SOCPA Forum · Riyadh · 14 May 2026
Workshop · 19:00

Accounting and auditing in the age of AI.

A field map. The tools. The benchmarks. What we built. What is real.

Amr ShadyFounder & CEO · Accord Partners
01 / 32
Question

How many of you have used ChatGPT or Claude for finance work this week?

Hands up.

02 / 32
Frame

What you'll get in 45 minutes

Three things, in this order.

01
The capability ladder
From ChatGPT in your browser to autonomous audit agents. Six rungs. What each does. When it wins.
02
The benchmarks
Which models actually win on finance tasks. Real numbers. Real exams. Honest gaps.
03
What we built
Accord OS for Saudi audit. Live demo. What it does. What it cannot.
03 / 32
Where the market is

Global · 2024 → 2027

72% → 99%

Companies using AI in financial reporting today, projected within three years.

Source: KPMG / Accenture, Global AI in Finance Survey, 2024–2025.
04 / 32
Where Saudi is

Saudi Arabia · 2025

75%

Saudi professionals using AI in everyday work — outpacing the global average.

PwC Middle East Workforce Survey, 2025. SOCPA has issued 300+ reforms since 2016. ZATCA FATOORA Phase 2 forces digital compliance on every VAT-registered business.

PwC Middle East · SOCPA · ZATCA.
05 / 32
Section

Part one of three

The ladder.

Six rungs. Each does something the rung below cannot. Each costs more.

06 / 32
The capability ladder

From browser to agent

Six rungs. Pick the right one for the work.

00
Chat in a browser
ChatGPT, Claude, Gemini — open the tab, ask the question.
01
Workspace integrations
Excel Copilot, Sheets Gemini — AI inside the tools you already use.
02
Vertical AI tools
DataSnipper, Trullion, Vic.ai — built for one workflow, deeply.
03
Finance-specific agents
Claude for Financial Services, OpenAI + PwC Agent OS, MindBridge.
04
Big 4 audit platforms
EY Canvas Agentic, KPMG Clara, Deloitte Omnia — internal-only.
05
Accord OS
Saudi audit, end-to-end, partner in the loop. What we built.
07 / 32
Rung 0 of 5

Rung 00 · Chat in a browser

ChatGPT, Claude, Gemini.

Open the tab. Paste the question. Read the answer. Three months ago, they passed all three CFA levels.

Wins at

Quick research. Drafting memos. Explaining a standard. Translating jargon. Brainstorming a risk list.

Loses at

Hallucinations on figures. No memory of your client. No access to your working papers. No audit trail.

Best at finance Q&A: o3, GPT-5.5, Claude Opus 4.7.
08 / 32
Rung 1 of 5

Rung 01 · Workspace integrations

Microsoft Copilot for Finance. Google Sheets Gemini.

AI inside the tool you already work in. Reads your spreadsheet. Writes formulas. Reconciles transactions.

Wins at

Variance analysis in Excel. Cash-flow narrative. Account reconciliation. Reading your data without uploading it elsewhere.

Loses at

Audit-grade accountability. Locked into one ecosystem. Saudi-specific knowledge near zero. ~$30/user/month on top of everything else.

Microsoft Copilot for Finance — GA October 2025. Connects to Dynamics 365 and SAP.
09 / 32
Rung 2 of 5

Rung 02 · Vertical AI tools

DataSnipper. Trullion. Vic.ai. MindBridge.

One workflow, done deeply. DataSnipper: $1B valuation, 175 countries, $1.4B saved for clients in 2025.

DataSnipper
Excel-native audit automation. Cross-references documents, ties disclosures. Disclosure + Excel Agents launched Oct 2025 with Microsoft.
$1B valuation · 175 countries
MindBridge AI
100% transaction analysis (not samples). Risk-scores every journal entry. KPMG strategic alliance since 2023.
Production · Big-4 embedded
Trullion
Revenue recognition (ASC 606 / IFRS 15) and lease accounting (ASC 842) automation. Extracts contracts to journal entries.
$34M raised · narrow but deep
Vic.ai · AppZen
Autonomous AP and expense audit. AppZen claims 87% autonomous processing. Vic.ai end-to-end invoice automation.
Production · client-side tools
Self-reported metrics. Independent audit pending.
10 / 32
Rung 3 of 5

Rung 03 · Finance-specific agents

Claude for Financial Services. OpenAI + PwC. Thomson Reuters CoCounsel.

Purpose-built for finance. Trained on filings. Connected to authoritative content (FASB, IFRS, IRS, ZATCA when ready).

Anthropic — Claude for Financial Services
Pitchbooks, regulatory filings, financial models, valuations. Moody's integrated. PwC Agent OS uses Claude.
Vertical launch · late 2025
OpenAI + PwC CFO Partnership
AI-native finance function. OpenAI's own CFO is "customer zero." Forecasting, controls, reporting via agents.
Announced May 2026
Thomson Reuters CoCounsel
AICPA, FASB, GASB, IFRS content baked in. 1M+ users by Feb 2026. Tax + audit + accounting workflows.
Production · 1M users
Wolters Kluwer CCH Axcess Expert AI
Embedded in CCH Axcess cloud. Audit Suite version GA Oct 2025. Tax research and audit documentation.
GA October 2025
Anthropic, OpenAI, Thomson Reuters, Wolters Kluwer — public announcements 2025–2026.
11 / 32
Rung 4 of 5

Rung 04 · Big 4 internal platforms

EY Canvas. KPMG Clara. Deloitte Omnia. PwC Agent OS.

Internal to Big 4 audits only. Massive scale. Massive investment. Not for sale to anyone outside their own audit engagements.

$1B
PwC commitment to OpenAI. KPMG's five-year AI program is similar scale. EY rolled agentic AI to 160,000 engagements in April 2026.
1.4T
Journal entry lines per year processed by EY Canvas globally — agents now embedded in that pipeline.
10,000+
EY MENA audit engagements in scope of agentic rollout — 14 MENA countries including Saudi Arabia.
EY April 2026 · KPMG Microsoft Copilot deployment · PwC + OpenAI announcements.
12 / 32
The KSA signal

November 2025

EY signed a partnership with HUMAIN.

HUMAIN is PIF's AI company, launched May 2025. EY MENA's AI business solutions are being embedded into HUMAIN ONE for Saudi government and enterprise clients.

What this signals
The Big 4 + sovereign AI infrastructure partnership is now formal in Saudi Arabia. The Kingdom is not waiting. The forcing function (ZATCA + Vision 2030 + HUMAIN) is in place.
The question for Saudi mid-tier firms: what's your answer when a Tier-1 client asks "are you using AI?"
EY MENA / HUMAIN press release · November 2025.
13 / 32
Section

Part two of three

The benchmarks.

Hard numbers. Real exams. Which models actually win.

14 / 32
The headline number

Chartered Financial Analyst exam · Dec 2025

97.6%

Gemini 3.0 Pro on CFA Level I. GPT-5 close behind at 96.1%. Six frontier models passed all three CFA levels.

Human CFA Level I pass rate: 43%. Source: Columbia / RPI / UNC peer-reviewed study, December 2025.

arxiv:2512.08270, December 2025.
15 / 32
Finance benchmark leaderboard

May 2026 · Public benchmarks

Different tasks. Different winners.

TaskWinnerScoreRunner-upScoreHuman
FinanceBench (10-K Q&A)OpenAI o3~90%GPT-5~88%—
Finance Agent (analyst tasks)Claude Opus 4.764.4%Claude Sonnet 4.663.3%—
FinQA (table reasoning)Fin-R1*76.0%GPT-4.1~68%91%
CFA Level IGemini 3.0 Pro97.6%GPT-596.1%43% pass
CFA Level III essayGemini 3.0 Pro92.0%Claude Opus~75%56% pass
DocVQA (document images)Claude Opus 4.793.8%GPT-5.491.1%~95%
Arabic reasoning (MMMLU)Claude Mythos92.7%Gemini 3.1 Pro~91%—

*Fin-R1 = open-source finance reasoning model (Alibaba). All other rows are general-purpose frontier models. Scores from Vals AI, awesomeagents.ai, arxiv:2512.08270, MindStudio (May 2026).

16 / 32
Three patterns from the data

What the benchmarks tell us

Three patterns. Each changes how you should staff.

Frontier models beat humans on closed-form exams. CFA, MMLU, document QA. The certification gap is closed.
Agents still cap at 64% on real analyst work. The Finance Agent benchmark — with tools — peaks at Claude Opus 4.7. Complex modeling remains human.
Arabic finance has no published benchmark. Zero. This is the trust gap for any Saudi audit firm deploying AI on Saudi-client work.
17 / 32
Section

Part three of three

What we built.

Three insights. One demo. Honest about what it cannot do.

18 / 32
Insight 1 of 3

Insight 01

Standardization comes before automation.

Today
Working papers
Step one
Standardize
Step two
AI layer

"The backbone of transformation is standardization. You've got to clean out your closet."

Jessie Kanter · Partner, Citrin Cooperman · Journal of Accountancy, 2026

When we acquire a firm, we are digitizing thirty years of institutional memory. The roll-up is the standardization play.

19 / 32
Standardization, applied

Concrete example

A real Saudi firm had 14 templates for one trial balance.

What we found
Each partner customized the master. Headers in different orders. Account codes in different columns. Some Excel, some paper.
The acquisition is the standardization play. When we bring a firm into Accord, we don't just buy revenue — we digitize 30 years of institutional memory and normalize it into one system. Then AI reads it.
Anonymized. Saudi target firm, 2026 diligence.
20 / 32
Insight 2 of 3

Insight 02

The auditor's job changes. It does not disappear.

Yesterday

Generates working papers. Tests samples by hand. Builds memos line by line.

Today

Designs the AI agent's scope. Reviews its output. Makes the judgment calls AI cannot.

More intellectually demanding. Not less.

Frame: Byron Patrick, Karbon TB Academy.
21 / 32
The honest list

What AI cannot do · 2026

Four things still need a human auditor.

Professional skepticism
Knowing when management's explanation is plausible — and when it is not.
Materiality judgment
What matters in this engagement, this year, this client. AI will not weigh it for you.
Client relationship
The conversation that changes the disclosure. Trust that took ten years to earn.
Final accountability
A signature on the audit report. A name on a regulatory inquiry. AI cannot sign.
22 / 32
Insight 3 of 3

Insight 03

Trust is the product.

AI speed
Hours, not weeks
Partner signoff
Named accountability
Client trust
SOCPA-licensed

Every AI output has a named partner who signs off. The AI moves fast. The partner stays accountable. We sell both.

23 / 32
What we built

Rung 5 · Accord OS

An audit operating system. Saudi-first. Partner in the loop.

A
Reads Arabic working papers
OCR + LLM extraction trained on Saudi templates and SOCPA forms.
B
Routes the right model
Claude for documents. o3 for filings. Gemini for exams. Right tool, right job.
C
Standardizes the file
Every engagement enters a single normalized format. Then audit logic runs.
D
Partner reviews. Signs. Owns.
Every output has a named human reviewer. No exceptions.
24 / 32
Live

Demo · 4 minutes

Live: Accord OS reads an Arabic working paper.

An internal tool. Not a product pitch. We are showing you the work.

25 / 32
Live demo

Switch to laptop

Switch to laptop screen / projector input now
26 / 32
Demo recap

What just happened

A junior auditor's afternoon, in 90 seconds.

90 sec
An Arabic working paper read, classified, structured, and flagged for one anomaly. Manual equivalent: two to four hours of junior work.

The junior auditor doesn't disappear. The afternoon disappears. The auditor moves to the harder question: is the anomaly actually wrong?

27 / 32
For students & early-career

If you are starting out

Your career does not shrink. It shifts.

Learn to question an AI output. That is the new audit skill.
Junior work changes. The bar moves up. Your judgment is paid for, not your typing.
We hire SOCPA candidates who are AI-curious. Find me after.
28 / 32
For partners & firm owners

If you run a firm

Modernizing alone is harder than modernizing with us.

Standardize before you automate. Or the AI will not stick.
The capital, the platform, the playbook — we have built it. You do not need to.
We are talking to Saudi firms quietly. If that includes you, talk to us after.
29 / 32
Open floor

Now

Questions.

Your questions are usually smarter than my answers.

30 / 32
One last number

Before we end

Three years from now, 99 of every 100 audited companies will be using AI.

The question is not whether the audit profession changes. The question is who is ready for the audit profession that comes next.

KPMG / Accenture, 2024–2025.
31 / 32
Thank you · شكراً لكم

The future is not something that happens to us.
It is something we build.

We are starting now. Find us after the session.

Amr Shady · Founder & CEO amr@accordpartners.ai accordpartners.ai
accordpartners.ai Replace before May 14
32 / 32