● 15-minute demo · for wealth & advisory firms in DIFC
One adviser, the output of three
Every screen shows one routine job at a wealth firm. Each one has the manual steps today, the AI steps, the time saved, and how a human checks the result before it reaches a client.
12 + 1
processes, plus the rulebook
~43 hrs
saved per adviser / month (estimate)
100%
human sign-off before anything goes out
Run sheet
| Time | Screen | The one line to say |
|---|---|---|
| 00:00 | Run sheet | "Twelve jobs your team does every week, then the rules that keep it safe." |
| 00:30 | 1 · What the industry uses | "52% of DIFC firms already use AI. Here's what the big banks use it for." |
| 01:30 | 2 · Pitch a new client | "A full prospect brief in 10 minutes instead of an afternoon." |
| 03:00 | 3 · Onboarding journey | "Enquiry to funded account, with AI doing the chasing at every step." |
| 04:00 | 4 · Fact-find form | "The form fills itself from the documents. You check, not type." |
| 05:00 | 5 · Portfolio AI | "Every portfolio checked against its model, every morning." |
| 06:30 | 6 · Equity & bond criteria | "Your house rules, applied the same way every time." |
| 08:00 | 7 · Research reports | "40 pages read, 1 page written, every line linked to its source." |
| 09:00 | 8 · News watch | "Only the news that touches a client's holdings." |
| 10:00 | 9 · Client email update | "Personal monthly updates for 200 clients, approved in one sitting." |
| 11:00 | 10 · Staff follow-ups | "Every promise from every meeting becomes a task with an owner and a date." |
| 12:30 | 11 · Verify & compliance | "AI drafts. The adviser decides. Everything is logged." |
| 13:30 | 12 · Next step | "Pick one process. Two-week pilot." |
| 14:00 | 13 · DIFC rules & safeguards | "Client data never touches the internet. Here is the rule, and here is what enforces it." |
● Process 2 · Pitching to new clients
Prospect brief before the first call
The executive types a name and a company. AI gathers what is public, maps it to your firm's services, and drafts the opening email.
Today, manual
- Google the person and company
- Read LinkedIn and news
- Guess which service fits
- Write the email from scratch
2 to 3 hours
With AI
- Enter name and company
- AI builds the brief with sources
- AI ranks the 3 services that fit
- Executive edits the draft email
10 minutes
Live demo · sample prospect
Prospect: Sarah M. (sample)CFO, logistics group, Jebel AliUK national, 6 yrs in UAE
› Reading public LinkedIn profile and company website…
› Company has ~400 staff in UAE → end-of-service gratuity liability likely
› UK national, UAE resident since 2020 → UK inheritance tax exposure under 2025 residence rules
› Matching to your firm's services…
✓ Fit 1: Corporate gratuity scheme (company) · Fit 2: Estate & IHT planning (personal) · Fit 3: Savings & investment
✓ Draft email ready · 4 sources attached · waiting for executive review
How it's checked: every fact in the brief carries its link. The executive opens any claim in one click. Nothing is sent automatically.
Example source types: company website, LinkedIn, Gulf News on the UAE gratuity savings scheme, UK IHT reforms for UAE expats
● Process 4 · Routine form (fact-find & KYC)
The form fills itself
The client uploads a passport, a bank statement and a payslip. AI reads them and fills the fact-find. It marks what it found, what needs a check, and what is missing.
Today, manual
- Chase documents by email
- Type details into the form
- Check source of wealth by hand
- Find gaps at compliance review
3 to 4 hours per client
With AI
- Client uploads to a secure link
- AI extracts and fills fields
- AI flags gaps on day one
- Adviser confirms each field
40 minutes
Client fact-find · sample client
Full namewaiting…
Nationalitywaiting…
UAE residencywaiting…
Monthly incomewaiting…
Source of wealthwaiting…
Risk profilewaiting…
UK property sale proofwaiting…
How it's checked: green fields came straight from a document and show the page they came from. Pink fields need the adviser's judgement. Orange means a document is missing, so the client gets an automatic request.
● Process 5 · Portfolio management update
Every portfolio checked every morning
AI compares each client's holdings with the agreed model. It flags drift, cash sitting idle and anything outside the risk profile. The adviser decides what to do.
USD 2.41m
sample portfolio value
+6.8%
year to date (sample)
2
drift alerts
Balanced
risk profile
Target model vs actual
Model
Actual
| Asset class | Model | Actual | Drift |
|---|---|---|---|
| Global equity | 45% | 53% | +8.0 ⚠ |
| Bonds | 35% | 29% | −6.0 ⚠ |
| Alternatives | 12% | 11% | −1.0 |
| Cash | 8% | 7% | −1.0 |
AI suggestion
› Pulling holdings from platform export (sample)…
› Equity is 8 points over the model, above the 5-point tolerance band
› Bonds are 6 points under the model
✓ Suggest: move ~USD 190k from equity funds to short-dated investment-grade bonds
✓ Suitability note drafted · sent to adviser for decision
How it's checked: AI only suggests. The adviser approves or rejects, and the reason is saved to the client file for the annual review.
Today
Quarterly spreadsheet check per client · 45 min eachWith AI
Daily check across all clients · adviser reviews alerts only● Process 6 · Looking into equity, bonds & companies · criteria to look
House rules, applied the same way every time
The firm sets its criteria once. AI screens the universe against them and explains each pass or fail in plain English. Move the sliders to see it work.
Criteria
Screen result ·
| Name (sample) | Type | P/E | Yield | Rating | D/E | Result |
|---|
How it's checked: every number links to its data source and date. An analyst signs off the shortlist before it goes on the buy list.
● Process 7 · Research reports
40 pages in, 1 page out
AI reads an annual report, a fund factsheet or a fixed-income note and writes a one-page summary in the firm's format. Every line points to the page it came from.
Input · sample
Annual report · 42 pagesQ2 results call transcriptFund factsheet› Reading 3 documents (61 pages)…
› Pulling revenue, margin, debt and guidance figures
› Checking each figure against the source table
✓ 1-page note ready · 14 figures · 14 page references
Research note · Gulf Logistics Holdings (sample)
Hold, with a watch on debt
| Revenue growth | +11% | p.14 |
| Operating margin | 18.2% | p.16 |
| Net debt / EBITDA | 2.9x | p.31 |
| 2027 guidance | "mid-single digit" | call, 14:20 |
Plain English: growing well and profitable, but borrowing rose to fund two new warehouses. Fits balanced clients. Not for cautious ones.
How it's checked: the analyst clicks any figure to open the source page. If AI can't find a figure in a document, it writes "not found" instead of guessing.
● Process 8 · Update of news
Only the news that touches a client
AI watches news, regulator notices and rate decisions. It matches each item to the clients who hold the affected asset or face the rule change, then rates the impact.
This morning's matched news
Headlines below are examples of the type of item, not live newsRule change
14 clients
UK inheritance tax now based on residence, not domicile
Affects UK nationals who left the UK recently. AI flags clients with under 10 years outside the UK.
HighAffects UK nationals who left the UK recently. AI flags clients with under 10 years outside the UK.
14 clients
UAE scheme
6 companies
Voluntary end-of-service savings scheme for private-sector staff
Corporate clients can move gratuity into invested funds. AI lists corporate clients that haven't reviewed it.
HighCorporate clients can move gratuity into invested funds. AI lists corporate clients that haven't reviewed it.
6 companies
Wills
22 clients
DIFC Wills now under exclusive DIFC Courts jurisdiction
Dubai Law No. 2 of 2025. AI flags clients with UAE assets and no registered will.
MediumDubai Law No. 2 of 2025. AI flags clients with UAE assets and no registered will.
Ross Naylor guide · confirm with DIFC Courts
22 clients
Markets
0 actions
Example: central bank holds rates
AI checks bond-heavy portfolios and cash yields. No action needed this time.
LowAI checks bond-heavy portfolios and cash yields. No action needed this time.
0 actions
How it's checked: AI links the original article and the regulator's own page where one exists. A rule change only becomes client advice after the adviser reads the primary source.
● Process 9 · Email to clients with their update status
200 personal updates, approved in one sitting
AI writes each client's monthly email from their own portfolio, their open actions and the news that affects them. The adviser reads, edits and approves.
To: James Whitfield (sample) · Subject: Your October update
Dear James,
Your portfolio is up 6.8% this year. Shares did well, so they now make up more of your portfolio than we agreed. I'd like to move part of that back into bonds to keep your risk where you want it. Can we talk this week?
Two other things:
1. We still need proof of your 2023 UK property sale to finish your file. You can upload it here.
2. The UK changed how inheritance tax applies to people who moved abroad. As you left in 2019, this may affect you. I'll cover it when we speak.
Kind regards,
Your adviser
Built from
Screen 5 · drift alertScreen 4 · missing documentScreen 8 · IHT newsHow it's checked: the figure 6.8% links to the portfolio report. No product is recommended in the email, and the adviser approves every one. Approved emails are saved to the client file.
✓ APPROVED · LOGGED
| Batch | Drafted | Approved |
|---|---|---|
| October updates | 200 | 141 |
● Process 11 · Verify, sources & compliance
AI drafts. The adviser decides.
This screen answers the question a DFSA-regulated firm will ask: what stops a mistake reaching a client?
1 · Sources on everything
Every number and claim links to a document page, a data feed or an article. If there is no source, AI says "not found".2 · Human sign-off
Nothing reaches a client without an adviser approving it. AI never sends, trades or gives advice on its own.3 · Full audit trail
Who approved what, when, and from which sources. Ready for a file review or a regulator visit.Audit log · sample
| Time | Item | AI action | Human action | Sources |
|---|---|---|---|---|
| 08:02 | Whitfield portfolio | Drift alert, rebalance suggested | Approved by adviser | 2 |
| 08:15 | Whitfield fact-find | 6 fields filled, 1 missing | 4 confirmed, 2 edited | 3 |
| 09:40 | Gulf Logistics note | Research summary | Analyst signed off | 14 |
| 10:05 | October emails | 200 drafts | 141 approved, 9 edited | — |
Data rules
- Client data stays in the firm's own systems or an enterprise AI account with no model training on it.
- No client names or documents go into free public chatbots.
- Access by role: advisers see their own clients only.
- Set up to fit the firm's DFSA conduct and data-protection duties. The firm's compliance officer signs off the setup.
Regulator references: DFSA rulebook (COB module) · DIFC Data Protection Law 2020
● Process 12 · Time saved & next step
Where the hours come back
Estimates for one adviser with about 40 active clients. A two-week pilot measures the real numbers in your firm.
| Process | Today / month | With AI / month | Saved |
|---|---|---|---|
| Prospect briefs (6) | 15 h | 1 h | 14 h |
| Onboarding (2 new clients) | 8 h | 1.5 h | 6.5 h |
| Portfolio checks | 10 h | 2 h | 8 h |
| Research notes (4) | 8 h | 2 h | 6 h |
| Client update emails | 5 h | 1.5 h | 3.5 h |
| Meeting notes, follow-ups & CRM | 6 h | 1 h | 5 h |
| Total | 52 h | 9 h | ~43 h |
These are planning estimates, not measured results. The pilot replaces them with your own figures.
Week 1
Pick one process
Most firms start with onboarding or client emails.Week 2
Run it on real files
Inside your own systems, with your compliance officer watching.Day 14
Measure and decide
Hours saved, errors caught, adviser feedback.Read the full implementation proposal One-page PDF
azizsaif
azizsaif.com · aziz@azizsaif.com · Dubai · UAE
● Process 1 · Research: what AI wealth firms actually use
This is already normal in DIFC
Before any demo, the facts. What the regulator found in DIFC, and what the largest wealth firms have rolled out.
52%
of DIFC firms use AI (33% in 2024)
+166%
growth in generative AI use in 12 months
79%
of use cases are internal operations
26%
using AI in critical areas have no governance
Source: DFSA Artificial Intelligence Survey 2025, 661 authorised firms, published 12 Nov 2025.
What leading firms have deployed
| Firm | AI use | Reported result | Source |
|---|---|---|---|
| Morgan Stanley | AI @ Morgan Stanley Assistant (research search) and Debrief (meeting notes, actions, draft emails, CRM entry, with client consent) | Around 30 minutes saved per meeting. 98% of adviser teams adopted the Assistant. | OpenAI case study · InvestmentNews |
| Bank of Singapore | AI agent drafts source-of-wealth reports for KYC | Drafting cut from about 10 days to about 1 hour. Staff still review each one. | Bank of Singapore release |
| Adviser firms | Jump, Zocks and Zeplyn: meeting notes straight into the CRM | Vendors claim 45 to 60 minutes saved per meeting. Roughly USD 67 to 199 per adviser per month. | Zocks vs Jump · Financial Planning |
Most common uses, in order
- Meeting notes and CRM updates
- Meeting preparation
- Onboarding and KYC paperwork
- Client emails and reviews
- Research summaries
- Portfolio monitoring and alerts
What this means for a firm of 5 to 10
The big banks built their own tools. A small DIFC firm gets the same benefit from off-the-shelf AI set up around its own process.
The DFSA's main concern is governance. Starting with a written AI policy and human sign-off puts a small firm ahead of the 26% the regulator flagged.
● Process 3 · Onboarding a client, end to end
Enquiry to funded account, nothing stuck
Eight steps from first contact to the first review. AI does the paperwork and the chasing. The adviser handles the relationship and the advice. Click a step.
4 → 1.5 wks
enquiry to funded (target, estimate)
0
files that wait silently on a missing document
1
place to see every client's stage
Pipeline today · sample
| Client (sample) | Stage | Waiting on | Days in stage | AI action |
|---|---|---|---|---|
| J. Whitfield | 3 · KYC | Proof of UK property sale | 6 | 2nd reminder sent, adviser alerted |
| R. Haddad family | 5 · Suitability | Adviser sign-off | 2 | Report drafted, in review |
| Nair Holdings (corporate) | 7 · Funding | First gratuity contribution | 1 | Payment instructions sent |
| S. Morgan | 2 · Discovery | Meeting on Thu | 0 | Meeting brief ready |
● Process 10 · Staff follow-up tracking
Every promise gets an owner
After each meeting, AI pulls the actions out of the notes and turns each one into a task with an owner and a due date. The manager sees what is late across the whole team.
Today
- Actions live in notebooks and inboxes
- Follow-ups depend on memory
- Manager asks "where are we?" in meetings
Things slip
With AI
- Actions pulled from meeting notes
- Owner and due date set automatically
- Morning digest per person, alert when late
Nothing slips
31
open follow-ups
4
overdue
92%
on time this month
1.2 days
average client reply time
Team tracker · sample
| Staff (sample) | Client | Follow-up | From | Due | Status |
|---|
› Reading meeting notes: R. Haddad family review, 10:00…
› Found 3 actions in the conversation
✓ Task: send DIFC will checklist → Adviser A · due tomorrow
✓ Task: request trust deed copy → Paraplanner · due in 3 days
✓ Task: book IHT review with partner lawyer → Adviser A · due in 7 days
How it's checked: each task links back to the line in the meeting notes it came from. Staff can edit or reassign before it's saved. The manager's weekly report is generated from the tracker, not typed.
● Process 13 · DIFC rules & what enforces them
Client data never touches the internet
The rule that makes a DIFC compliance officer say yes. Two worlds: an offline world for anything with a client in it, and a normal desk for public work. Below it, the laws this meets and the controls that enforce it.
World 1 · offline, inside the office
Client data and the AI that reads it
- Encrypted file server in the DIFC office, one folder per client
- AI model runs on a machine in the same room, no internet cable
- Screens 3 to 7 and 9 to 11 run here: onboarding, KYC, portfolio, research, emails, follow-ups, audit log
- Two encrypted backup drives, one in the office safe, one off-site, swapped weekly
- Staff see only their own clients. Every open and edit is logged
NO LINK
World 2 · normal internet desk
Public work only, no client names
- Prospect research, market news, regulator updates (screens 1, 2 and 8)
- Prompts and skills from azizsaif.com/ai-wealth
- Platform exports downloaded here, carried to World 1 on an approved encrypted drive
- Rule in the AI policy: no client name, document or account number, ever
The rules this has to meet
| Rule | What it asks | What the setup does |
|---|---|---|
| DIFC Data Protection Law No. 5 of 2020 | The firm is the controller: lawful basis, security, client rights, breach reporting | Data never leaves the firm. Access logged. Retention set |
| DIFC DP Regulation 10 (autonomous systems) | Tell clients an AI system is used and name it. Keep a human in control. Keep records. High-risk uses may need an Autonomous Systems Officer | AI drafts, adviser signs. One line in the client agreement. The audit log is the record |
| DFSA GEN 5.3, outsourcing 5.3.21 and 5.3.22 | Written agreement with any supplier, risk assessment, exit plan | Written scope, NDA, all logins handed back at go-live, runbook for the firm's IT |
| DFSA record keeping | Client files, suitability reports and logs kept and retrievable, 6 years cited | Server plus two offline backups |
| DFSA AI Survey 2025 expectations | Written AI policy, clear accountability, an AI register | Policy, register and named owner written before any AI runs |
Summaries, not the rulebook. The compliance officer confirms each line against the current DFSA Rulebook and DIFC Commissioner guidance. Sources: Regulation 10 text · Clyde & Co · DFSA GEN · DFSA AI Survey 2025
What Claude Code can enforce
Claude Code is the tool that builds and runs these workflows on the firm's own machines. Its controls are settings files the compliance officer can read, not promises.
Ask before actingEvery file write, command or email is shown first and needs a person's approval. Permission mode is set per firm, not per user.
Deny listsSettings can block tools outright:
WebFetch, WebSearch, Bash(curl*). In World 1 the model simply has no way to reach out.Hooks that logA hook runs on every tool call and writes who, what, when and which files to the audit log. A second hook can block any call that touches a path outside the client folder.
SandboxCommands run in a sandbox that limits which folders can be read or written, so a mistake can't reach the wrong client's files.
Rules in plain EnglishA
CLAUDE.md file carries the firm's rules: never recommend a product, never send without approval, always cite the page.No training on your dataUnder Anthropic's commercial terms, API and enterprise data is not used to train models. Zero-data-retention is available for enterprise accounts.
Honest limit: Claude Code itself needs a connection to the model. For a firm that accepts an enterprise cloud account with these controls, it runs as shown. For a strict no-internet firm, World 1 runs an open-weight model on the office machine with the same approval, deny and logging rules, and Claude Code is used to build and test the workflows, never with live client data.
● Before any AI touches a client file
1 · AI-use policy
What AI may do, what it may never do, who signs off, how clients are told.
2 · AI register
One line per use: purpose, data touched, human owner, risk level, date reviewed.
3 · Impact assessment
For onboarding and KYC first, since that is where personal data is heaviest.
Demo only. All client names, portfolios, companies and figures are invented sample data. Nothing here is investment advice or live market data. Linked sources were checked on 4 to 6 Oct 2026 and should be re-read before relying on them.