Sultan Mahmood is the Global Financial Services leader of artificial intelligence and robotics at PwC. He also leads the development of PwC UK’s AI solutions and advises clients across many industries on identifying and delivering business results enabled by disruptive technologies such as AI and robotics.

Mahmood was a panelist at the recent Fortune CEO Series event, held at Salesforce’s London World Tour. We spent time talking with him about artificial intelligence and the future of work.

Salesforce: Artificial intelligence (AI) is a powerful new technology with the capacity to impact on business processes across the organization. What type of business problem is AI particularly well-suited to solving?

Mahmood: I think the first thing to point out is that “artificial intelligence” can mean a whole range of things. There are three main ways in which AI can be incorporated into a business. Assisted intelligence is all about improving what people and organizations are already doing, and is relatively common today. Think of how Gmail automatically sorts your email into “primary,” “social,” and “promotion” tabs.

Second is augmented intelligence, which is beginning to emerge now. Augmented intelligence will enable companies to do things they couldn’t otherwise do. A good example is Netflix’s content suggestions, in which machine learning enables the company to offer a superior service to anything that was possible before.

Third is autonomous intelligence, which is still being developed, and is all about creating and deploying machines that will act on their own. Self-driving cars and automatic language-translation services are a couple of examples.

While the possible impact of AI is very broad, the technology won’t be applicable to all business problems. There are examples of AI in the creative fields (Flow Machines, an AI system from Sony, released a Beatles-sounding pop song in Japan in 2016), but business processes that require discussion and debate with other humans require a certain level of empathy that AI can’t yet provide.

Take a doctor’s office. AI technology is capable of taking you through a set of questions to analyze what’s wrong with you, and can probably do so more accurately than a doctor can. But that AI won’t necessarily be able to empathize with a patient like a human doctor can. Our recent global survey on AI and health found that while 55% of patients are willing to use AI and robotics as part of their care plan, 47% feel that the human touch is needed in healthcare.

Where AI is already having an enormous impact is taking menial and repetitive tasks off the plates of human workers. Consider medical-image classification, automated insurance-claims processing, or even some elements of customer service delivery. When a customer first calls a company, customer service agents typically have to walk them through a series of questions so the customer can be added to various systems. It’s only then the agent can deal with the specific problem at hand. Introduce an AI that can automate the “onboarding” process, and the experience becomes easier and more efficient for both employee and customer.

Where AI is deployed like this — to take on somewhat menial and repetitive tasks — our experience to date indicates that the existing workforce tends to really like the addition, because the robots take away a lot of the menial activities that they probably didn’t want to do in the first place.

Salesforce: In a world of elevated customer expectations, how can Artificial Intelligence help companies deliver a superior customer experience?

Mahmood: I think one of the most important use cases for AI technology is giving companies the ability to build a deeper and more nuanced view of their customers. That deeper knowledge enables companies to offer a more useful and valuable service to individual customers at scale, and deliver customer experiences that are far smoother and more pleasant than was previously possible.

Consider the current experience for consumers making travel plans. Currently, apps are able to give customers information on flight times, costs, and connections. But if the app provider knows their customer far more intimately, they’re able to look at previous behaviors and declared preferences, and offer personalized advice on the best routes and schedules for every customer. Because the company  understands the customer in far more detail, the solutions they provide can be more relevant and valuable, and offered in service of the customer reaching their goals.

Salesforce: What would you suggest companies looking to incorporate AI into business practices should be doing first?

Mahmood: A crucial first step is for companies to ensure they are aligning AI with the core strategic goals of your company. Companies should avoid allowing a series of different pilots and initiatives be set up around a company in isolation. Instead, take a company-wide look at how AI could disrupt their business as a whole, and begin to define the threats and opportunities it presents. It’s important at this stage that companies consider the societal and ethical implications of any new strategy.

A second first step is for companies to consider what skills and resources they need to build up internally. We’ve recently released our 2017 Digital IQ survey, which identifies a looming skills gap. While 54% of companies are making substantial investments in AI today, only 20% feel they have developed the relevant skills within their business.

Third, given the adoption of AI will be an emotive subject both within a company and in society as a whole, it’s important to consider how you can build trust with all relevant stakeholders. Build a control, monitoring, and compliance framework into the solution from the beginning, rather than once AI systems are in operation.

Fourth, consider the model you’re planning to use for your AI organization. Is a center of excellence a good fit, or is a “develop and steer” model preferable? Once that decision is made, it’s critical to ensure there is clear communication to the rest of the business.

It’s, of course, important that a company has done the relevant groundwork, so the fifth step is to cleanse and label data, and then allocate the resources to train both machines and human employees. It’s equally important to ensure you can demonstrate compliance with relevant regulations — like the EU’s General Data Protection Regulation.

The sixth step is to integrate assurance into your AI operating model. It’s not enough to simply embed AI into operational processes. Companies must have the evaluation function in place to measure outcomes, to identify potential risks, and to spot new opportunities as they develop.

Salesforce: How should business leaders think about educating and training their workforce to ensure it’s fit for purpose in an AI-powered world?

Mahmood: There are a couple of things to consider here. AI is such a disruptive and transformational technology that it’s increasingly becoming a boardroom topic. If your board is not already considering the relevance of AI on your business strategy, it’s important this is addressed as soon as possible

At the next level it’s important to train up those people who’ll be responsible for incorporating AI into business processes and operations. This requires building agile teams combining business process knowledge with industry knowledge, and knowledge of the possibilities AI can provide. While quite a lot of organizations have data scientists, they tend not to have that breadth of experience within the business, so it’s not a matter of leaving all the work to technical specialists.

Equally, those with employees with broad business knowledge tend to lack the depth of expertise in AI. Both types of employee need to be brought together and will need further training to succeed. Delivering it effectively will allow the organization to rapidly implement AI-enabled solutions across their enterprise

From there, companies must also think about training those people directly impacted by those AI solutions. One of the things that we’ve seen quite a lot is that the implementation of AI is freeing up time for the human employees they work directly with. That gives those employees the time to train others, and thus help other departments implement similar AI technologies. Those employees are changing from becoming the people who are running a particular process or undertaking particular activities to people who are now helping the rest of the organization make use of AI technologies.

For more insight on the impact of AI on your business, download the new Salesforce Research report, The AI Revolution: Data-driven insights into the next era of customer relationships below.