Just two years ago, few companies had appointed a Chief AI Officer (CAIO). According to IBM’s latest CEO survey, 76% have now done so, up from 26% last year. Rarely has a senior role been created at such a pace and adopted so widely.
So, it is worth reflecting on what this role encompasses and where it might be heading. Adopting and deriving value from epoch-defining technologies while avoiding major pitfalls is probably a fair working job description of today’s CAIO role. But as these technologies evolve, so must the role.
A crowded C-suite
I have been involved in the building and implementation of high-risk, high-profile technologies for some of the world’s largest banks, energy and commodities trading houses for over 25 years and have seen a fair few senior technical roles develop over that time, starting with the stalwart CTO, joined by the CISO and the CIO, and more recently the CAIO. Then, companies began to ask themselves, “Who is actually responsible for the quality of the data used by our enterprise?” That gave us the CDO, and latterly some organizations have combined the CDO and CAIO roles to create a CDAIO. A lot of new roles for sure, but who is responsible for what?
AI tooling is typically both a consumer of organizational data and a producer of new data outcomes. So, who takes responsibility for the accuracy of those outcomes when they depend so specifically on the prompts used and the context available to the models? And who takes the fall when outcomes are deemed unacceptable, or dangerously wrong?
Frontier models are undeniably powerful but the stakes at C-level are also very high, with boards asking, “Why aren’t you cutting costs by leveraging AI more?” and “Why haven’t you shrunk that five-year roadmap to 18 months”?. That is a lot of pressure on a newly created role, which relies on technologies which – at least in terms of enterprise-wide ROI – remain largely unproven.
If the role runs on a 24-36-month cycle so typical of some CDO role holders, it is long enough to launch initiatives – as evidenced by the abundance of AI pilots, which often surface some early-stage value – but rarely long enough to establish conclusive, production-grade ROI and to be accountable for those results. Furthermore, since AI laggards will inevitably seek to hire the senior staff of AI leaders at a premium, the person holding the CAIO title today may well have moved on before anyone can tell whether these initiatives yielded meaningful business value.
Where the results are real
Yet, within this otherwise bleak landscape, there are some beacons of hope: some teams are getting extraordinary results with the same AI tooling. I can say that with certainty, because my own teams, working closely with our clients, have achieved plenty of them. At one of our global banking clients, two engineers used AI tooling to build reusable automated testing infrastructure in under three months, instead of the more typical two or three squads of six or more people taking over a year. That is measurable ROI and we have many such examples.
That same client now tests large parts of its financial markets software estate, faster and more cheaply, more or less in perpetuity. Contrast that with a large post-trade firm where a thousand engineers have similar tooling but cannot get deployable products out of the door, because they need to devote masses of scarce subject-matter-expert time to reviewing, correcting and often rejecting AI-generated code.
I posit that organisations will need to concentrate on building that domain-rich, contextual understanding into the prompt-conditioning cycle – or better still into the models themselves – thereby embedding the organisation’s innate knowhow into the machine itself, if results are to improve measurably. However, the responsibility for preserving and enhancing that enterprise intelligence is shared across the CDO who manages the data, the CIO who owns the information, the CISO who protects it and the business owners who rely upon it. It should not reside within third party AI tooling and no tooling will deliver a penny of savings or a dollar of innovation unless used correctly and with clear organisational ownership.
Have we seen this cycle before?
The first generation of AI chiefs has not yet moved on, so this is a hint at a possible outcome rather than a firm forecast. But I have watched successive waves of CTOs and CIOs arrive and mandated to drive innovation, only to resort to cost-cutting via outsourcing to ever-cheaper offshore locations, sometimes collecting a large bonus for signing the contract before departing to the next bank seeking to “control costs.”
That role was in effect the economic chief executive of the technology function, and it left behind an ageing estate nobody wants to maintain, patched by ever-cheaper hands with less domain knowledge at every pass. Citing figures I set out in my book, Transform!, maintaining the world’s legacy systems costs around a trillion dollars a year, and failed attempts to escape them cost roughly another trillion, and both are, if anything, conservative estimates.
The AI-enabled version of this future could be much worse, because the bets are larger, the landscape stranger and the pace of change, ever faster. Enterprises are locking in multi-year token commitments without knowing what each unit of spend returns, choosing models whose availability can shift with geopolitics, in a market where state of the art lasts perhaps three months and the full extent of the risks is literally unknowable.
Boards will soon run out of patience with the token-driven economy, the lack of tangible savings and the absence of measurable innovation. If the verdict is “We have invested in AI for three years and productivity has not improved,” heads will roll. You can guess whose head it will be; spoiler alert, it is the one with “AI” in the title.
What to do about it
The AI leaders who last, will combine three things rarely found in one person:
- a research-level understanding of models and data
- hands-on experience of the enterprise’s own data estate and who uses it and how
- a clear-eyed view of the strategic cost of building a company on rented intelligence, by which I mean capability that lives inside someone else’s model, priced by somebody else, and, withdrawable on somebody else’s timetable.
Frontier models are extraordinarily powerful for specific applications, but no organization of scale can run on rented intelligence alone or should expect a third party to deliver context-aware AI to replace the current diet of POC’s.
If the AI bet is to pay off, it must touch every role and function from technology, operations, talent and more. Questions abound; do you still take on graduates? How do you mentor people fluent with machines but poor in context, and grow future talent as the enterprise’s “intelligence” gets baked into each iteration? And when something goes drastically wrong, who takes the hit?
The AI chief may select the models to be used, but they may know very little about how they were trained. In truth the CAIO is at best the conductor of the organisational orchestra, they are not the orchestra itself, but they may be judged as such.
Beneath all of this sits one important principle: treat your organisation’s context, its accumulated knowledge of itself, as an asset you own and grow, not something you rent and hope it comes back as future ROI. Know where it lives, in which systems and whose heads, and make sure that what your people and their tools learn from each project, carries forth into the next, rather than walking out with a contractor or a departing executive, or generating value for rented intelligence platform owners, because that knowhow is the one asset a competitor cannot easily mimic. In my opinion, that stewardship is the work truly worthy of the title Chief (Artificial) Intelligence Officer.
Ian is the founder and CEO of Digiterre, the data-engineering firm that has built the systems inside the world’s largest banks and trading houses, from Deutsche Bank and LSEG to Shell, and Anglo American. His book, Transform!, has been endorsed by Siemens Chairman and former SAP CEO Jim Hagemann Snabe, recently appointed Special Envoy for Industrial Artificial Intelligence by the European Commission, and Sir Ron Dennis, Founder of McLaren Applied and McLaren Automotive.
