Building an AI-native finance team

The phrase "AI-native" is overused, but it points at something real. A finance function designed today, without a legacy of twenty years of process and headcount built around manual work, would not look like the finance function most CFOs inherit. The work has changed. The team that performs it has not always changed with it.

Building an AI-native finance team is, for most CFOs, a transition rather than a fresh start. The transition has identifiable components, identifiable failure modes, and a cost that is most often underestimated in the first business case.

What "AI-native" should mean for a finance function

An AI-native finance team is one where the structure, the roles, the hiring, the training, and the measurement are designed around what AI can and cannot do. It is not a team that uses AI tools. Many teams use AI tools without changing structurally. The native version goes further.

Structurally, an AI-native team has the following characteristics:

• The high-volume, rule-based, transactional work is handled by automated systems, with humans monitoring and exception-handling rather than executing.
• The analytical work is performed by analysts who direct AI tools rather than perform the work the tools would perform.
• The judgement and relationship work — business partnering, board interaction, qualitative variance explanation, ownership of decisions — is concentrated in roles that are larger and more demanding than the previous equivalents.
• The governance and review layer is staffed and resourced as a first-class function, not as an afterthought.

The structural changes that are not optional

Several structural changes flow from the work, not from organisational preference.

• Compression of the transactional layer. Roles that consisted primarily of reconciliation, data entry, or routine analysis will reduce. The reduction is not a marginal efficiency. It is the work being performed elsewhere.
• Expansion of the governance layer. The review of AI outputs, the model risk management, the audit support, the documentation discipline — these consume more time than they did pre-AI and require people with a specific skill profile.
• Reorganisation of the analyst function. Analysts no longer specialise in producing outputs. They specialise in directing and challenging the production of outputs. The job description is different.
• Concentration of business-partnering capability. The interface with the business, where qualitative context and relationship matter, is the part AI cannot perform. It is also the part that decides whether finance has influence in the business it serves.

None of these changes is conditional on a particular AI vendor or technology. They are consequences of the work AI now performs.

The hiring profile that the new team requires

The profile that suits an AI-native finance team differs from the traditional profile in identifiable ways.

Technical fluency matters. Not deep engineering — that is a different function — but the literacy to evaluate a model's output, to recognise when the model is operating outside its competence, and to articulate what is needed from the technology team. An analyst who cannot read a citation in a model's output, or who cannot describe what "confidence threshold" means, will struggle in the new function.

Comfort with ambiguity matters. The pre-AI analyst worked with structured tasks that had defined inputs and defined outputs. The post-AI analyst more often works with open-ended questions that require the analyst to define what success looks like before the model can be directed.

Communication matters more, not less. The output of the AI-augmented analyst is, more often than not, a narrative that explains what the underlying data shows. The narrative quality is a constraint on the analyst's effectiveness in a way it was not when the analyst's output was a reconciliation tied to a balance.

Domain depth matters. The analyst who knows how the business actually works — what drives revenue, what creates cost, what the operational rhythms are — can challenge an AI output that the analyst without context will accept. Generic technical skill, by itself, is not sufficient.

The pathway problem and how it must be solved

The traditional finance career pathway started at reconciliation, progressed through variance analysis, and arrived at business partnering. Each step taught what the next step assumed. If the first step is automated, the pathway is broken — not because the next steps are unnecessary, but because the people doing them have not been through the formation the previous steps provided.

The pathway problem is the single most underestimated consequence of AI in finance. It is not solved by hiring laterally from outside; the supply of mid-career analysts is finite and competitive. It is solved by designing a different pathway for the new function.

What that looks like in practice:

• Junior roles defined as analyst-development positions, with deliberate exposure to the underlying work the AI now performs — through inspection of model outputs, through manual sampling, through exception handling — so that the judgement layer has the foundation it needs.
• Rotation through functions, not progression through volume. The analyst who has spent twelve months on AI review, twelve on business partnering, and twelve on regulatory reporting is better equipped than the analyst who has spent thirty-six on the same task.
• Mentorship structures that compensate for the reduction in incidental learning that used to occur through the volume of work.
• Investment in formal training where the volume of practical experience is no longer doing the training informally.

Performance measurement that reflects the new work

The performance measures that suited the previous function — throughput, error rate, time to close — do not capture the work the new function performs. New measures are required.

Output quality matters more than output volume. An analyst who produces a small number of high-quality, business-changing analyses is more valuable than an analyst who produces a large volume of moderate-quality routine outputs.

Review quality matters. Where the analyst's job is to review AI outputs, the measure is the proportion of errors caught, the proportion of false positives escalated, and the speed of legitimate exception clearance.

Business influence matters. The analyst whose work changes a business decision is producing value the financial measures may not capture. Performance frameworks should make this visible.

The transition cost and what it pays for

The transition to an AI-native finance team is not free. It involves redeployment costs, training investment, in some cases severance, in most cases recruitment in a competitive market. The CFO who presents the AI deployment to the board with a productivity benefit and no transition cost is presenting a number that will not survive.

What the transition cost buys is the right team for the work the function now performs. The alternative is the wrong team performing it badly. The cost is paid either way. The first version pays it deliberately; the second version pays it through attrition, audit findings, and missed opportunity. The discipline is to budget the first version.

This piece sits inside the CFO in AI framework. See also where AI cannot replace the CFO and AI in the financial close. Lorna writes from practice at IMPT. The verified page records what is and isn't published here.