This article will cover a range of topics from introducing the narrative of artificial intelligence (AI) as an extension of digital transformation to exploring the potential upside for businesses considering engaging in their own AI initiatives. By the end of this piece readers will also be familiar with the idea that no AI journey is the same but that there are several key success factors leaders should be aware of when they are ready to engage in their own endorsement of a technology considered by many to be responsible for instigating a ‘4th industrial revolution’. Given the Darwinian nature of business competition, what could the opportunity cost of losing first-mover advantage in AI be?

1. The story of digital transformation so far

In recent decades the most significant impetus for most companies’ strategic direction has been to engage in programs of digital transformation to remain relevant in response to the online phenomenon. These programs have involved following a set of processes, methodologies, and tools (Ribero, J. 2020) to optimize operational activities and use technology and data to focus on providing customers with improved services. The first wave of these transformations not only increased the breadth of offerings to customers and boosted productivity among employees, but systematically began to incorporate big data into workflows. Ever since, the production and collation of data has been a valuable commodity at the service of companies’ business strategy opening new channels for customer interaction, user analytics and avenues of growth. For those looking to prepare their business for the future and gain an early competitive advantage however, there is a new epoch on the horizon in the form of artificial intelligence, which is getting ever closer.

The impetus for anticipating this shift toward incorporating AI into digital transformation strategies cannot be understated and consultancies have been assembling resources and skillsets for what they see as an inevitable part of their clients’ future business models. It is widely acknowledged that as data and AI become increasingly important, any given companies’ success will likely depend on how they manage these capabilities. Indeed, experts Gerbert, P., Mohr, J-H. and Spira, M. (2019) have even come up with their own metric for how well prepared their clients for this shift using their amusingly named ‘Big Mac Index for the digital age’. Using a self-admittedly crude measuring technique they calculate the share of data scientists in a company’s workforce to gauge its digital and AI maturity. This is followed by a somewhat less crude regression analysis which indicates to a high level of confidence that a one-point increase in the number of data scientists correlates to a two-point increase in EBIT margin and therefore uplift in business performance. Coincidence? Most likely not.

Many companies who embarked on significant digital programs in recent decades are ramping up AI activities in combination with any digital strategies already underway.

Many companies who embarked on significant digital programs in recent decades are ramping up AI activities in combination with any digital strategies already underway. Although successes have thus far been relatively modest it is close to an inevitability that this technology is set to transform work. It will be the companies adopting AI now and whom have aggressive implementation plans who will be best placed to reap benefits likely to surpass those of forward-thinking early embracers of data-driven analytics. These innovators who made such step changes in using AI to analyse customer data leveraged this newfound understanding to increase sales revenue and marketable screen time by recommending more relevant products and personalizing users’ media content feeds. With AI set to shake the foundations of entire industries and stakes rising, those at the helm will increasingly find themselves asking the question, are we doing enough?

2. What value will be derived for AI empowered organizations?

It has been estimated that AI will add $13 trillion to the global economy by 2030 (Fountaine, T., McCarthy, B. and Saleh, T. 2019), this is no insignificant sum and until recently the potential for developing new solutions and driving growth by combining both human and artificial intelligence has received relatively little in the way of attention (Lichtenthaler, U. 2018). Narratives have typically focused on AI capability as being a competitive rather than complementary force to human skill and have neglected to explore the multitude of high potential use cases for well synthesized and applied forms of the technology. The value of AI to businesses can take many forms depending on the integration but most orient around increasing productive efficacy and capitalising on data for strategic purposes coupled with associated positive externalities around workforce job satisfaction. As data is organised in preparation to endorse cognitive technology projects there is an impetus to redesign outdated workflows to accommodate new divisions of labour between human and AI tasks. This will ensure humans and machines augment each other’s respective strengths and compensate for weaknesses. It is often in business units that have been on the receiving end of these early organizational activities including the reimagining of processes and workflows where the first returns of value are generated.

Cognitive insight is another area where prediction-based value is produced by AI as algorithms are trained to detect patterns within vast siloed datasets and provide analytical substance to their findings. Unlike previous analytic-focused work, pre-trained models will be able improve over time through either supervised or un-supervised learning to identify errors or assign data labels more accurately. This will have a dramatic impact on data curation which has typically been labour and cost intensive and allow cross-comparisons between information appearing in slightly different formats across databases (Davenport, T.H. and Ronanki, R. 2018). To realise true value from AI the question that needs to be asked is ‘how and where does the technology reduce costs?’ As with most technological advances the task for which AI provides a solution is that of providing analytical prediction at eventually comparative low cost, taking data which hadn’t previously been organized and applying AI to generate information and insights which were formerly hidden. Although a long way off yet due to the unique nature of needing to customize each AI application to each individual use case it is evident that one day the idea of artificial intelligence-as-a-service allowing companies to purchase problem-solving power from highly trained AI programs to get an edge over competitors may not be as unrealistic as it currently seems.

It is worth bearing in mind that depending on format, AI functionality will perform tasks and not entire jobs, most cognitive tasks currently derivable from data-trained machine learning models will augment human activity by performing a narrow task within a wider job such as guiding robots to collect items in warehouses for humans to then package and post. This is not to underestimate the paradigmatic impact this task assistance will have on job KPIs including output, review and cycle times, the availability of insights and employee wellbeing as well as resource churn. In the meantime, as businesses focus on new ways to cultivate the amount of data at their disposal, the applicability of AI to a broader variety of tasks will continue to grow with it (Agrawal, A., Gans, J.S. and Goldfarb, A. 2017).

3. Why hasn’t the future arrived yet?

There are several reasons as to why the fabled omniscience of artificial intelligence hasn’t yet come to fruition, the primary of which is the misinformed impression that AI can be ‘turned on’ or bought ‘off the shelf’. In fact, the application of AI is more akin to a variable slider (Benjamin, A. 2017) with both a gradual and phased tuning up of functionality through incremental improvement. As the slider moves, businesses will need to adapt, revaluating the implication of change and determining how best to approach futureproofing activities. One of the biggest mistakes any leader can make is to regard AI as a ‘plug and play technology with immediate returns’ (Davenport, T.H. and Ronanki, R. 2018). The reality is that the process is neither easy nor fast, as the price to pay for the pace of innovation is patience.

One of the biggest mistakes any leader can make is to regard AI as a ‘plug and play technology with immediate returns

In essential form, mismanaged expectations boil down to the existence of disconnect between needs and wants. To assess the use cases wherein cognitive applications would create genuine value and contribute to business success key questions must be answered, for example, how would addressing this problem fit into our overall strategy? How challenging will it be to implement the potential AI solution we require on a technical and operational level? Will the outcome outweigh the considerable cost of inputs and resources required to staff such a project? These answers must then be carefully examined to establish which approach will be most appropriate and compliment said organization’s competitive advantage and wider objectives.

For this to happen a clear understanding of which AI technologies perform different types of tasks as well as the limitations and strengths of each must be analysed. For instance, companies in certain industries may be limited by the amount of deep learning they are able to engage with due to the highly regulated nature of their governing bodies and lack of transparency over how these models are created (Davenport, T.H. and Ronanki, R. 2018). On the other hand, rule-based modules can follow pre-programmed orders but are incapable of learning and improving in the same way their machine learning counterparts are. Companies equipped with a sound understanding of the different facets of AI at hand will be better positioned to determine the best fit for their requirements and therefore will start on the right track when it comes to engaging with the technology and service vendors aligned to their ambition.

Pursuing overly aspirational initiatives which are unduly complex to implement or execute have in the past led to costs spiralling out of control. This is not to say that anything other than low hanging fruit should be condemned but that decision makers should be aware of practicable reality and avoid the allure of AI science fiction in its various forms. The importance of having a clear change management plan for whatever foray into AI is being undertaken will become apparent upon realising that the capability of key employees and willingness to learn will need to be leveraged, a task made much easier when tangible understanding and belief in the project is present.

One of the more significant misconceptions from a knowledge perspective is underestimating the collaborative power of an AI implementation. Partnering and establishing dynamic ways of working with the right specialist contractors and external service providers is a critical success factor. For these parties to work effectively together they should be supported by a coordinated team of internal resources with clearly defined roles located in both a centralized IT or strategy team and in independent business units across the organization. To facilitate this, a collective business-wide endorsement of the common goal is usually necessary since it is unlikely that the budget from a specific function will be enough to support any individual AI initiative, let alone a portfolio of initiatives with varying time horizons (Davenport, T.H. and Ronanki, R. 2018).

4. How can businesses prepare for a future with AI?

Level setting expectations and having a specific approach and objectives are among the determining preparation activities a business ought to partake in ahead of any AI initiative. Fundamentally, a clear sense of purpose and mission for what is hoped to be achieved benefits from the application of design thinking principles. These include understanding customer or user needs, considering multiple avenues of experimentation or alternatives, and working in an iterative or agile manner. It is known that companies tend to perform better when they take an incremental approach to developing and implementing AI within a wider transformative vision (Davenport, T.H. and Ronanki, R. 2018), whilst focusing on empowering human capabilities instead of replacing them. The most valuable synergies are often found at the collaborative intersection between technology experts and owners of businesses process in scope for automation. Since AI technologies frequently support individual tasks rather than entire jobs thought must be given to plans for potential scaling up. This activity most often involves integrating with existing systems and processes and has been reported to be the greatest challenge faced by such initiatives.

From a skilling and resource perspective AI represents what will eventually be a shift in the capabilities of employees most valued by their employers. Judgement-based skills will be front and centre of corporate desirability as in more and more circumstances the role of managers will be to determine how best to apply AI, asking questions such as where do opportunities for prediction lie? What information should we gather to facilitate different analytics? What method of learning (deep, machine or natural language processing) should be applied for predictions to improve? Executives will increasingly be concerned by how to train staff from a focus on prediction to judgement related activities whilst developing management processes that complement the new division of labour between judgement-oriented colleagues and prediction-oriented AI tools (Agrawal, A., Gans, J.S. and Goldfarb, A. 2017). Successfully seeing through the transition and onboarding of AI software will require a dedicated change management function focusing on feedback-led comms, training, and other change agent activities. Contrary to popular Hollywood depictions of AI dystopia, organizations will continue to demand people capable of responsible decision making and who can demonstrate emotional intelligence to engage with customers, use creativity to diagnose issues and identify fixes, and find new opportunity areas.

A test-and-learn mentality can help to reframe mistakes as sources of discoveries

It is important to appropriately adapt ways of working where relevant to cater for an AI implementation, the progression of which will invariably experience bumps in the road. A test-and-learn mentality can help to reframe mistakes as sources of discoveries, reducing the fear of failure and encouraging developers and data scientists to ‘fail better’ as well as being more open and communicative about shared learnings. Performing early user acceptance testing and incorporating this into upcoming iterations and versions of the product will prevent minor issues from becoming costly problems. For this, proof-of-concept pilots are best suited to serving testing initiatives with high potential business value as they allow firms to test multiple variations of AI technology simultaneously (Davenport, T.H. and Ronanki, R. 2018). One pitfall for leaders is to avoid is applying too much pressure on senior executives who can be susceptible to agenda-driven vendors coercing them to ‘doing something AI related’ which may or may not actually be relevant to business objectives. In most cases professional scepticism should be applied to anything bearing the label ‘quick win’ and priorities should be given to endeavours which are realistic about the attributable AI value and timescale in which breakeven impacts can be achieved.

Explaining the ‘why’ behind AI is a change management activity that must be commenced early in the planned roadmap. Leaders can setup projects for success by conveying a compelling narrative with clear employee or customer benefits, paying attention to education and adoption, and considering the company’s operational complexity and AI maturity when determining the pace at which change is expected (Fountaine, T., McCarthy, B. and Saleh, T. 2019). Incentives for change need not always be provided but should at the very least be used to guide the production of communication and training materials for change resistant staff. Demonstrating how AI will positively impact roles is often easier when the reality of it is framed as an empowering technology rather than an undermining one such as helping direct reviewers to errors in promotional materials to avoid regulatory bodies imposing heavy misconduct fines. Unlike other technologies in the past where businesses have been able to afford to wait for proven results and falling costs the unique nature of AI development and implementation to specific business use cases has brought back the significance of first mover advantage and we are already seeing corporate strategies containing synergies of both digital and AI ambitions.

5. Key takeaways for your AI revolution

Here I will discard some elements of the formality in the sections above and rather lean on first-hand experience gleaned from being part of an AI product design, development, and delivery team. The takeaways are in no way context specific and some of them will inevitably be more transferable than others.

Top 5 (if you remember nothing else, remember these):

1. It’s all about user centricity and addressing pain points – these should be mappable against the majority of key decisions and proposed functionality.
2. ‘Off the shelf’ doesn’t exist - progress ultimately depends on the quality of AI learning and collaboration between highly skilled developers as part of the tech team, the business/implementation team and end users.
3. Test, test, test – you can never run too many pilots or too much user acceptance testing with the right audience, iterative feedback is gold dust.
4. Set realistic expectations and initiate change management activities early - communication is key, as well as conveying a clear narrative and providing incentives for change.
5. Partnership and collaboration are fundamental to any successful business transformation and AI implementation - between client, vendors, and third-party service providers.

Diary of an AI Implementation:

• A clear objective and purpose must be maintained throughout, the project can’t afford to be pulled in different directions by overexcitement or vocal executives competing for alternative use cases.
• Each workstream should have equal power and voice in project discussions between business, technology, and client teams.
• For AI learning data is king and algorithmic models will need a constant flow of it so map out what is needed and share a request roadmap with relevant points of contact in advance.
• Anticipate two steps ahead, there will be third party vendors involved so identify risks and areas of functionality where you have less control so you can mitigate or find alternative solutions.
• Hold your course until proven otherwise – an AI project shouldn’t change direction off the back of one or two user interactions, sometimes it takes longer to share perspective and see the bigger picture.
• If something is going wrong or not working it must be called out early, testing and learning is the epitome of an iterative process.
• Early workshops are critical for both establishing the current state and determining business rules so that testing can focus on issues and training AI models.
• Don’t underestimate the process benefits associated with preparing for AI - these include current state assessments, the identification of inconsistencies, data standardisation, the realisation of opportunities across the business and updating ways of working.
• Having senior stakeholders who truly believe in the ‘AI mission’ being undertaken, especially from IT, is critical for when things get tough.
• Don’t underestimate integration efforts into existing systems, this can involve multiple parties with misaligned incentives and is unlikely to be plain sailing. Introduce change slowly to build up functionality and accuracy over time, users tend to more concerned about false positives (such as inaccurate performance) than an AI being fully comprehensive in the case of error-flagging review tools.
• For every feature, functionality and capability consider how the business and users will benefit or act upon it.
• From a methodology perspective pursue agile over waterfall, with AI almost every workstream is interlinked and interdependent on one another in some way so nobody should work in isolation.
• At the end of the day any implementation is about people and culture, this boils down to the success of the change management workstream – reskilling employees so that they are prepared and ready to work alongside AI is a critical enabler of change management.
• Elements of most AI solutions will be transferable for different use cases whether that be in the form of employee upskilling, individual trained models or the learnings taken from their development so don’t lose sight of what is being achieved along the way.