The situation
Safari grew into a multi-brand agency serving the group’s own brands as internal clients. Leadership’s own estimate: 90%+ of marketing agencies run with no structure — the work depends on individual memory, personal style, and WhatsApp threads.
What was broken
- New employees took months to “absorb the vibe” before producing.
- Everyone worked their own way — quality changed from client to client.
- Requests arrived by WhatsApp, email and call; some vanished with no owner.
- Evaluation ran on impressions, creating unfairness and turnover risk.
How I ran it
- 01
Diagnosed first. Sat with the CEO and the department managers to find where work actually broke, rather than designing from theory.
- 02
Mapped the client journey into 9 stages — sales → onboarding → assets → research → strategy → execution → reporting → archive — with role-by-role SOPs against each.
- 03
Designed the ticket system as the communication backbone: SLA priorities set by the Account Manager, a no-silent-downgrade rule, and 5-level escalation ending at the Founder.
- 04
Defined “Done.” “The person after you doesn’t need to ask you anything,” enforced by a unified pre-delivery checklist instead of an opinion.
- 05
Compressed onboarding into one structured meeting: a 16-section intake questionnaire and a 22-block handover to the Marketing Manager.
- 06
Built the Strategy & Planning stack — five documents narrowing from business strategy down to a Monthly Plan of Action with owner, date and KPI on every deliverable.
- 07
Kept it alive. Updates after every explanation meeting, shipping in Arabic and English — a living document, not a PDF nobody opens.
0 sectionslive and governing daily work — V1, July 2026
0 meetingfull client onboarding, no call-backs
Owner + SLAon every single request
Scorecardsnumeric evaluation replaced gut feel
Shows I can take “we have no structure” and ship the actual structure — documents, rules, tools, adoption.
The situation
One marketing team produces for six brands — a Quran education academy, e-commerce, and agency clients — with work scattered across Drive, Sheets, WhatsApp and memory.
What was broken
- Double data entry and endless status meetings.
- No data answers: “how many posts did we publish for brand X in May?”
- Total person-dependence — one operator leaves, delivery stops.
How I ran it
- 01
Audited where the work actually lived — Drive, Sheets, WhatsApp, and people’s memory — before designing anything.
- 02
Designed a 16-database architecture on one spine: Business → Quarterly Plan → Monthly Plan → Objectives → 3 Labs, so every daily task traces to a quarterly goal.
- 03
Built the 3 production pipelines: Content (7 stages, idea → script → design → review → schedule → publish), Ads (with a buyer approval gate), and Setup.
- 04
Separated backend from frontend. The relational complexity stays hidden; each role works from one filtered dashboard — adoption dies the moment operators see the machinery.
- 05
Set the operating cadence: ~20 monthly objectives per brand (10 content + 10 ads), weekly reports and meetings, and a daily loop so the team finds its work ready.
- 06
Wrote the operating manual — 10-minute response SLA, locked deadlines with auto-logged completion dates, filtered sharing, everything recorded as data.
- 07
Trained the team through 3 recorded sessions (~3.5 hours) that now double as onboarding material.
0 brandsone system, full traceability
0 databasesquarterly plans down to daily tasks
Per-persondelivery measured from auto-logged data
Replaceablework continues when people change
Shows Multi-brand systems architecture — plus the change management that makes it stick.
The situation
Strategy documents, content briefs and customer responses all consumed senior time, and brand voice shifted with whoever happened to be writing.
What was broken
- Every strategy and brief started from a blank page.
- Brand voice shifted with every writer.
- Replies and bookings for a large student base ate staff time.
- Hiring consumed weeks of HR capacity per role.
How I ran it
- 01
Traced where the hours went before buying or building anything — blank-page strategy work, inconsistent voice, and manual reply handling.
- 02
Built per-brand Claude plugins with skills for quarterly strategy, monthly planning and weekly content delivery — fed with 11 programs, 3 books and the brand assets.
- 03
Standardised the pipeline research → quarterly strategy → strategic brief → content, delivering designer-ready briefs into the Content Lab via Notion and MCP.
- 04
Chose buy over build for the customer-facing bot: scoped requirements, evaluated providers, subscribed, and folded it into the academy’s response and booking workflow — speed and cost won.
- 05
Extended it to hiring. An AI recruiting pipeline writes the job post and application link, screens applicants, and runs the first interview with no human in the loop.
Unified voiceany creator can be replaced, tone stays
Designer-readybriefs generated in a fixed format
Automatedreplies, bookings and first-round interviews
Fasterstrategy and briefs no longer start from zero
Shows Practical AI adoption tied to operations — plus build-vs-buy judgement.
The situation
The group’s online academy serves 1,000+ students across countries and time zones, and generates most of the group’s revenue. Scheduling, reminders and supervision ran by hand every single day.
What was broken
- Reminders sent to students and teachers by hand at 12:00, daily.
- A team leader distributed sessions across supervisors manually, every day.
- Postponement requests lost in WhatsApp threads.
- No numbers: who attended, who stopped, how many trials?
How I ran it
- 01
Spotted it in the operations data, not in a complaint — the manual 12:00 send and the manual daily distribution.
- 02
Sized the risk before escalating. The academy generates most of the group’s revenue, so scheduling failures were a revenue problem, not an admin one.
- 03
Ran the research on buy versus build, then took the recommendation to the CEO and got approval.
- 04
Sourced the developer through the network built across previous roles, and wrote the requirements brief: automation rules, roles, and the student lifecycle.
- 05
Controlled the build with a written approval system. Scope had been agreed verbally and forgotten; I replaced that with a documented approve step on both sides.
- 06
Specified the automation core: sessions auto-generated per package, automatic time-zone conversion, WhatsApp reminders with reply capture, and equal distribution by shift.
- 07
Reviewed and accepted delivery against the brief, then rolled out with a 3-guide documentation series.
- 08
Trained the operating team — 4 supervisors and 2 team leads — and stayed on it through the first cycles.
0+ studentsacross countries and time zones
0 people4 supervisors + 2 team leads operating it
0 guidesrole-based rollout documentation
Live filtersattendance, retention, trial → subscription
Shows Product-minded operations: an operational mess turned into a specified, delivered, adopted system.
The situation
A CEO overseeing four departments was pulled into daily detail, and reducing exactly that was the organisation’s stated goal.
What was broken
- No consistent picture of who delivered what, and when.
- Meetings consumed by status updates instead of decisions.
- The CEO was the integration point for everything.
How I ran it
- 01
Named the real problem: the CEO was the integration point for four departments, and reducing that was the stated goal.
- 02
Implemented daily and weekly structured Trello reports from each manager to the CEO, replacing ad-hoc chat with one regular, dated source of truth.
- 03
Rebuilt the weekly meeting around a prepared agenda issued a day ahead — the top 2–3 operational problems, each arriving with a proposed solution.
- 04
Documented every decision with an owner and a deadline, followed up the next week so nothing depended on memory.
- 05
Maintained a delivery tracker showing each manager’s on-time versus late record — accountability from data, not impressions.
0 source of truthregular, structured, dated
Owner + deadlineon every action item
0–0 mindecision meetings, not status theatre
Patterns visiblerecurring delays surfaced per department
Shows Managing upward, designing accountability, and freeing leadership bandwidth.
The situation
A UAE engineering training academy with a flat social presence, no calendar, no briefs, and no numbers anyone tracked.
What was broken
- Sporadic posting with no calendar, briefs, or review loop.
- No paid strategy; reach hovered near 13K with 32 inbound conversations a month.
- Performance was invisible — nothing measured.
How I ran it
- 01
Audited the account first — sporadic posting, no calendar, no briefs, no review loop, no numbers.
- 02
Rebuilt execution as a system: a monthly content calendar, scripts, and briefs for the freelance designer and video editor, with a review loop before publishing.
- 03
Managed the freelancers to the brief rather than to taste, so output stayed consistent without me producing it.
- 04
Launched Meta campaigns optimised for started conversations — the metric tied to enrolment, not to vanity reach.
- 05
Iterated spend against cost per conversation week over week, tracking before/after windows in Meta analytics so the result was verifiable.
0× reach12,988 → 111,042
0× conversations32 → 250 started
0× followers78 → 1,422
EGP 0cost per started conversation
Shows I can own a growth scope end to end — and prove the result with data.
The situation
The academy had passed 1,000 students and stalled there. Trial sessions kept filling but subscriptions did not follow, and the Egyptian market — the largest addressable one — had collapsed almost entirely. Nobody could say why.
What was broken
- Out of every 100 students who took a trial, only 7 subscribed.
- Egypt had stopped converting; the room assumed demand was gone.
- The Marketing Manager resigned, leaving no campaign for the next month.
How I ran it
- 01
Went to the data, not the opinions. Requested operations data per cohort from the Operations Manager instead of debating causes in a meeting.
- 02
Found the pattern. Saudi students were attending trials in volume and then not subscribing — the drop-off was concentrated, not general.
- 03
Challenged the obvious answer. The room assumed pricing. The data said the loss happened after the trial, which pointed at the session itself.
- 04
Formed the real hypothesis: the Egyptian teachers’ dialect was the friction for Saudi students — a delivery problem wearing a pricing problem’s clothes.
- 05
Tested it cheaply. Recruited 2 Yemeni teachers whose accent sits far closer to the Saudi ear, and routed Saudi trials to them.
- 06
Took the pricing question separately. For Egypt the data did point at price — a large market priced for the Gulf. Rebuilt Egypt pricing with cost control so the lower price still cleared margin.
- 07
Held the line when the team broke. With the Marketing Manager gone, I wrote the campaign strategy myself in one day and it was approved first pass (July 2026).
0 → 0trial-to-subscription per 100 · 3.1×
0%+growth in the Egyptian market
0 dayreplacement campaign, approved first pass
0 teachersthe whole cost of the test
Shows I am not only a systems builder. I read operational data, isolate the cause behind a revenue number, and carry the fix through execution.