Key learnings from Andrew Ng’s AI Transformation Playbook
Artificial intelligence (AI) has become a popular buzzword in recent years. Amid the media hype and high profile debate between Elon Musk and Jack Ma on whether we’re in for an AI utopia or apocalypse, there’s broad consensus that AI will have a profound impact on business and society in the not so distant future.
AI technology is poised to transform every industry, just as electricity did 100 years ago. By 2030, it’s expected to create an estimated $13 trillion of GDP growth. While it has already added tremendous value in leading tech companies such as Google, Amazon, Microsoft and Facebook, experts say much of the additional waves of AI value creation will go beyond the software sector.
“ Deep Learning is a superpower. With it you can make a computer see, synthesise novel art, translate languages, render a medical diagnosis, or build pieces of a car that can drive itself. If that isn’t a superpower, I don’t know what is. “
- Andrew Ng, Founder of deeplearning.ai and Coursera
The trouble with AI adoption
According to Accenture, over 80% of the world’s largest companies know that they must scale AI in the next five years or risk going out of business. Yet most still struggle with large scale AI adoption. Many companies either don’t understand what AI can do for their business (often described as ‘a solution seeking a problem challenge’) or, when they do experiment with AI, can’t easily scale the technology beyond siloed pilot projects.
A recent O’Reilly global survey on the state of AI adoption in 2020 found that 85% of companies are shifting deployment from prototype to production and more than 50% of global organisations identify themselves as ‘mature’ AI adopters. However, almost 25% of respondents cite cultural problems, like lack of institutional/C-level support as the biggest obstacle to broader AI adoption, closely followed by lack of skill sets and talent shortages.
Meanwhile, companies that are already strategically scaling AI, report nearly three times the return from AI investments, compared to companies pursuing siloed proof of concepts.
Plenty of well known companies are already innovating in unprecedented ways, leveraging AI’s “superpowers” to reach, engage, convert and retain customers through smart content curation, voice and image search, analytics, ad targeting, chatbots and hyper personalisation, optimise operations and improve their products and services:
- Netflix — uses machine learning to improve streaming quality and power its content recommendation and personalisation engine.
- LinkedIn — uses AI for its recruiter and jobs recommendations engine.
- Novartis — utilises AI to accelerate drug discovery and healthcare delivery.
- AT&T, Verizon, Comcast as well as just about every other large-scale telco use AI for enhanced customer service as well as network optimisation and predictive maintenance.
- Sephora’s — launched an AI powered chatbot to let customers book makeup appointments, increasing bookings by 11%.
- Unilever — used AI-powered marketing tools to create Ben & Jerry’s new line of cereal flavoured ice cream.
- Nike — uses AI to let customers to design their own shoes in-store and create custom sneakers in under two hours.
- Alibaba — launched a “FashionAI” concept store using intelligent garment tags, smart mirrors and a virtual wardrobe feature in its app to help customers uncover product information, get fashion recommendations and build entire looks in-store or at home.
- Chase Bank — embraced AI tools for its ads when trials showed machine-written copy outperformed humans.
- BMW — introduced an AI-enabled personal assistant into their cars.
Which begs the question — how do some companies successfully launch, scale and generate major ROI from AI? What are they doing differently?
What makes a true AI company
Full disclosure: having been a technology lawyer for many years, and then an e-commerce entrepreneur, I’ve crossed paths with artificial intelligence and machine learning along the way. I’ve always found it fascinating but never had the opportunity to dive deeper. So I recently took “AI For Everyone“ — an online course on Coursera taught by AI guru, Dr. Andrew Ng, a world-renowned AI technology leader, founding lead of Google Brain and a prominent AI educator. The course demystifies basic AI concepts for non-engineers, and skilfully explains a few complex ideas underlying machine learning. It also breaks down the practicalities of the AI product workflow and explains how to spot AI growth opportunities in your company.
Andrew Ng makes a simple yet profound observation: that simply trying to bolt AI technologies onto a typical company, won’t make it a true “AI company”. Just as in the internet era, simply having a website doesn’t make a company a true “internet company” (so, for example, a shopping mall + a website doesn’t equal Amazon).
Today, it’s common knowledge that a true internet company optimises for things that the internet does really well, such as:
- using A/B testing to trial new website features/pages;
- having short iteration times so it can release products and test the market fast;
- making data-driven strategic decisions based on advanced analytics to analyse and improve digital metrics such as customer conversion, satisfaction, engagement and retention; and
- using technology to deliver a better end-to-end customer experience.
Also, a true internet company organises itself with digitally focused roles (product managers, UX designers and software engineers), each with respective responsibilities and workflows that map back to digital goals and KPIs. Similarly, argues Ng, for your company to become great at AI, it’ll need to be organised to do the things that AI lets you do really well:
- Sufficient understanding of AI: There should be general understanding of AI, with appropriate processes in place to systematically identify and select valuable AI projects to work on. It’s important to have the right people with the right skills in place to be able to analyse what AI can or cannot do for your business and identify and execute the highest value projects.
- Strategic direction: In true AI companies, the company’s strategy is broadly aligned to succeed in an AI-powered future. Ng recommends devising a thoughtful AI strategy after (not before) you’ve experienced working with AI: “Once teams start to see the success of the initial AI projects and form a deeper understanding of AI, you will be able to identify the places where AI can create the most value and focus resources on those areas.”
- Allocate financial and human resources to systematically execute on multiple valuable AI projects: true AI companies have the outsourced and/or in-house technology and talent to systematically execute on multiple AI projects that deliver direct value to the business. AI projects involve intensive data collection, model training and iteration, deployment to test users and further iteration and maintenance to keep the model accurate and useful. While AI can give your product or service superpowers, it’s not a magic bullet and requires commitment, investment and plenty of trial and error to validate assumptions and get to the right solutions.
5 key elements of an AI transformation plan
“The important thing is to get the flywheel spinning so that your AI team can gain momentum.” — Andrew Ng
Andrew Ng’s AI Transformation Playbook sets out a simple, practical roadmap for leading your company into the AI era. The key point is that AI technologies aren’t just for companies like Google, Netflix and Amazon. As with the internet, everyone can — and should — make use of AI’s superpowers, no matter what sector you’re in.
The 5 key steps in a successful AI transformation plan according to Andrew Ng’s AI Transformation Playbook are:
1. Execute pilot projects to gain momentum
- Choose the right problem to solve — Your first few AI projects don’t have to be large or deliver huge business value — they should be meaningful enough to help your company gain familiarity with AI and convince others in the company to invest in further AI projects (but they should not be so small that others would consider it trivial).
- Think about the biggest AI opportunity (i.e. not necessarily the biggest problem the business is facing) — consider current AI technologies (e.g. speech recognition, image recognition, natural language processing) and how they might be used to improve your product or service.
- Aim for short term success — Build AI solutions that start showing traction within 6–12 months. Don’t aim for projects that will take years to show results as that would impede your momentum.
- Check feasibility — Get a trusted AI team to do due diligence on a project before kickoff to confirm feasibility (i.e. can AI technology do what you think it can do within your time frame and other constraints?).
- Set measurable goals and clear metrics — have a clearly defined and measurable objective that creates business value (but, again, it doesn’t have to relate to the main revenue generating aspect of your company; it might be easier to tackle a less important feature/service first, gain confidence in the technology and move on to bigger goals).
2. Build an in-house AI team
- Outsourcing to AI experts might help you get faster results for your pilot project BUT you should look to form a centralised, multidisciplinary AI team that can help the whole company accelerate its AI competitive advantage. This AI team could sit under the CTO, CIO, or CDO function or be led by a dedicated CAIO (Chief AI Officer).
- The new AI team will execute an initial sequence of cross-functional projects to support different divisions/business units with AI projects and move on to deliver larger and more valuable AI projects down the line.
- In conjunction with hiring for an AI team, you’ll want to develop company-wide platforms and standards (e.g. unified data warehousing and data governance frameworks) that are useful to multiple divisions/business units and essential for a successful AI strategy.
- An effective AI team would include the following new roles: a Machine Learning Engineer, Data Engineer, Data Scientist, and AI Product Manager — those are different to digital teams from the pre-AI era.
3. Provide broad AI training & education
There are many misconceptions out there about what AI is — many fear mass job losses and don’t fully understand AI’s true potential to improve their ability to perform their jobs, and how it can positively impact the company’s success and bottom line. It’s vital to educate people across all lines of business (not just in the exec or tech teams) about the power, benefits and potential risks of AI.
- Online courses taught by AI experts such as on Coursera, Udacity, Udemy, Data Camp, A Cloud Guru, Amazon and others make it is accessible, easy and cost effective (especially in the current remote working era) to train up large numbers of employees in new skills such as AI, machine learning, data science and programming. This can also be combined with on-site expert training.
- Executives and leaders should understand AI concepts, how to craft an AI strategy and what AI-skills their employees might need in the future.
- Set up a wider “AI PR” initiative — how might you increase your non-technical staff’s general awareness of AI in order to drive a wider understanding and adoption of AI technology across the company.
4. Develop an AI strategy
- Once teams start to see the success of the pilot AI projects and form a deeper understanding of AI, identify where AI can create the most value for the business and focus resources on those areas with the goal of using AI to build your competitive moat.
- A good place to start is to leverage AI to create an advantage specific to your industry sector (rather than trying to compete with Amazon or Google) and to build several AI assets that would be tricky for your competitors to easily copy.
- Align your strategy with the “virtuous circle of AI” - a positive-feedback loop which makes it harder for competitors to break into: a better product leads to more users which lead to more data which results in a better product and so on.
- Establish a forward-looking data strategy, including acquiring data strategically (i.e. think about what data is valuable and how you can monetise it), set up unified data lakes/warehouses and identify what data is and isn’t valuable.
5. Develop internal and external communications
Communicating with the outside world and within your company about your AI strategy, AI product and AI vision is key — think about your messaging and positioning to investors, government relationships, the press, customers/users, partners and talent/recruitment.
“ AI will likely bring significant benefits to your customers, so make sure the appropriate marketing and product roadmap messages are disseminated. “
- AI Transformation Playbook, How to Lead Your Company Into the AI Era by Andrew Ng
It’s clear that to achieve AI success at scale and speed, it’s key for companies to do 3 things well:
1. Intentionally drive AI strategy — successful AI companies have a clear AI strategic plan and instil the right capabilities and mindset within the organisation, regardless of the company’s size. This involves having a clear enterprise vision, accountability, metrics, and governance to break down silos.
2. Tune out “the noise” surrounding data — recognise the importance of business-critical data and identify financial, marketing and customer data as priority domains. The more adept you are at structuring and managing data, the more likely you’ll be to wield a larger, more accurate data set which, in turn, will lead to better AI product performance.
3. Treat AI as a team sport — AI strategy should be championed by the CEO (not just the CIO or IT or data teams) who must make clear that AI is a priority for the organisation. This can be achieved by embedding multi-disciplinary “AI teams” throughout the organisation which have clear sponsorship from the top to ensure alignment with the C-suite vision. These teams are most often headed by the Chief AI, Data or Analytics Officer and comprised of data scientists; machine learning, data and AI engineers; AI product managers, data governance and other specialists. This enables faster cultural and behavioural changes to take place and signals your AI leadership goals both internally and externally.
One thing is certain - companies that wait to adopt AI may never catch up. The AI Transformation Playbook would be a good place to start.
- Be Shrewd About Those 5 Myths on Scaling AI
- HBR: Why Companies That Wait to Adopt AI May Never Catch Up
- AI: Built to Scale, Accenture Research report Nov 2019
- McKinsey: Global AI Survey: AI proves its worth, but few scale impact
- Forbes: Most Executives Fear Their Companies Will Fail If They Don’t Adopt Ai