The next time a finance team reports that AI has shortened its month-end close, the
board should ask what happened to the days it saved. Did management spot a cash
shortfall earlier? Did it challenge an investment that was underperforming? Did it settle a
tax position before it became expensive?
At Oblique we work to a simple principle. AI prepares the work. Qualified professionals
review it, decide and sign off. The value ofAI in finance depends on how well that review is
designed, and on what the business does with the time it releases.
The surveys suggest few functions have reached that point. In a June 2025 Gartner
survey of 183 CFOs, 7% reported high or very high impact from AI. Among GCC
organisations surveyed by Deloitte, 10% had an enterprise-wide AI strategy and
governance. Adoption is moving faster than the operating design around it.
Review has a cost
Reviewing AI output takes time. A reviewer needs the source evidence, the expertise to
challenge the output, the time to investigate and the authority to reject it. Corrections,
model and prompt changes, and data clean-up take more. The capacity released is what
remains after all of that. The financial return must also cover software and integration
costs. A business case needs both measures, alongside the gain in quality. A business
case that counts only the preparation hours saved overstates the gain.
The right level of review depends on the risk. In our practice we set a level of autonomy for
each task by its materiality and the cost of an error.
- Assist. AI helps and a person does the work, as in tax research and board commentary.
- Prepare and review. AI prepares and a person reviews every item, as in material
journals and tax computations. - Exception-based. AI processes and a person reviews the exceptions and a sample, as
in bank reconciliations and invoice coding. - Bounded autonomy. AI acts within tested limits and a person monitors the results, as in
payment reminders and daily cash reports.
Payment release, tax filings and sign-off of the accounts stay with accountable
professionals at every level.
Oversight takes real effort to build and to keep working. A reviewer who sees AI get it
right a hundred times can stop checking the hundred-and-first. In the pilot we rotated
reviewers, seeded known errors into test samples and tracked each reviewer’s catch rate.
A falling override rate needs explaining. It can mean the AI has improved. It can also mean
reviewers have stopped challenging plausible answers.
What it changed for one client
We piloted this on one client’s VAT return pack. The pack covers the return, an exception
report, commentary on movements from the previous return, and the emails and memos
that go with it. Turnaround fell from about nine calendar days to five. Team hours on the
pack fell by around 35%. Much of that saving came from drafting: the commentary, emails
and memos arrived as drafts to check and correct.
The client now sees its VAT position four days earlier. Earlier information gives a CFO more
time to act: to plan cash around the VAT payment, query an unusual movement or resolve
an exception before filing.
The rules already ask for evidence
FTA Decision No. 13 of 2026 took effect on 1 October. Before deducting input VAT, UAE
businesses must verify their suppliers and the supplies they receive. They must also keep
a documented verification policy that names the people responsible. From 1 January 2027,
UAE businesses with revenue ofAED 50 million or more will exchange B2B invoices as
structured data.
Each rule asks the same thing of finance. Show the work. Name who checked it. Keep the
evidence. A well-designed review process supports compliance with those rules, using
the same evidence and controls that produce the management numbers. Shared
evidence and controls can lower the cost of compliance.
What to demand
Released time needs a job. Give the staff hours released a defined purpose: investigating
margin erosion, improving collections, testing a funding scenario or challenging an
investment case.
The same standard applies to advisers. As preparation takes less labour, clients should
expect more senior attention, clearer reasoning and earlier intervention. CFOs can test
this when they appoint an adviser:
- Who will exercise judgement on our work?
- What evidence will support the conclusions?
- How will quality be tested, and how much review time is priced in?
- What capability stays with our team after the engagement?
Finance leaders should now demand three things from their teams and their advisers:
accountable judgement, traceable evidence and demonstrable business value. AI makes
all three easier to deliver. The design of the human review decides whether they arrive.
About the author
Wahaj Siddiqui is Founder and Managing Director of Oblique Consult, a Dubaiheadquartered tax, accounting and advisory firm serving groups, family offices and founder-led businesses across the UAE and Saudi Arabia. He leads the firm’s tax and regulatory advisory. He began his career in KPMG’s tax practice and held finance roles at Etihad Airways, Emirates and Al Hilal Bank before founding Oblique in 2018. He hosts the Beyond Decimals podcast. Oblique’s white paper, Humans in the Loop: The AI-Native Finance Function, is available at obliqueconsult.com.
