The Mobilink Microfinance Bank cohort, NIBAF Islamabad, 23 September 2026
thinkFRST Mobilink Microfinance Bank · 23 September 2026
One day · NIBAF, Islamabad

Twenty people. One day. Fifteen working prototypes.

Not slides about AI. Fifteen things the bank can actually test, built by the people who have the problem, before five o’clock.

thinkfrst-mmbl-enablers.vercel.app
thinkFRST| STRATEGY BEFORE AI.

Your AI Enabler workbook.

7 tools. One free ChatGPT account. One rule: AI asks, you think.

START HERE · YOUR GROUP

Pick once. Every tool loads your department’s tasks and problems.

Pick your group ...
STARTED

01.

LEARN · CHATGPT + SUNO

Learn CO-STAR. Build a song

Your own song in fifteen minutes.

OPEN TOOL →

02.

APPLY · CHATGPT

CO-STAR on a real task

The same six parts on a task from your week.

OPEN TOOL →

03.

DEFINE · CHATGPT

Define your problem

One problem, one sentence, one root cause.

OPEN TOOL →

04.

IDEATE · CHATGPT

Idea round. Pick one

4 people, 3 ideas, 4 rounds.

OPEN TOOL →

05.

PROTOTYPE · GOOGLE AI STUDIO

See your idea work

AI Studio builds a small app.

OPEN TOOL →

06.

PRESENT · GAMMA

Prototype to pitch

Six slides with your app.

OPEN TOOL →

thinkFRST← WORKBOOK HOME

Make your first project using prompting.

Fill 6 gaps. Get lyrics from ChatGPT. Generate the song in Suno.

C · CONTEXT

Why are you writing this song?

Pick one ...
O · OBJECTIVE

What should the song do?

Pick one ...
S · STYLE

Pick the genre.

Pick one ...
T · TONE

Pick the mood.

Pick one ...
A · AUDIENCE

Who hears this song?

Pick one ...
R · RESPONSE

How long should it be?

Pick one ...

YOUR PROMPT · LIVE PREVIEW

Write a song.

Context. We want to laugh about how chaotic our weeks get.

Objective. Make us sing along. Style. qawwali. Tone. cheeky.

Audience. Our batch mates in this room. Response. A 90-second song.

COPY THE PROMPT
thinkFRSTGROUP 1 · AUDIT, COMPLIANCE, INFOSEC

Define your problem.

Pick one problem your table lived last week. ChatGPT X-rays it in three questions.

1 · THE PROBLEM · TAP THE ONE SOMEONE AT YOUR TABLE LIVED

FROM YOUR SURVEY
Technical work is rewritten by hand into formatted reports
FROM YOUR SURVEY
Audit sampling and evidence chasing takes 5 to 10 hours a week
PROPOSED
New SBP circulars are read and mapped to internal policies by hand

WHO DID THIS LAST WEEK

Pick a role ...

HOURS A WEEK

Pick one ...
THE SENTENCE, SCORED 8/10

Internal auditors spend five to ten hours a week rewriting the same information into several report sections.

thinkFRSTGROUP 1 · AUDIT, COMPLIANCE, INFOSEC

Idea round. Pick one.

Four people, three ideas, four rounds. Your table picks one.

Drafts the condition, cause, risk and recommendation from a rough noteHIGH IMPACT · LOW EFFORT
A shared template library everyone fills in by handLOW IMPACT · LOW EFFORT
A workflow system wired into the audit platformHIGH IMPACT · HIGH EFFORT
thinkFRSTGROUP 1 · AUDIT, COMPLIANCE, INFOSEC

See your idea work.

Your idea becomes a small working app in Google AI Studio. Then the next table tries it.

THE PROMPT · CARRIED IN FROM 03 AND 04

Your problem. Internal auditors spend five to ten hours a week rewriting the same information into several report sections.

Your idea. Drafts the condition, cause, risk and recommendation from a rough note.

Who uses the app. Internal auditor.

Risk and control. AI may misread detail. Human verification required before use.

COPY THE PROMPT → OPEN AI STUDIO

BUILDING IN AI STUDIO…

thinkFRSTBUILT IN THE ROOM · 23 SEP 2026

Audit Finding Writer.

A WORKING PROTOTYPE, BUILT BY AN INTERNAL AUDITOR

ROUGH NOTE IN

“checked 40 vouchers, 6 had no approval sig, policy says dual sign over 50k, told the branch twice already”

CONDITIONSix of forty vouchers sampled carried no approval signature.
CAUSEDual-signature policy above the fifty thousand threshold is not enforced at branch level.
RISKUnapproved disbursement. Repeat finding, raised with the branch twice before.
RECOMMENDATIONSystem-level block above threshold without a second signature.
HUMAN CHECK REQUIRED. The auditor confirms every number, every piece of evidence and the final risk rating before this leaves the file.
thinkFRSTGROUP 1 · AUDIT, COMPLIANCE, INFOSEC

Prototype to pitch.

Six slides with your app inside them. Present in three minutes.

1The problem
2Who it is for
3What the AI does
4What the person still does
5The one metric
6The fourteen day pilot

15PROTOTYPES BY 5PM

One person. Eight hours.

  • A problem defined in a sentence they can defend
  • An idea their table chose over two others
  • A working app, tested by someone else in the room
  • A six slide pitch with the app inside it
  • Five AI buddies and twenty ready prompts for Monday
WORKBOOK HOME

This is the real workbook they used, replayed step by step. Open the live tool →

20management trainees in the room
7browser tools built for the day
15working prototypes finished
12ready to pilot as they stand
0software licences bought
The Mobilink Microfinance Bank cohort with Think FRST, NIBAF Islamabad, 23 September 2026
The cohort, NIBAF Islamabad, 23 September 2026.
BY FIVE O’CLOCK15

This is what they built.

Each one chosen by the person who has the problem. Hours are their own estimate. Readiness is our call, for the bank to confirm.

01READY TO PILOT

Circular Clause Extractor

Regulatory Compliance

Reads every new SBP circular and pulls the clauses that bind the bank, with the page number.

6 to 8 hrsback a week, from 10+
02READY TO PILOT

Evidence to Observation Matcher

Internal Audit, IT Audit

Matches returned evidence against the observation sheet, closes what is satisfied, re-issues the rest.

1 to 2 wksper audit area, cut to checking only
03READY TO PILOT

Contract Data Standardiser

Procurement

Twelve departments, twelve formats, one output. Flags anything it is unsure about.

10 mina month, down from 2 to 5 hours
04READY TO PILOT

Smart Proposal Designer

Business Banking

Builds the client proposal structure and the visuals, in bank branding.

2 to 3 hrsback a week per officer
05READY TO PILOT

Evidence Request Generator

Internal Audit

Groups sample transactions by department and drafts the chase emails, with a deadline.

1 to 2 hrsback a week, from 5 to 10
06READY TO PILOT

AI Exploratory Tester

Quality Assurance

Reads the requirement, writes the test cases, produces the bug sheet.

3 to 5 hrsback a week, from 5 to 10
07READY TO PILOT

Conversation Notes to One Pager

Information Security

Turns scattered meeting notes into the standard one page format.

5 to 8 hrsback a week, from 5 to 10
08READY TO PILOT

AI Stack Recommender

AI Associate, Tooling

Matches a use case to the tools and models that fit inside bank constraints.

5 to 8 hrsback a week, from 5 to 10
09READY TO PILOT

Email to Tracker Parser

Strategy, Project Tracking

Reads forwarded project mail and writes tasks, decisions and owners into one sheet.

3 to 5 hrsback a week, from 5 to 10
10READY TO PILOT

Client Follow-up Writer

Finance, Receivables

Writes a personal recovery reminder for each client from the ageing data. Nothing sends itself.

1 to 2 hrsback a week, from under 5
11READY TO PILOT

AI Manual Risk Scanner

Risk

Scans branchless banking manuals for access, data and control gaps before the meeting.

2 to 4 hrsback a week, from 2 to 5
12READY TO PILOT

Automation Opportunity Scanner

Credit Operations

Finds the manual steps worth automating, and skips what the loan system already covers.

1 to 2 hrsback a week, from 2 to 5
13METRIC TO SET FIRST

Executive Report Standardiser

Information Systems

Turns plain technical notes into a first draft report for the right audience.

Not yetmeasured. Baseline set in the pilot
14NEEDS A DECISION FIRST

Merchant Scoring Matrix

Risk, Merchant Compliance

Scores merchants against a rules matrix and buckets them high, medium or low.

Not yetmeasured. Scoring rules come first
15STAGED BUILD

AI Credit Risk Engine

Credit and Risk

Assesses affordability, repayment risk and collateral, and proposes a grade with the evidence attached.

5 to 10 hrsanalyst time back a week

The moment it stopped being training

Nobody told them to put a human check in it.

A participant presenting her prototype
A participant presenting her prototype
A participant presenting her prototype
A participant presenting his prototype

All fifteen did it anyway.

Three minutes each, standing up, taking questions from the people they sit next to. One builder was asked straight out whether customer CNICs went into her tool. They did not. Amounts and ageing only.

The hard questions came from the floor, not from us. Three people named what stood between their build and a real pilot before anyone asked.

The method

Training ends with notes. This ended with fifteen things to test.

CONVENTIONAL AI TRAINING A trainer’s slidessomeone else’s examples A generic exercisenobody’s real job A quiz and a scoremeasured on recall A certificatefiled, then forgotten Nothing to show on Monday. The capability leaves with the trainer. Same eight hours. Same room. Same people. THE THINK FRST METHOD THEIR OWN PROBLEM surfaced a week before, by survey DEFINE one sentence or it fails IDEATE three ideas, the table picks BUILD free tools, no procurement BREAK IT the next table tests your app DEFEND three minutes, peers question 15 prototypes by 5pm Every step runs inside one workbook, and the workbook stays. WHAT THE BANK GETS A shortlist it can pilot in a fortnight, each with a named owner and one metric. WHAT THE PERSON GETS Something they built and defended, and the tool to do it again on Monday. WHAT NOBODY GETS A certificate, a recording, or a folder they never open again.

The rules of the day

Build inside the bank as it is, not as it might be.

The bank hires a cohort of AI Associates and Business Graduates every year. This time it wanted them building on day one.

Their problem, not oursEach person brought a task from their own week. Nobody was handed an exercise.
No customer dataNo names, no CNIC, no account numbers. Mask before you paste.
Pilot in fourteen daysEvery build had to be testable in a fortnight, next to the way the work is done now.
Participants building their prototypes on their own laptops
A group working through the define step together
Usman Zahid running the day at NIBAF, Islamabad

Who ran the day

One rule all day. AI asks, you think.

Usman built the workbook the cohort used, ran the room at NIBAF, and pushed every table from problem to working prototype by five o’clock.

Usman ZahidPartner and Chief AI Officer, Think FRST

The workbook is still open

A day produces prototypes. The workbook is why the next fifteen won’t need us in the room.