Meet the loops: software that pays attention so we don't have to remember to.
We're giving the company a set of tireless digital watchmen — AI agents that check our numbers on a schedule, tell the right person what changed, and never touch anything without a human saying yes.
Prepared by IT · Nish Faria · Discussion draft — nothing here is switched on until we agree it should be.
01 — Why we're doing this
Our biggest leaks weren't caused by bad decisions. They were caused by nobody looking.
These four numbers are real, from our own systems, verified this month. None of them happened because someone did their job badly — they happened because no one's job was to keep watching.
AED 0
spent on online ads over 14 months with zero verified sales — because no tracking was ever connected to tell us what worked.
Marketing
AED 0
of warehouse stock that hasn't moved in over six months — about 24% of everything on our shelves. Invisible until we queried for it.
Inventory
0%
of our 1,144 CRM deals have no scheduled next step. Deals don't die from a "no" — they die from silence.
Sales
0
our main data pipeline sat broken without anyone noticing — found by accident while researching this very project. It's now fixed.
Data
02 — What is a loop?
Think of a night watchman who never gets bored.
A loop is an AI agent given a repeating job: wake up on a schedule, walk the same route through our systems, compare what it finds against rules we wrote, send a short message to a named person, and go back to sleep. Watch one do its rounds:
⏰
Wake
On a schedule — every morning, every Monday
👀
Look
Reads our systems: ERP, CRM, dashboards
📋
Check
Against thresholds written by us
💬
Report
Short digest to one named person
✋
Ask & sleep
Actions wait for human approval
Watch the loop do its rounds…Round 1
The four kinds of loop, in plain words
Kind 1
Ask once
You ask, it does the job once and stops. Like asking a fast colleague for a report.
Example: "Summarise yesterday's orders."
Kind 2
Goal loop
You set a finish line and a maximum number of tries. It keeps working until it crosses the line.
Example: "Propose supplier lead-times for 50 items; stop when all 50 have evidence attached."
Kind 3
Timed loop
The watchman: the same checks every day or week, forever, until we switch it off. Most of our loops are this kind.
Example: "Every Monday: what stock hasn't moved in six months?"
Kind 4
Proactive loop
Reacts to what it notices without being asked — but still asks permission before acting.
Example: "This ad is wearing out → here are three replacement drafts. Approve any?"
03 — The three promises
Every loop keeps three promises. No exceptions.
1 Never autopilot
No loop sends a customer an email, changes ERP or CRM data, publishes anything, or spends a dirham on its own. Loops watch and draft; a named person taps Approve before anything happens. Watching is automatic — acting never is.
2 It comes to you
You don't log in anywhere or remember to ask. The digest arrives on its schedule — as a plain email to begin with, and as a card on the company feed as that rolls out. If it's not useful, tell us — loops are cheap to change and cheap to kill.
3 Never silent
"Nothing found today" is a delivered result. If a loop stops reporting, that itself sets off an alarm — we watch the watchmen. We learned this the hard way: our data pipeline sat broken for 11 days because nothing was assigned to notice.
04 — What you'll actually see
A 20-second read with your morning coffee.
This is the entire experience for most people: one short message, written in plain language, from a loop that spent the early morning reading systems so you don't have to. Green means checked and fine. Coral means look at this. Buttons mean a decision is yours to make — nothing happens until you say yes.
Digests will reach you on whichever channel fits: as cards like this, on the company feed (feed.huxapps.com) as it rolls out — or, to begin with, as a simple email. Same numbers, same plain language, whichever way it arrives.
HX
Huxberry BizOps
bot · Monday 08:30
Dead-stock watchdog
Morning ☀️ Weekly stock check done.
3 items newly crossed 6 months with no movement (AED 41K)
Top sleeper: Visco foam blocks 40D — AED 18.2K sitting since Dec
2 items revived — sold last week after 8 months idle
Total unmoved 6+ months: AED 986K (▼ AED 3K vs last week)
08:30
Suggestion — needs your OK
Draft liquidation shortlist ready for the top 5 sleepers. Want it as a spreadsheet to review?
Yes, send itNot now
08:30
Data check
✓ All pipelines green — numbers above are as of this morning, 07:15.
08:31
To begin with — the same digest, as an email
The company feed is still being finished, so the first loops will report by email — nothing to install, nothing to learn. Here's the exact same Monday morning, in your inbox:
Weekly stock check — 3 new sleepers (AED 41K) · rest all quiet
Good morning — weekly stock check done. Three things worth your eyes:
3 items newly crossed 6 months with no movement (AED 41K total)
Top sleeper: Visco foam blocks 40D — AED 18.2K sitting since December
2 items revived — sold last week after 8 months idle
Total unmoved for 6+ months: AED 986K (down AED 3K vs last week).
A draft liquidation shortlist for the top 5 sleepers is ready — one click and it lands in your inbox as a spreadsheet to review. Nothing is discounted or moved until you decide.
Send me the shortlist
✓ All pipelines green — the numbers above are as of this morning, 07:15. You're receiving this because you're the named recipient of the dead-stock watchdog. Reply to this email to tune what it shows you.
05 — What's in it for you
Every loop reports to a person, not to a system.
You own the thresholds: "stop showing me X, add Y" is exactly the feedback we want. Here's who hears from which watchmen.
SS
Shahbaz
Finance
Which invoices just went overdue — with ready-drafted chase text
Bank facility deadlines & monthly compliance checklist
Monthly ad-spend vs invoice vs bank reconciliation
Expense claims stuck in the review queue
LF
Lerris
Purchase & Stores
Weekly dead-stock report with liquidation candidates
Lead-time & reorder-point proposals to approve in batches
"Order fabric by this date" alerts for long-lead items
Twice-monthly supplier & wool-price intelligence
RF
Rajiv
Customer Success · CRM
Daily: deals that lost their next step, before they go cold
Weekly pipeline movement across all four BUs
Lost deals missing a reason — so we finally learn why we lose
Nineteen company loops, seven Huxberry marketing loops, and a three-loop B2B pipeline. Expand any of them.
Each one says what it watches, when it speaks, and what it will never do. Status shows how ready it is: ready = data verified, can be built now · needs prep = a small investigation or setup first · needs plumbing = new data connection required before it can see anything. Implementers: the full technical designs — verified data sources, queries, thresholds, build notes — are on the build sheets page.
ReadyNeeds prepNeeds plumbingWave = build order
Data trust
reports to Nish
DATA-0Pipeline sentinel
ReadyWave 1
Watches
Every automated data pipe we have — did last night's syncs actually run, and how fresh is each number the company sees?
Speaks
Only when something needs you: a problem, or a one-time "recovered" note when it clears. Green days stay out of your inbox — a machine-level dead-man's-switch proves the sentinel itself is alive, and Monday's digest carries a one-line "data pipelines: 7/7 green."
Why it exists
Our warehouse sync broke on June 27 and sat broken for 11 days — every inventory dashboard quietly showed stale numbers. This loop makes that impossible to repeat.
Never touches the pipes — it only reports. Repairs are done by people (or approved separately).
Sales & CRM
reports to Rajiv + BU leads
CRM-1Deal hygiene watchdog
ReadyWave 1
Watches
Every open deal in the CRM. Today 93% have no scheduled next step — this loop catches a deal the day its follow-up goes missing, sorted by value.
Speaks
Daily 07:30 per team; Monday full report with trends ("no-next-step: 93% → 78%").
Can offer
Batch fixes — "schedule a follow-up on these 12 deals?" — applied only after an Approve tap.
Never contacts a customer, never edits a deal without approval, never blames — it reports numbers, not names, to anyone but the team's own lead.
CRM-2Pipeline pulse
ReadyWave 3
Watches
Week-on-week movement of each BU's pipeline: what advanced, what stalled, what was won and lost — the first pipeline visibility the BU leads have ever had.
Speaks
Monday morning: one card per BU on the company feed + a short message to each lead.
Waits for
Two weeks of CRM-1 clean-up first — today's pipeline numbers are junk-inflated, and we won't report numbers we don't trust.
Never ranks salespeople against each other — each lead sees their own BU.
CRM-3Lead intelligence — hospitality first
Needs prepWave 3
Watches
Public announcements of GCC hotel openings, refurbishments and tenders — then qualifies them against our ideal customer profile.
Speaks
Weekly shortlist to Ifham with sources and a suggested angle; a construction version for Yasar follows.
Why hospitality first
Marriott & IHG dropped the Hypnos brand globally — a large share of hospitality revenue needs replacing, and Ifham currently prospects alone.
Never adds a lead to the CRM without Ifham approving it, one by one.
Inventory & purchasing
reports to Lerris
INV-1Dead-stock watchdog
ReadyWave 1
Watches
Every item on our shelves. First verified run already found AED ~989K (24% of stock value) unmoved for 6+ months — 898 items.
Speaks
Monday 08:30: what newly went to sleep, what woke up, top sleepers by value, and a liquidation shortlist on request.
Never discounts, moves, or writes off anything — liquidation ideas are suggestions for humans to act on.
INV-2Stock-settings builder (lead times & reorder points)
Needs plumbingWave 3
The problem
Only 9% of items have a supplier lead time on record and 0% have reorder points — so the ERP can't warn us before we run out. This is also homework the Odoo migration needs done.
How it works
A goal loop: takes 50 items at a time, studies our purchase history, proposes values with evidence, and hands Lerris a spreadsheet to approve. Repeat until covered.
Never writes settings into the ERP itself — approved values go in through the migration team.
INV-3Reorder-risk watchdog
Needs plumbingWave 4
Watches
Stock levels vs. supplier lead times for long-lead materials (fabric can take 40–95 days) and open purchase orders.
Speaks
Daily, only when action is needed: "order X by the 15th or production waits."
Waits for
INV-2's lead-time data — it can't compute risk without knowing how long things take to arrive.
Never places a purchase order.
INV-4Supplier & commodity intelligence
Needs prepWave 3
Watches
Our key material suppliers (wool, Tencel, foam) and world wool prices — the intelligence briefings we used to get before the old agent was retired in May.
Speaks
1st and 15th of each month: a short brief that also files itself into the company knowledge base.
Never contacts a supplier.
Finance
reports to Shahbaz (Rajiv oversight)
FIN-1Receivables watchdog
Needs plumbingWave 3
Watches
Every unpaid invoice, bucketed by how overdue it is. Flags the moment an account crosses 30/60/90 days.
Speaks
Daily: what newly went overdue, top 10 by amount, with polite chase-email text pre-drafted for the team to copy.
Waits for
One data connection: invoice data isn't in our reporting warehouse yet. A single extension unlocks this loop and FIN-2's calculations together.
Never emails a customer. Drafts only — always sent by a person.
FIN-2Bank compliance calendar
ReadyWave 2
Watches
The calendar of obligations attached to our bank facilities — reporting deadlines, review dates, and the monthly self-checks our covenants require, including which customers' payments must route through which bank.
Speaks
Monthly checklist to Shahbaz; every confirmation is logged, building the audit trail our banks like to see. Escalates to Rajiv if a month goes unanswered.
Never sees or touches bank accounts — it manages the calendar and the evidence, people manage the money. (Facility details are kept off this page deliberately.)
FIN-3Ad-spend reconciliation
Needs prepWave 4
Watches
Three numbers that should always match: what ad platforms say we spent, what we were invoiced, and what left the bank.
Speaks
Monthly reconciliation table with any unexplained gap flagged.
Why
This is the loop that makes another AED 720K structurally impossible.
Never approves or blocks a payment.
FIN-4Expense-queue nudge
Needs prepWave 3
Watches
The new expense-capture app's review queue: claims stuck waiting more than a week, corrections nobody answered.
Speaks
Weekly, only when something is stuck.
Waits for
The expense app itself finishing its pilot.
Never approves or rejects a claim.
Operations & construction
reports to Yasar, Ifham, factory leads
OPS-1Retention & DLP money tracker
Needs prepWave 2
Watches
Construction projects where clients hold back retention money. The final 50% becomes claimable exactly one year after handover — a date currently tracked in people's heads.
Speaks
Weekly: money becoming claimable in 90 / 30 / 0 days, with amounts; plus handover-stage retention not yet invoiced.
First step
A one-time check of the billing system's records; if handover dates are missing, the loop's first job is politely collecting them, five projects a week.
Never invoices anyone — it makes sure we never forget to.
OPS-2Site-milestone speed monitor
Needs plumbingWave 4
Watches
How long site paperwork stages take (delivery inspections, work inspections) — the steps the business has already said it wants to speed up.
Honest status
The stage dates may not be captured anywhere yet. If they aren't, this becomes a small feature request for the billing app first — we'll say so rather than pretend.
Never chases site staff directly — findings go to Yasar.
OPS-3Protector replenishment radar
ReadyWave 2
Watches
Hotel customers' purchase history for mattress protectors — an annually recurring order we currently leave to memory.
Speaks
Monthly to Ifham & Mithun: "these hotels bought 10–14 months ago and haven't reordered — worth a call."
Never contacts a hotel.
Marketing — Huxberry.com
reports to Sherif
HUX-L0Tracking foundation (one-time project)
Needs prepWave 1
What
Not a loop — the plumbing that makes marketing loops honest: connect the shop to ad platforms and analytics so a sale can finally be traced to the ad that caused it. The AED 720K happened because this was never done.
HUX-L1Ad creative factory
ReadyWave 2
What
Give it a one-line brief; it designs ad variants and copy, checks every piece against the Huxberry brand rules, and files them in the photo library for Sherif to approve.
Never uploads to Meta or Google — approved humans publish.
HUX-L2Search-ranking & competitor watch
ReadyWave 2
Watches
Where huxberry.com ranks on Google for the money keywords (we're #2 for "organic mattress Dubai"), and what competitors change on their sites and prices — daily.
Never touches the website.
HUX-L3Ads account watchdog
Needs plumbingWave 3
Watches
Every dirham in the ad accounts, daily: spend without results, disapproved ads, budgets pacing wrong. Example rule: "any ad set over AED 500 spend with zero sales in 7 days turns coral."
Waits for
Read-only access keys to the ad accounts, and HUX-L0 so "results" means something.
Never edits campaigns or budgets — read-only by construction.
HUX-L4Content pillar loop
ReadyWave 3
What
Weekly: drafts one planned article (the Gulf-climate sleep guide is first — a gap no GCC competitor has filled), runs its own quality checks, and sends it for review. Publishes only after sign-off.
HUX-L5Creative refresh (proactive)
Needs prepWave 4
What
The composite: when the ads watchdog sees an ad wearing out, it automatically asks the creative factory for replacement drafts. Detection to draft with no human effort — publishing still needs Sherif's tap.
HUX-L6Negative-keyword janitor — the first approved edit
Needs prepWave 4
Watches
The ads watchdog's weekly list of Google search queries that spent money and produced zero sales — e.g. any query over AED 100 spend with no conversions in 30 days.
Does
Weekly: shows Sherif the junk-query list with the spend evidence, and on his Approve, adds them as exclusions ("negative keywords") — so we permanently stop paying for searches that never buy.
Why this is the one edit we allow first
Smallest possible blast radius: the worst case is that we stop paying for a junk search — and it's reversible in one click. Every other kind of ad edit stays human-only.
Unlock rule
Only switches on after the ads watchdog's findings have matched Sherif's own judgment for 4 straight weeks — audits earn their place before any edit does.
Never touches bids, budgets, audiences, or live ads. Approval stays batch-by-batch — it never runs on its own.
Marketing — B2B pipeline · Hospitality, OEM & Construction
reports to the B2B marketing owner — to be decided
Three of our four business units sell B2B — and none has any paid marketing motion today. These loops are designs on the shelf, built on a proven public playbook (ColdIQ's ads-skills, run at $300K/month of managed spend — bookmarked for when we build, its B2B machinery transfers to us almost as-is). They switch on when two decisions are made: a B2B marketing budget exists, and someone owns it. The one free step worth taking now: requesting LinkedIn's advertising API access, because that approval is the slow part. The bigger picture — signal-driven outbound, the market moment, and the phased plan — has its own page: Automated go-to-market →
B2B-L4Outbound assistant — the signal engine, human-sent
Needs prepPhase 1
Does
The full go-to-market engine: watches project databases for buying signals (a hotel entering early planning, a tender award), finds and verifies the decision-maker, quality-gates the record, drafts a message referencing the specific project and stage, and hands the ready-to-send package to Ifham or Yasar — who send it themselves. On a reply, the contact is proposed into the CRM with its source stamped.
Why human-send
Every LinkedIn sending tool violates LinkedIn's terms (accounts get banned — a major vendor's own page was banned in March 2026). Software watches, finds, and drafts; a person presses send. Zero ban risk — and it's how the outreach stays personal.
Also reports
Weekly outreach health for the humans: acceptance rate (pause below 30%), reply rate, stale invites to withdraw.
Never sends anything, never adds a CRM record without a named person's approval, never imports raw lists.
B2B-L1LinkedIn account watchdog & audit
Needs plumbingOn decision
Watches
A future LinkedIn ads account, weekly: spend pacing, audiences that spend without converting, bids leaving efficiency on the table — and LinkedIn's unique data: which job titles and company sizes actually convert (hotel procurement heads for Hospitality, buying-office roles for OEM).
Baseline
On day one it runs a 35-point account audit from the playbook and ranks the fixes by impact — so the account starts healthy instead of drifting into the usual LinkedIn waste (the most expensive clicks in advertising).
Speaks
Weekly to the B2B marketing owner + the relevant BU lead.
Waits for
Budget + owner decision · a LinkedIn ads account · LinkedIn API approval (days — request now, costs nothing).
Never edits campaigns, bids, or budgets — read-only by construction, exactly like the Huxberry ads watchdog.
B2B-L3Lead-form capture — leads with receipts
Needs plumbingOn decision
Does
When LinkedIn lead forms go live: each new lead is qualified against our ideal-customer profile and proposed into the CRM one at a time, with its source and campaign stamped on the record from the moment it's created.
Why that stamp matters
Today 100% of our CRM deals have no source recorded — we literally cannot say which marketing produced which deal. B2B leads born through this loop are the first that arrive with receipts attached.
Approval
The BU lead approves every record before it enters the CRM. Shares its qualification machinery with the hospitality lead-intelligence loop (CRM-3).
Never contacts a lead, never creates a CRM record without a named person's approval.
B2B-L2Audience builder from our own CRM
Needs plumbingOn decision · last
Does
Builds precise advertising audiences from lists we already own — e.g. GCC hotel chains for Hospitality account-based campaigns, big-box buyer roles for OEM — and uploads them to LinkedIn (and Meta where relevant) as matched audiences.
Safeguards
This is both an edit and a data export, so it goes last in the sequence: every upload approved batch-by-batch after the list contents are reviewed; business contacts only, never consumer data; a UAE data-privacy (PDPL) check with Rajiv before the very first upload.
Waits for
The CRM hygiene loop first — audiences built from a dirty CRM aim at the wrong people. Then B2B-L1 proving itself.
Never uploads a list that hasn't been human-reviewed, and never touches consumer/customer personal data.
IT & governance
reports to Nish
IT-1Company-knowledge hygiene
ReadyWave 2
Watches
The company wikis and the AI "brain" that answers from them — broken links, contradictions, stale pages, and whether recent edits actually made it into the brain's index.
Speaks
Weekly summary of what it fixed and what needs a human.
IT-2Workspace security posture
Needs prepWave 4
Watches
Google Workspace safety basics across all 45 accounts: two-step verification coverage, admin access, third-party app permissions, leavers' accounts properly closed.
Speaks
Monthly, deltas only — what changed since last month.
IT-3The loop-of-loops
ReadyWhen ≥4 live
Watches
The watchmen themselves: which loops ran, what they cost, which are stalled, noisy, or no longer earning their keep.
Speaks
Weekly scorecard. Killing a useless loop is a success, not a failure.
07 — Rollout
Four waves. Every loop pilots quietly for two weeks before anyone relies on it.
Wave 1 · now
Fix the pipes, start the first watchmen
The broken warehouse sync is already repaired. Next: the pipeline sentinel (DATA-0), the CRM hygiene watchdog (CRM-1 — zero blockers), and the dead-stock watchdog (INV-1). Marketing tracking (HUX-L0) starts in parallel.
Wave 2 · weeks 2–3
Money and knowledge
Retention tracker, bank compliance calendar, protector radar, knowledge hygiene — plus the first Huxberry creative and search-watch loops. Wave-1 pilots get promoted to always-on.
Wave 3 · month 2
Connect the missing data
One data extension unlocks the receivables watchdog and the stock-settings builder (which doubles as Odoo migration homework). Pipeline pulse and hotel lead-intelligence go live.
Wave 4 · month 3
Complete the picture
Reorder-risk, supplier intelligence, ad-spend reconciliation, security posture, the proactive creative-refresh — and the loop-of-loops reviewing them all weekly.
Beyond · gated on conditions, not dates
Waves 5–7, saved for when their doors open
Wave 5 — the B2B go-to-market ladder (starts the day it has an owner and a budget; can run alongside waves 2–4). Wave 6 — loops the Odoo migration unlocks: email→lead and WhatsApp→lead capture, customer feedback collection, per-BU profit reporting, and the big one — automating the ~979 quotation spreadsheets. Wave 7 — deepening what's proven: earned write permissions, event-driven triggers, faster approvals. After that, new loops come from what the loop-of-loops observes — not from a master plan. Details: the future backlog.
08 — Questions people actually ask
Fair questions, straight answers.
Will a loop ever email a customer or supplier?
No. Loops draft text for us; a person always sends it. The same goes for changing ERP/CRM data, publishing content, or spending money — every action waits for a named person to tap Approve.
What if a loop gets something wrong?
It will, occasionally — usually by flagging something that turns out to be fine. That's why loops only report and suggest. Tell the loop's owner and the threshold gets tuned; each digest also states exactly how fresh its data is, so you can judge it.
Is this watching me?
No. Loops watch numbers and records — stock, invoices, deals, deadlines — not people. Team-level figures go to that team's own lead, not broadcast. There are no individual performance leagues here, and we've deliberately kept HR and payroll out of scope.
Do I need to learn a new tool?
No. To begin with, digests arrive as ordinary email — nothing to install, nothing to learn. As the company feed finishes rolling out, weekly summaries will appear there too. Your only "controls" are replying to a message and telling us what to change.
What does it cost?
Mostly subscriptions we already pay for; routine checks run on our cheapest AI capacity and the smarter models are reserved for judgment. Every loop's cost is reviewed in its first week and weekly thereafter — a loop that doesn't earn its keep gets switched off.
Is this replacing anyone's job?
No — none of this work is anyone's job today, which is exactly the problem. The dead stock, the unfollowed deals, the untracked retention money: they slipped because everyone already has a full-time role. Loops do the watching; people keep the judgment, relationships, and decisions.