Artificial intelligence is an end-to-end technology that can dramatically increase the efficiency of business processes across a wide range of industries and lines of business. Successful digital companies can grow dramatically to become almost out of reach for competitors.
Meanwhile, more companies are still trying to launch their transformation and large-scale implementation of AI at a strategic level. Industry statistics show that 70% of digital transformation initiatives do not end with success and 87% of AI projects do not reach commercial operations.
Ilya Filinykh, CEO of Brainy Solutions discusses common mistakes in the implementation of AI that can slow down digital transformation.
Mistakes when developing a strategy
Strategic mistakes in any initiative cascade to subprojects, and the introduction of new technologies is always characterized by significant risk. A well-developed strategy is critical to successful AI adoption.
- Lack of vision and desynchronization of goals between teams
According to MarketWatch, the capitalization of the Top-5 digital companies (Facebook, Apple, Google, Microsoft, and Amazon) exceeds the capitalization of the other 282 companies from the S&P 500. Moreover, these companies have successfully leveraged network effects of data, creating significant economies of scale and a “moat” against the competition, to borrow from Warren Buffet’s terminology.
Instead of engaging in large-scale transformation and developing their own products and services with meaningful network effects, legacy companies are spending resources on local modernization initiatives and haphazard innovation, while still labeling it as holistic digital transformation.
Some of these projects do not benefit the brand perception, and some even worsen it.
For example, an international telecommunications company wanted to launch an AI chatbot to automate customer support. The local team in each country chose its own vendor and implemented it as a local project. As a result of this approach, the company repeatedly overpaid for training each team, all of them made mistakes due to the lack of knowledge sharing.
Additional resources and time were spent on integrating different vendors within the same platform for storing customer dialogues. Customers traveling to other countries continually complained about an inconsistent user experience and the inability of the “foreign” chatbot to respond to familiar “home” commands.
- Form a shared vision and transformation goals through strategy sessions, idea management platforms, workshops, surveys and interviews
- Lead a distributed but systematic search for AI opportunities under the leadership of a champion from the senior management team responsible and accountable for AI implementation
- Organize regular communication between teams, local offices, corporate innovation lab, general IT department and company leadership
- Premature investment in expensive infrastructure
Premature investment in expensive infrastructure leads to the “freezing” of a large budget, which is difficult to recoup even in the medium term, and creates a mismatch between the infrastructure capabilities and project needs.
One of the largest international retailers spent more than $ 10 million on acquiring a license for and 2.5 years on creating a centralized data warehouse. When the storage project was completed, it turned out that only two teams took advantage of it and built their own layer of data interaction over the storage, not leveraging its own pre-existing similar capabilities.
As a result, the former CIO was fired due ineffective IT budget management and lack of positive business results.
- At the beginning of the journey, the company should focus on real business cases and on, first, attaining business results from “low-hanging fruit”
- Success early on in AI initiatives will allow them to receive further funding, and thus self-finance AI transformation while attracting interest from other executives and employees
Mistakes in strategy implementation
Having a good strategy doesn’t always guarantee success. Most of the mistakes are made by companies at the stage of implementation, since AI requires qualitatively new principles of work, competencies and technologies.
- Insufficient funding and an abstract approach to budgeting programs for the introduction of AI
In AI projects, a significant cost item is the raw data and its markup, as well as computing resources for training machine learning models and utilizing these trained models to process new data. Moreover, unlike typical SaaS, AI is characterized by a more complex structure of variable costs that can negatively affect the economics of a business case upon scaling.
Insufficient funding can hinder scaling up when a data-driven AI solution starts working.
As an example, it took Uber more than $ 10 billion to build its network of drivers and create accurate AI algorithms for data-driven matching of passengers and drivers at scale. It is worth noting that over-funding can be harmful, as the project team will not feel financial constraints and will not be able to fine-tune the AI model for efficient use of computational resources and, hence, attractive unit economics.
The budget should correspond to the task being solved and be allocated gradually based on project metrics and achievement of key milestones, as is generally the case for startup investment rounds. For funding purposes, there are seven phases of an AI project:
- Case Study
- Technical Specification
- A / B Testing
- Launching Commercial Operations
- Lack of support from senior management, without which the transformation can fade or slow down
For example, a project may need to obtain access to data under the control of another department, priority access to server computing resources or upgraded hardware, or additional funding to test hypotheses. In all these situations, a leader with an appropriate level of authority and/or a significant social capital within the company is able to rewrite history and help.
- Highlight a digital leader among the top managers, who will be responsible for the systems implementation of AI (usually CDO, CAO or CAIO)
- Form a working group of executives and project managers that will regularly discuss barriers to transformation and find solutions, as well as serve as a “check and balance” to confirm the decisions of the digital AI leader
- Inconsistency in the choice of technologies and the absence of a centralized AI platform
A lack of clarify in technology choices leads to the inability to meet the growing demand for AI within the company after the initial positive results are obtained. In this case, the initiative to qualitatively change the company through AI lacks a centralized technology roadmap and remains at the level of several projects.
- Beginning with the first projects, the company should think about the issues of reproducibility of experiments for training machine learning models, simplifying the process of putting these models into operation, setting standards for data storage, testing and post-release monitoring of systems, and avoiding code duplication so that the company is able to successfully handle the computational load during “peak times” of massive simultaneous demand for AI-driven business applications
- In practice, for one of our clients, we reduced the budget by 40% (from $ 2.9 million to $ 1.76 million) and the implementation time by 33% (from 12 to 8 months) by allocating the code repeated in several projects to an AI platform hosted on cloud services
- Hiring ineffective employees / Lack of standards
Instead of hiring a large number of ordinary specialists without the understanding of how they will interact and who will be responsible for what, it is better to create effective “special forces” teams and provide them with attractive working conditions and creative freedom.
Due to the rapid development of AI, ordinary specialists often do not have time to adapt to changes without first acquiring the necessary skills.
For example, we faced a situation where a solution developed by us as consultants satisfied all the requirements and goals of a client company, but its employees could not support the solution after its transfer due to the high complexity and a large number of new technologies that were used for development.
As a result, management made a decision to partially abandon its own requirements and certain strategic goals, and we spent several more months in order to simplify the solution and make it “adoptable” by the company’s incumbent IT department.
- Digital teams should be small and consist of highly-qualified specialists who can quickly test hypotheses, adapt to the changing business environment, and create novel solutions for new problems.
- The task of the CHRO (e.g. the recruiting team leader) is to clearly define the roles and requirements for AI specialists, design the composition of typical teams for AI projects of varying complexity, and then conduct an inventory of available skills within the organization
- The missing skills can be gained by narrowly hiring specialists or expanding the talent pool by working remotely with freelancers and/or outsourced teams, which has become the norm during the COVID era
- A linear approach to transformation and non-acceptance of agile methodologies
During the technological revolution that we are now experiencing, new standards and approaches to solving problems are created in the process. Therefore, it is naive to expect that all the prerequisites of the hard-set strategy can be fulfilled.
Speed, flexibility, data and the ability to derive unique insights from data are the main competitive advantages of digital companies.
Let’s look at an example. For a major telecommunications company, our team ran a series of strategic sessions to articulate the goals of an AI transformation. As a result, a roadmap for AI was formed eight months in advance. However, in the course of analyzing customer data a month later, it turned out that the target variable needed for modeling and machine learning was not in the data – it was simply not recorded – and the project could not move on.
Together with the client, we ran additional sessions over six weeks to identify opportunities for AI implementation and planning, and then started implementation. Upon completion of the modified project, it turned out that the new direction was even more profitable – we saved the company a further $10 million.
- Develop the corporate strategy, processes and culture so that you can respond to new information and adjust plans without threatening to interrupt the entire digital transformation
- Instead of one project, the implementation of AI should start immediately with a portfolio of 3-5 projects of varying complexity: several simple projects with a low risk level and high chances of success (2-3 months for each project), several projects of medium complexity (4-6 months for each project) and one complex and highly profitable project (i.e. 6-9 months duration)
- As practice shows, in the process of implementation, some projects may be suspended due lack of necessary data already at the stage of collecting requirements, others will turn out to be unprofitable after the implementation of the prototype and determination of the cost of computing resources
- However, in general, the introduction of AI in the portfolio will show profitability, which will allow one to reinvest funds in scaling the success of remaining pilot projects throughout the company
There is no time to wait. It is necessary to develop solutions that will generate network effects for data as early as possible, and carry out a large-scale introduction of AI and overall digitalization for all business processes.
It is imperative to create an AI platform and be sure to change the corporate culture in order to truly transform and minimize business risk (up to and including the company being made completely irrelevant) due to the threat from digital competitors. The listed errors and recommendations for their elimination will allow you to move faster and increase the chances of survival in this new competitive world.