made or an action being taken based on it.

In today’s fast-paced world, the value of an insight can diminish rapidly.

Companies are making it simpler for anyone to train ML models using AutoML.

“Automated Machine Learning,” often shortened to AutoML, refers to the process of automating the end-to-end tasks of applying machine learning to real-world problems.

Essentially, the goal of AutoML is to make it easier and faster to build and deploy machine learning models, even for people who aren’t expert data scientists. It aims to automate many of the time-consuming and often repetitive steps involved in a typical machine learning workflow.

This isn’t just about adopting new software; it’s a strategic move towards democratising AI capabilities.

This trend suggests that the ability to leverage machine learning for predictive or identifier tasks is indeed on its way to becoming a basic skillset for the knowledge worker of the future, much like MS Excel proficiency did in the past.

Consequently, as the authors point out with examples like Kroger, proactive training and upskilling of the entire workforce to understand and utilise these tools effectively becomes a critical organisational commitment. It’s about preparing everyone for this new future of work.

The idea that many “All-In” on AI companies are adopting Standard Operating Procedures (SOPs) for how ML models are built – from solution engineering through model development to deployment, as seen at Kroger’s subsidiary 84.51° – is a mark of maturing AI adoption.

When the authors say, “Models are framed, developed, and deployed in the same way that a well-managed manufacturing organisation might create physical products.” it signifies a shift towards discipline, quality control, reproducibility, and scalability in AI initiatives, moving beyond ad-hoc experimentation.

The creation of comprehensive internal platforms, like DBS Bank’s ADA platform, is another vital enabler.

Consider buying the book for a deeper dive into this and several other real world examples of applying AI in the organisation.

Such platforms streamline and govern the entire AI and analytics lifecycle: data ingestion, ensuring security, managing storage, implementing governance, enabling visualisation, and handling the management of AI/analytics models themselves.

A key objective here, is to enable as much self-service as possible, empowering various teams to create and maintain their own AI models efficiently and within established guidelines.

The computational demands of sophisticated AI, especially advanced forms of Predictive AI and large-scale model training, are significant. Many companies are migrating AI systems to hybrid clouds and this reflects the need for flexible, scalable, and often faster processing capabilities.

Furthermore, access to High-Performance Computing (HPC), whether on-premise or through cloud providers, is often a necessity, not a luxury, for organisations pushing the boundaries with AI.

Perhaps one of the most crucial points in such a rapidly evolving field is that the technology environment for AI is anything but static. As the authors say, “No organization should expect that they can establish a technology environment for AI and let it ride for a decade.“

Continuous monitoring of external technological offerings and a constant reassessment of their match (or mismatch) to internal strategic needs is absolutely critical. Agility and a willingness to adapt the tech stack are key to staying “AI Fuelled” in the long term.

Going beyond data and tech. stacks, the authors make a crucial point: while technical expertise is essential for building and deploying AI models, the direction, purpose, and ultimate business value of those AI initiatives need to be driven by strategic conversations at higher levels.

Questions like “How can AI improve our business?” “What can we do with AI to create new offerings to help us grow?” and “How can we make money with AI?” are fundamental strategy questions, not just technical ones.

And as the authors point out, these aren’t easy questions.

This means that senior managers, strategy departments, and even strategy consultants must be deeply involved, bringing their understanding of business situations and strategic possibilities into dialogue with what AI can offer.

No single person today can be expected to have an all encompassing view of AI, whether it’s solopreneurs like me or massive companies – all of us are living through a time where an increasing knowledge of AI is important, foundationally at least, how each kind of AI works, what it can/cannot do and so on.

The modern knowledge worker will need a blend of both deep business insight and a solid understanding of AI’s transformative potential.

And even so we need to maintain intellectual humility due to the dynamic nature of this technology.

Truly becoming “AI-Fuelled” is as much about fostering this strategic dialogue and organisational learning as it is about mastering the algorithms and data.

Objective 1: Build something new

This can be either a new (or improved) product, a new customer experience, or even a new business model in a new market.

The authors take us through several case studies of several companies: Loblaw and their PC Health initiative, Toyota and its “Chauffeur”, “Guardian” and “Team-mate” driver assistance systems, Morgan Stanley and its next-best-action (NBA) system.

But the really serious companies go beyond merely augmenting existing business processes through AI and instead build entire platforms, even business ecosystems with AI at the core – such as Ping An and its massive healthcare ecosystem that uses AI at several steps, Airbus and its Skywise data platform, Shell and its Open AI Energy Initiative, and Japan’s SOMPO that has established not one but five ecosystems.

Objective 2: Become more efficient

Although important, the authors position this a lower impact goal versus the previous one of building something new “… companies that use AI primarily to explore and create new forms of business value are 2.7 times more likely to improve their ability to compete with AI than those who use AI primarily to improve existing processes.“.

Here too the authors take us through a few cases such as Kroger and its “Restock Kroger” strategy of whose four components two were dependent on AI, while the “Kroger Precision Marketing” program sought to use AI to deliver more personalised and relevant marketing messages and promotional offers to customers across channels.

Objective 3: Influence customer behaviour

Here too the authors show how companies like FICO with its credit score, Progressive Insurance with its driving score, or Manulife with its health score nudge customers into living with greater financial responsibility, better road manners, and increased mindfulness about their health respectively.

They show how AI, enabled by data tracking technologies such as smart watches or vehicle telematics, has the ability to quickly assess if your actions are in line with desired actions and then help nudge you towards the latter. Such nudging is beneficial for customers in more ways than one – for instance, a vehicle insured under usage based insurance (UBI) is beneficial because it not only makes the driver drive more safely and this behaviour then means that the driver’s insurance premiums are reduced.

If your people see AI as not very helpful, or worse, a threat to their jobs then no matter how strong your AI strategy, it will fail to execute. Like the technology itself, a culture that embraces AI must be engendered top down. The authors give the example of the DBS’ Piyush Gupta (who also happens to be an IIM Ahmedabad alumnus 😀) as a role model for other leaders looking to make their organisations “AI Fuelled”.

A critical lesson from Gupta’s example is that he started experimenting and learning about AI early and did not abandon effort even though his first attempts were failures.

This gave the rest of the company the signal and also the permission that it was okay to fail with AI as long as you were learning something.

As a result “AI Skunkworks” sprouted in the company with different departments testing out this new technology for business gain, even if they were traditionally composed of non-technical personnel such as HR. I expect AI Skunkworks to be the norm instead of the exception in the coming years especially with AutoML.

“Hackathons” and other events also normalised the idea that AI was something that the organisation dabbled in from a learning POV.

No one lost employment because of AI putting to rest fears about AI driven layoffs. While any success or gains achieved due to AI were publicised internally.

Upskilling employees was also seen as a central requirements as the bank sees the role of human operators central to extracting maximum gain from AI.

Funding was also enabled for this departments who wanted to go a step further and hire key AI resources and setup other supporting infrastructure. The initial experiments with AI also did not require financial justification as Gupta understood the presence of a learning curve with AI.

Of course, for all this change to flow from the top, the leader themselves must be well versed with AI, and its promises and perils as well as with IT more generally. “Some of the best AI leaders are technologists at heart.“.

Finally, change management is also critical when the AI is deployed into production – “… identifying stakeholders, gaining clarity about objectives and performance expectations … communicating frequently … demonstrating prototypes … retraining/upskilling workers who will be users of the new system.“

A company does not become “AI Fuelled” all of a sudden – it’s a journey of “… experimentation, development of capability over time, fits and starts, mistakes and setbacks …”.

There are five levels on the “AI capability ladder”:

1. Level 1 – Underachievers: This is the starting level and here companies are experimenting with AI but have deployed nothing into production and thus have zero economic gains from AI.
2. Level 2 – Starters: These companies are only slightly better than the Underachievers, they have plans to deploy AI or may even have a few models in production but the economic gains from them are still zero.

3. Level 3 – Pathseekers: These companies have deployed a few AI systems and are now realising first hand how production deployment and economic gain from AI is a continuously iterative affair. Even if they have economic gain from AI, they are slight at best.
4. Level 4 – Transformers: Like I mentioned earlier, these are the companies that are finally seeing economic gains from AI, they have multiple AI models deployed in production.
5. Level 5 – AI Fuelled: These are the companies that are not only using AI to boost business, but building businesses around a core of AI – examples include Ping An, Shell, Airbus as I mentioned previously. They are becoming learning machines.

The authors take a deep dive into showcasing how Ping An, Scotiabank, Anthem, Progressive Insurance, and Manulife became “AI Fuelled” including leadership vision/buy-in, training employees at all levels, setting up dedicated AI research units, hiring key resources in AI (some even directly reporting to the CEO), building an ecosystem for collecting and synthesising data and more.

Of course, all these companies are not doing AI for AI’s sake – there is a clear intent first to build models that will realistically deploy, and then to have any deployment translate into tangible economic impact. They’re going for incremental but consistent improvement, “… no “big bang” projects, just those that involve continuous improvement …”.

With the foundational components as described above in place, a company can reach the final level on the AI capability ladder – AI Fuelled – by travelling on one of the three paths available to it (the authors actually outline four paths in the book, but I found paths 2 and 3 very similar, so I’m covering them as one).

The authors illustrate these paths by taking the examples of companies like Deloitte, CCC Information Solutions, Capital One, Well and more.

Path 1 – AI + Humans

In this journey the organisation makes a conscious decision to use AI for the purposes of enabling and empowering its human workforce across a broad spectrum of tasks. This is the most encompassing of all AI paths as all parts of the company are expected to learn and use AI.

Significant investment is made on employee upskilling to the point that training centres may be created by the company for this express purpose.

Path 2 – AI + Analytics

On this path the company applied AI, specifically ML, to enable better data analytics, insights and decision making. Key resources such as PhD-level data scientists get hired to build ML analytics, and a key strategic focus is to build an exceptionally strong data backbone.

The sophisticated decision tree enabled by applying AI to the data backbone is used to deliver automated and seamless experiences to customers at every touchpoint.

Path 3 – AI at the core

On this path, AI is not applied to the company as much as a company is applied to AI. The authors illustrate this with the example of Well, a behavioural health startup.

• Know what you can accomplish with AI.
• Start with analytics.
• Reduce technical debt.
• Put data and apps on cloud.

• Marshal some data assets.
• Create an AI governance structure.
• Develop centres of excellence in AI.

• Be prepared to invest.
• Work with an ecosystem.
• Build solutions across the entire organisation.

The major downside to the application of AI in companies, and that is that it might inadvertently lead to bias and lead to certain members of society (and sometimes the company’s own employees) to be unfairly discriminated against. There are other risks too such as data privacy and responsible use, and “automation bias” leading to “AI solutionism” that ends up harming the company or its customers. And if the benefits that a company reaps from AI application are significant there is also the risk of, well essentially, the leadership getting greedy and using AI to make money at the expense of moral or ethical values.

To address any unintended fallouts from these risks, companies that are serious about AI must also be clear around how they will ensure AI is used ethically.

They may choose to create guiding principles, policies, assign responsibility, always requiring a human in the loop (HITL), and even create independent AI ethics councils. The authors showcase Deloitte’s Trustworthy AI Framework as a starting point for companies looking to ensure ethical application of AI.

The framework includes seven elements:

1. Private: User privacy is respected, and data is not used or stored beyond its intended and stated use and duration; users are able to opt-in / out of sharing their data.
2. Transparent & Explainable: Users understand how technology is being leveraged, particularly in making decisions; these decisions are easy to understand, auditable, and open to inspection.
3. Fair & Impartial: The technology is designed and operated inclusively in an aim for equitable application, access, and outcomes.

4. Responsible: The technology is created and operated in a socially responsible manner.
5. Accountable: Policies are in place to determine who is responsible for the decisions made or derived with the use of technology.
6. Robust & Reliable: The technology produces consistent and accurate outputs, withstands errors, and recovers quickly from unforeseen disruptions and misuse.
7. Safe & Secure: The technology is protected from risks that may cause individual and / or collective physical, emotional, environmental, and / or digital harm.

Companies can also seek to join AI Consortia such as the one by the World Economic Forum or “The Partnership on AI”.

The authors also share a case study on how ethical and responsible AI was implemented at Unilever.