Umar M. Sharif.
Selected work

Analysis that moved the decision.

7 pieces from marketplace, fintech and services work. Client names withheld where under NDA; numbers shown as published ratios and indices.

Types of work: Diagnosis Programme design Research Point of view
Flagship roles

Where I built these methods.

TelecomTelenor and Jazz2010 to 2020

Loyalty, personalisation and offer design at national scale

Read the case →
Telco bundle penetration grew from 17% to 25%.
Before
17%
After
25%
+5% ARPU
Bundle penetration; ARPU vs a control group. Internal figures.
FintechDiagnosis2026

The activation bottleneck

Where the install-to-paid funnel of a regulated investing and learning app breaks, and the operating model built around that finding.

Read the case →
99.85% of installs never become paid, and most are lost before the second session.
Install
100%
Active session
22.6%
Viewed holdings
3.7%
First asset
1.3%
Setup done
1.1%
Paid
0.15%
Source: Mobile analytics event export, 192 days, anonymised
MarketplaceProgramme design2025

The first 60 days

An incubation programme for new marketplace vendors, built around speed to first order.

Vendors who get a first order fast generate about 6× the volume.
Orders per vendor, first 30 days (fastest = 100)
First order < 10 days
100
10 to 20 days
28
> 30 days
16
Source: Group vendor cohorts, 2024–25, anonymised
Read the case →
MarketplaceOrg design2025

Who should incubate new vendors?

3 organisation options compared; one incubation team with early graduation to account managers is selected.

The payoff from incubating the right vendors ranges from 4× to 21× by market.
21×
16×
11×
4×
Source: Ratio of 30-day orders, top-class vs same-grade vendors, A-grade, Markets 1 to 17, sorted
Read the case →
ServicesStrategyProposal2026

Where the AI gain lands

Role exposure, who keeps the gain, and a hybrid pricing model for an HR outsourcing and staffing group.

Automation removes billable hours before it removes cost.
Payroll officer HR operations administrator Billing and invoicing analyst Recruitment coordinator Back-office processing associate Recruitment consultant Compliance and statutory officer HR consultant Product and platform engineer Site supervisor and coordinator Executive search consultant Client relationship manager Field, facilities and support staff
13 role families. Filled = high exposure, outline = partial, dim = low. Ratings are analyst judgement, to be validated with the firm's own data.
Read the case →
Research synthesisEvidence2026

−19% to +35%: what the evidence says about AI at work

7 studies, 1 measured range. Licences alone can land anywhere on this line.

The range is the finding.
Slower / worseFaster / better
−19%
−19%
~0
+12.2%
+13.8%
+26.1%
+35%
Source: 7 peer-reviewed or preregistered studies, 2023–2026; not like-for-like, so no median is shown.
Read the case →
Programme designProposal

An executive AI programme built on leaders' own work

4 sessions where leaders work on their own material, with hands-on time held above a floor.

Every session is designed to clear the 65% hands-on floor.
83%
Apply
83%
Redesign
85%
Build
79%
Validate
Hands-on share of session time, planned. Dashed line: 65% floor.
ProposalCommitment: 12 leaders, 12 tested prototypes, 1 ranked pilot pipeline.
Point of view

Directing AI now has 4 levels

From wording a question well to wiring loops together with human checkpoints.

04Graph engineeringWire loops together with routing, permissions and human checkpoints.
03Loop engineeringInstruct, act, check against a standard, correct.
02Context engineeringFill the window with the right files, examples and standards.
01Prompt engineeringWord the question well.
How I work with data
01

Diagnose before building

02

Show the finding in one chart

03

Tie every number to an owner and a date

Bring one problem. Leave with the next move.

Book a 30-min AI Audit