The GCC System Trends That Will Actually Change Your Company, and the Ones That Will Not

A managing director in Sharjah forwarded me a consultancy deck in the spring. Seven trends, one slide each, each with a diagram and a verb. Composable architecture. Agentic AI. Embedded finance. Sustainability reporting. Hyperautomation. Industry clouds. Data mesh.

He had one question, which was the right one: "Which of these changes what my production planner does next month?"

The answer was one and a half of the seven, and neither of the halves was on the slide with the nicest diagram.

That is the test worth applying to everything in this article and everything else you will be sent. A trend matters to your company if it changes what somebody types, signs, checks or is prevented from doing. If it only changes an architecture diagram, it may still be true, and it is not yet your problem. A 200-person joinery works in Al Quoz and a 600-person cable manufacturer in KIZAD do not have a technology strategy problem. They have a small number of specific, dated obligations and a finite amount of attention.

Here is the sorting, with the reasoning attached.

The short version

Force	What it actually is	Does it change a mid-market GCC system by 2028	What to do about it
Regulatory digitisation	E-invoicing, corporate tax, structured filings, machine-readable records	Yes, and on a published calendar	Treat as a dated project with an owner, not a compliance line item
Configuration replacing customisation	Vendors shipping as standard what used to be built	Yes, quietly and to your benefit	Re-check your customisation list at every version
AI in narrow, checkable tasks	Extraction, drafting, matching proposals, anomaly flags	Yes, in a smaller way than advertised	Start where failure is loud and volume is countable
Arabic-first and bilingual operations	Documents, master data, interfaces and reports that work in both languages	Yes, and it is consistently underestimated	Design it in, do not translate it on afterwards
Payment rails changing underneath	Instant domestic payment schemes and national card infrastructure	Partly, and mostly as a reconciliation change	Watch your bank's roadmap, not the fintech press
Partner market consolidation	Implementation firms merging, being acquired, or quietly shrinking	Yes, for your support contract specifically	Contract for continuity, not for a named individual
Composable and MACH architecture	Best-of-breed components assembled over APIs	Rarely, at this size	Ignore the label; the underlying integration question is old
Sustainability and ESG reporting	Structured environmental and governance disclosure	Only if a customer or lender demands it	Wait for the demand, then treat it as a data project
Industry clouds, data mesh, digital twins	Packaging and architecture vocabulary	No	Nothing

The rest of this piece explains the rows where the reasoning is not obvious.

Regulation is the only trend with a calendar

Every other force in the list is optional in the sense that a determined company can defer it for two years and survive. Regulatory digitisation is not, and it is the single reason most GCC mid-market systems will change between now and 2028.

The shape is consistent across the region: tax authorities are moving from receiving summaries to receiving transactions. That is a different demand on your system entirely. A summary can be assembled by a competent accountant from imperfect records. A transaction stream cannot.

In the UAE, the e-invoicing regime is now on a published timetable. The pilot and voluntary phase opened on 1 July 2026. Businesses with annual revenue of AED 50 million or more have until 30 October 2026 to appoint an accredited service provider — a date that was extended from an earlier July deadline — and must be issuing compliant invoices from 1 January 2027. Smaller businesses follow, with government entities later still. The Ministry of Finance published an updated version of its electronic invoicing guidelines in June 2026. The mechanics of what this asks of your system, corner by corner, are set out in what UAE e-invoicing actually requires of your ERP, and the appointment decision itself deserves more scrutiny than it usually gets.

Saudi Arabia is further along and demonstrates where this ends. ZATCA's integration phase has been rolling out in waves since 2023, and the threshold has descended steadily: wave 24 brought in taxpayers whose VAT-subject revenue exceeded SAR 375,000 in 2022, 2023 or 2024, with an integration deadline of 30 June 2026. That is not a large-enterprise threshold. It is nearly everybody.

Oman is the third. The Tax Authority was approved as a Peppol Authority in January 2026, adopting Peppol as the framework behind its Fawtara system, with the first mandatory wave — a named set of large taxpayers — from August 2026 and further waves through to 2028.

Alongside this sits corporate tax, which is no longer new but is now real in a way it was not in 2024. A UAE business with a 31 December 2025 year end had to file and pay by 30 September 2026, through EmaraTax, at 0 per cent up to AED 375,000 of taxable income and 9 per cent above it. Small Business Relief, the election available to resident businesses at or below AED 3 million of revenue, is currently scheduled to end for tax periods after 31 December 2026 unless extended. For a company that has been leaning on that election, the 2027 close is a different exercise from the 2026 one, and the system needs to support it.

Put those together and the practical consequence for a mid-market company is not "buy a compliance module". It is four uncomfortable requirements that touch the core of the system.

Your master data has to be correct at transaction level, because it is being transmitted rather than summarised. Your invoice has to carry structured, validated fields at the moment of issue rather than at the moment of filing. Your archive has to be retrievable in the transmitted form, not just as a PDF someone can reprint. And your close has to produce numbers that reconcile to what was transmitted, because the authority now holds a copy of the detail.

None of that is exciting. All of it is dated. If you are choosing where to spend attention over the next twenty-four months, this row is the answer, and the e-invoicing readiness check is a cheaper starting point than a workshop.

Configuration is eating customisation, and this is good news you have to claim

There is a slow, real shift that nobody puts on a trend slide because it does not sell anything: the products are absorbing what used to be built. Features that were custom modules three versions ago — deferred revenue schedules, structured invoice transmission, production scheduling views, reordering horizons, lot and serial handling — increasingly arrive in the box.

This matters because it changes the arithmetic of your existing customisations, and almost nobody re-runs that arithmetic. A modification built in 2021 to fill a genuine gap may now be duplicating standard behaviour, at a cost you pay at every upgrade, every support call and every time a new person has to understand why your system does something odd.

The discipline is simple and almost never done: at each version, take your customisation register — if you do not have one, that is the first finding — and ask of each item whether the product now does it. Retire what it does. That is the cheapest cost reduction available in an ERP estate, and it is the argument underneath where the line between configuration and custom code actually sits.

The counter-pressure is honest and worth naming. Vendors also use "it is standard now" to justify redesigning something that worked for you. Absorbing a standard feature is not free; it is a process change, a retraining, and a period of people complaining. It is usually still worth it.

AI, with the volume turned down

I have written the longer version of this separately, in what AI agents genuinely do inside business systems and what remains a demo. The short form for a trend list is this.

The capability is real and has shipped into products a mid-market company actually runs. Odoo 19 added configurable agents and natural-language querying over your own records. SAP has been bringing Joule agents to general availability through 2026. Microsoft's 2026 release wave one put autonomous agents into Dynamics 365 Finance and Business Central, including payables and payment agents.

The value that lands in the next two years is narrow and unglamorous: reading supplier documents, drafting text a human sends, proposing matches and codings for a human to release, and flagging drift in prices and payment behaviour. Every one of those has a countable volume and a loud failure mode.

The value that will not land in the next two years, for a company of this size, is anything that commits money or posts to a ledger without a person releasing it. Not because the technology cannot; because your master data cannot carry it and your audit trail is not designed for a non-human actor.

The honest planning position for a 200-person factory is that AI is a productivity line item worth low six figures of attention, not a strategy. Anyone selling it as the latter is selling you a re-implementation you did not need.

Arabic is an operating requirement, not a translation task

This is the row most consistently underrated by companies headquartered elsewhere and most consistently underestimated by implementation teams, including experienced ones.

Bilingual operation is not a language pack. It is a set of design decisions that reach into the data model: which fields carry both languages and which carry one, what appears on a customer-facing document versus an internal screen, how a name is stored so that it can be searched in either script, and what your tax authority expects on the face of an invoice.

Then there is the category of defect that no server-side test catches. Right-to-left rendering reverses the visual order of "number slash number" — a ratio of four over eight can present to the reader as eight over four while the stored value is entirely correct. Reports laid out for one direction break in the other. Sort orders do odd things. A system that passes every test in English and is signed off by an English-speaking steering committee can be quietly unusable for the people who actually run the warehouse.

The remedy is not clever. It is that somebody walks the screens in Arabic, with expected figures worked out in advance, before go-live rather than after. Budget for it as a workstream, not as a checkbox.

The payment rails are changing, and it reaches you as reconciliation

Instant domestic payment infrastructure is genuinely being built. In the UAE, the Aani instant payment platform and the Jaywan domestic card scheme have moved beyond consumer transfers: in August 2026 the Ministry of Finance adopted both as payment channels for federal service fees and fines, under a 2026 Cabinet Resolution, and described it as paving the way for other federal entities and collection banks to follow.

For a mid-market company this is worth understanding but not worth a programme. B2B settlement moves slowly, and your customers' finance departments will not abandon their existing habits because a new rail exists. What does reach you, and sooner than you expect, is the shape of the remittance data. Faster and smaller payments mean more of them, arriving with different references, which lands squarely on whoever does your cash allocation. If your receipts are already allocated by a person reading a bank statement PDF, more payment channels makes that job worse, not better.

The practical move is not to redesign anything. It is to ask your bank what formats and identifiers they will support over the next two years, and to make sure your system can consume a structured statement rather than a printout. That is a small piece of work with a real return, and it is the sort of thing that gets skipped because it is nobody's exciting project.

The partner market is consolidating, and your support contract is the exposure

This one is not a technology trend at all, and it is the one most likely to actually hurt you.

The regional implementation market has been through a period of expansion, and periods of expansion end. Firms merge, get acquired by larger groups, are absorbed into consultancies with different economics, or quietly lose the two people who knew your system. From the buyer's chair, all four look identical: the person who answers your ticket changes, and the new person does not know why your system does the odd thing it does in March.

Guard against it in the contract and in the documentation, not in the relationship. Specifically: source code and configuration documentation held by you rather than described to you, a written handover obligation, a defined response commitment that survives a change of ownership, and the ability to take your database and your custom modules elsewhere. This is not distrust. It is that the median tenure of a good consultant is shorter than the life of your ERP, and the considerations that govern choosing a partner apply with more force to keeping one.

What to ignore, and why

Composable, MACH, best-of-breed assembled over APIs. The idea is sound at large scale. At mid-market scale in the GCC it usually means running five products and owning the integration between them, which is a permanent engineering obligation you do not have the staff for. The interesting question underneath — which capabilities genuinely deserve a separate system — is old and unglamorous and is worked through properly in the integration paradox. The label adds nothing.

ESG and sustainability reporting, until a customer or a lender asks. Odoo 19 shipped an ESG application; other vendors have equivalents. If you supply into a European customer or borrow from a bank with disclosure requirements, this becomes a real data project on a real deadline. If neither is true, building a carbon reporting capability now is a hobby.

Industry clouds, data mesh, digital twins. For a 200-person factory these are vocabulary. There is a real version of the digital twin idea in heavy process manufacturing at a scale where a sensor estate already exists. If you do not already have the sensors and the historian, you are being sold the picture rather than the thing.

"Rip out and rebuild with AI." This appears in every conference season under a new name. The cost is never the software; it is the years of organisational learning encoded in your configuration and your workarounds, which no model reconstructs and no migration carries.

What to actually do with the next twenty-four months

Take the regulatory dates that apply to you and put them on a plan with a named owner and a budget, because they are the only items here that will not wait. Re-run your customisation register against what the product now does as standard, and retire what has become redundant. Pick one or two narrow AI use cases where the volume is countable and a wrong answer announces itself, and measure the correction rate rather than the vendor's accuracy claim. Get somebody to walk your system in Arabic with expected numbers in hand. Ask your bank what is changing in statement formats. And read your support contract as though your partner were about to be acquired, because one day it will be.

That is a two-year programme that fits inside a mid-market company's actual capacity. Everything else in the deck can wait for the version of it that arrives with a date attached.

If the harder question is whether the system you are running can carry any of this at all, that judgement is worth making deliberately rather than by default, and it is the first thing we do on a digital transformation engagement.