The slogan “AI first” is often used in keynote speeches, corporate agendas, and business writings. Google now considers itself an AI-first company, and AI-first design is raising its profile in the blogosphere. But you have a business to run. So how do you practically approach AI first? Independent of your business discipline, you’ll find my 7 Cs of AI-first operations helpful in organizing your thinking for AI-powered business and customer success.
1. Customer value
Even if you’re not a technical expert — but especially if you are — you need to be able to articulate how your solution creates value for its user in simple, clear terms. That user can be staff, partners, customers or prospects, and even other digital systems.
Creating value is a broad topic, but the basic question is: What problem or desire are you solving for? Basic types of value like education, empowerment, optimization, connection, expression, and entertainment are all good starting points. AI’s big promise is taking any of these value propositions and making them more predictive, more personalized, and more proactive. This in turn can make experiences more meaningful and powerful — to the point of being almost a magical, invisible layer on reality.
Just make sure you set expectations properly. Exceeding low expectations makes for a better customer experience and satisfaction than failing to reach goals that are set too high.
2. Collecting and managing data
“Garbage in, garbage out” has an outsized effect with AI, where scalable, predictive models depend on the data input. Deep and rich data are desired, even if the initial set exceeds your requirements. Attribution, cleaning, labeling, and storing are also key considerations.
Better data means smarter AI, which in turn creates demand for more and better data. To speed things up, ease data management, and streamline integration, AI is now getting integrated directly into databases like MLDB.ai or Microsoft’s new SQL server.
Quality data is crucial in AI, maybe even more so than in algorithmic design. Thanks to high levels of digital activity, a relaxed attitude to privacy, and thus lots of data readily available to marketers, China may match or surpass the U.S. in AI capability soon. Depending on whether you build in-house capabilities or use AI-as-a-service only, your requirements for your data team vary. This post has great points on the topic.
3. Cognitive resources
Cognitive resources, algorithms, and learning models are now widely available through APIs and cloud services. IBM, which has been in the AI game since the ’50s, leans on Watson and Bluemix, but Amazon, Google, and Microsoft are also heavily invested and have strong offerings.
And it’s not a game for traditional big tech only. Marketing platforms like Salesforce and Adobe are also starting to offer AI as platform, and a vast number of specialized startups are popping up. OpenAI and related nonprofit initiatives also offer resources and training tools.
There’s not going to be one right answer for all. Like a web services stack, your cognitive stack will be guided by your needs, learning models that fit the job, expertise of your scientists and technologists, data sources and formats, and integration needs.
A great example of how AI is becoming ever more accessible is that of a Japanese farmer who, without skills or budget, built a cucumber classification engine with TensorFlow.
One of the core elements that powers our brains, the internet, and AI is massive connectedness. Connecting your smart services across not just your own organization, but also across other organizations, your industry, and related industries for better intel is key to your success.
With both physical and digital products, cross-systems functionality is crucial. If your AI-powered service can’t navigate your brand’s key legacy systems with relative ease, there will be a crack between “smart” and “dumb” services — and thus in your customer experience.
With ubiquitous connectivity, privacy and security have been hot topics. For example, DDOS bots have taken over connected gadgets left to default factory settings. People in networking, threat and log analysis, and related areas are also looking to machine learning for solutions.
The boldest areas in connectivity are Facebook’s and Neuralink’s initiatives for a brain-machine interface. This application, while intriguing, is naturally still some way away.
5. Computing architecture
Running AI applications is not yet a trivial task, and full-scale solutions require massive parallel computing resources. Organizing your computing for AI locally and globally is thus crucial, both in terms of how smoothly your service will run — and how much you’ll pay for it.
GPUs by Nvidia and AMD have been darlings of the AI world — and Wall Street — but new players are entering the fray. Google’s secret weapon against Amazon is its new TPU chip, which is optimized to run and train AI applications, and on offer only through Google’s cloud.
Novel computing form factors are also popping up. Intel’s Movidius is the world’s first deep learning kit in a self-contained USB stick. Microsoft are building AI processing directly into the next iteration of HoloLens for speed, offline functionality, and resulting improved mobility.
Two contrasting trends emerge: More cloud computing services are available for heavy lifting, but devices are also growing more capable, and some computing is moving to the edges.
6. Code of ethics
With fire we had a few thousand years for trial and error before smoke alarms, sprinklers, and fireproof materials. With today’s powerful, pervasive platforms, every digital action can trigger a cascade of unintended interactions, and a lot of damage can happen very quickly.
As AI unleashes immense power into the hands of companies, they will also bear great responsibility. Marketers need to have a code of ethics for safeguarding people’s privacy and safety, their own legal liability, and even the smooth functioning of the economy and society.
And this doesn’t just concern deploying the services, but developing them, too. Applications that have gender biases, or that only identify Caucasian faces because of their training and trainers, need to be addressed for a smart world that’s also fair and equal.
Microsoft’s CEO Satya Nadella is one of the more prominent tech leaders who has laid down some preliminary rules to address the ethical dimension of AI development.
7. Continuous reconfiguration
The AI field is arguably the fastest developing field in technology, and if there’s one certainty, it is that the world of machine learning will look a whole lot different in the next 12 months. Come to think of it, it’ll look a whole lot different within the next three months.
This pace of change requires constant re-engineering of development operations. Some leading minds and companies are calling for a shift to the Design Ops model, which emphasizes relentless focus on customers and customer value, rigorously proven by constant testing and validation.
A lot of tactical metrics will be key inputs to, and even drivers of, AI strategy. The data you gather, the things you make available to customers, and how this all performs at the tactical level become keys in shaping goals and focus on a daily basis.
In an AI-first world, the organization and its people — not just the machines — need to be constantly learning.
Sami Viitamaki is the executive director for digital at Havas, an advertising firm.