Spokesperson: Hemant Kapadia, Chief Financial Officer, Anaplan
AI use in finance has more than doubled in two years, with more than three quarters of
organisations now using it in financial planning, reporting and commercial analysis. For
many finance teams, the current quarter close may be the first where AI is doing real work.
I’ve sat through enough closes to know that’s a bigger shift than it sounds. AI’s increased
role deserves a higher level of scrutiny to match its level of responsibility. AI tools and
agents are becoming more adept at drafting variance explanations, summarising
reconciliations, and flagging anomalies. But handing this workload over to AI, especially
autonomous AI, introduces risk that most organisations haven’t built controls for yet.
An AI-generated variance explanation, for example, can read as clean, complete and
certain, without providing evidence that the numbers underpinning it are correct, auditable
and reproduceable. Finance teams need to know the figures are grounded in current, correct data, and can be trusted. The challenge is therefore not just deciding whether an AI output is useful, but whether finance teams can properly audit and stand behind it.
Nowhere is that truer than in financial consolidation. Throughout my career, I’ve helped close the books across a dozen entities spanning different currencies and charts of accounts. A consolidated variance could look perfectly clean at the group level and still be hiding an error at the entity level, an intercompany elimination that didn’t net to zero, or a currency translation booked in the wrong period. In a close, that means tracing an answer back to the underlying entity and assumption, not treating a convincing explanation as evidence itself.
The importance of deterministic vs probabilistic AI
Finance teams need to be clear about whether they’re dealing with AI that generates or AI
that calculates.
It can be easy to forget that LLMs are simply prediction engines selecting the statistically
most likely sequence of words, given their training and context. Yet what makes them
extraordinary at synthesis and explanation also makes them unsuited to anchoring financial
planning and reporting.
If finance teams rely on probabilistic AI for financial reporting, it’s only a matter of time before an internal audit reveals that a projection used to support a decision was wrong.
Even subtle errors can lead to incorrect decisions and put an organisation’s compliance standing at risk,especially once results are rolled up across multiple entities.
Deterministic AI works on the opposite principle. Rather than generating answers or
numbers aggregated from public information and training data, it calculates from your
organisation’s specific internal data. This materially changes a finance team’s relationship
with AI tools in three significant ways:
- It’s verifiable: A deterministic model returns a single, traceable answer to a query,
grounded in your data. It also cuts the risk of an invented figure or something
irrelevant imported from outside.
- It’s reproducible: If you ask the same question twice you get the same answer,
which is the baseline any close or audit requires, and one that probabilistic systems
can’t achieve.
- It handles precise, high-stakes questions: It can answer something as specific as
the financial impact on a long-range plan if revenue exceeds target by 8%, or how a
currency shift changes a consolidated result, using your actual entity-level data, not
an approximation.
Deterministic doesn’t automatically mean correct. The calculation can be completely
reproducible and still produce the wrong conclusion if the underlying data is incomplete,
outdated or based on the wrong assumptions. Finance teams therefore need to apply the
same discipline to the inputs as they do to the outputs: knowing where the data came from,how it has been transformed and which assumptions are being used.
What the teams getting the most from AI actually do
In my experience, the finance leaders getting the most value from AI during close are the
ones who are the most deliberate about where it’s employed. They use language models for what they’re best at, such as interpreting results and writing the narrative that makes an output make sense to a non-technical audience, but approach the arithmetic and
calculations deterministically. No CFO or analyst is going to issue a set of financials that is
95% right or close their books on calculations they can’t trust completely.
Probabilistic and deterministic AI in tandem produce better answers than either alone. One
brings speed and accessibility, the other consistency, traceability and confidence. Once the
deterministic system has established the numbers, the LLM can help teams interrogate and
translate the results. This separation creates a clearer division of labour: AI can handle the
repetitive work of finding, calculating and explaining information, while finance professionals remain responsible for LLM-assisted judgement, exceptions and decisions.
That distinction matters most during close, when speed is valuable but an answer still has to withstand scrutiny after the fact. Automate the parts of the process where the outcome can be verified,and leave the rest with the people accountable for it.
The principle is the same upstream of the close. A period can only be closed as quickly as
the operational data feeding it allows, and the numbers that move the business are driven by headcount, pipeline, capacity and volumes rather than by the ledger. Holding those drivers to the same standard of verification is what connects a faster close to better decisions in the business. That’s the part most finance teams have yet to work through.