A company that has decided it needs an AI Officer usually cannot say what the seat produces. That is not a failure of the company. It is a failure of everyone selling into it. The category is eighteen months old, the title arrived from government before it arrived from industry, and the market filled the gap with readiness scans.
Every readiness scan comes back saying yes. That is what a scan is for. It is the first invoice in a sequence, and a scan that recommended stopping would end the sequence.
What follows is the alternative: four engagement shapes, what each one puts on the table, and the published band each trades at in this market. A leader on the Atlas register may run all four or one of them. The shapes are what a buyer should be able to name before the first call.
One. The decision audit
Two weeks. The published UAE band for a two-week AI audit runs AED 45,000 to 75,000.
The audit does not score the company against a maturity model. It measures the cycle time of the ten to fifteen decisions that set the pace of the business. Measured, not estimated. How long from a customer complaint to a resolution owner. How long from a supplier quote to a purchase order. How long from a month-end close to a number the board can act on.
Against that map, three outputs.
A ranked list of candidate use cases, scored on data readiness rather than on enthusiasm. Data readiness is the largest single driver of AI failure and it is normally discovered last, after the engineering money has been spent.
The kill list. The pilots already running and the vendor proposals already on the desk that will not reach production, named in writing, each with the reason. This is the deliverable that separates an audit from a scan.
One workflow, selected, with a measured baseline attached to it. That baseline is what the next engagement is judged against, which is why the sequence only works in this order.
The audit ends in a one-page board memo: what to start, what to stop, the expected return, and the confidence level attached to it.
Two. The production sprint
Sixty days, one workflow. Published UAE bands for a scoped implementation of three to six months run AED 100,000 to 250,000, and a sixty-day single-workflow sprint sits in the lower half of that.
The workflow goes live inside the system of record. The ERP, the CRM, the core banking platform, the practice management system, whatever the company actually runs on. Not a standalone tool sitting beside it. MIT's NANDA study found generic tools pilot widely and cross into production rarely, while tools embedded in one specific workflow cross over.
The sprint delivers the workflow live, a before and after measurement against the audit baseline, a runbook, a named owner inside the client who has been trained on it, and the failure modes written down.
Payment structured in three tranches, with the last one contingent on the measured outcome, is the arrangement that aligns the two sides. It also requires a baseline both parties signed before the work started, which is the dependency that makes the audit non-optional.
Three. The embedded operator
Two to three days a week, six-month minimum. UAE monthly retainers for this seat run AED 40,000 to 120,000.
A sprint that ships leaves the company with one workflow in production and nobody who owns it. The embedded operator runs the next workflow, owns the metric, sits in the leadership meeting and reports to the board. The title travels: Head of AI, Chief AI Officer, Director of Automation.
The argument for the part-time version is arithmetic. A full-time technology executive at this level costs AED 61,000 to 92,000 a month in base salary before housing, transport, bonus, visa and end-of-service gratuity, and the search itself takes months. 90% of GCC organisations reported skills gaps in 2025. The seat is real. The hire, at most mid-market companies, is not yet affordable.
Four. The AI register
Three weeks.
DIFC Regulation 10 has been in force since January 2026 and the federal PDPL applies from 1 January 2027. Both require the record before they require anything else, and most companies do not have it.
The register is an inventory of every model, tool, agent and shadow-AI use in the business, each one named, owned and classified by risk. With it: a map of which entity sits onshore, in DIFC or in ADGM and which obligations attach to each; impact assessments for the high-risk use cases in the form Regulation 10 asks for; an internal AI usage policy written to be adopted rather than filed; and a human-oversight mechanism for automated decisions affecting individuals.
One scope line belongs on every version of this engagement. It is assessment and documentation. The company's counsel reviews it. A leader who describes this work as a legal opinion is describing work they are not licensed to do, and the licensing question sits underneath every AI engagement written in this market.
The order is the argument
The four shapes sell in sequence because each one produces the input the next one needs. The audit produces the baseline the sprint is measured against. The sprint produces the workflow the operator inherits. The register runs alongside all three and it is the one a regulated entity should buy first.
A buyer reading a proposal that starts at the sprint, with no audit and no baseline, is reading a proposal that cannot be held to a number later. That is worth saying out loud on the first call.
Where this fits on Atlas
Atlas does not deliver these engagements. Atlas publishes the leaders who do, by name, with their rate band on their own page, and the standard every verified badge is checked against. Reading the register is free. Listing is free. A company pays one month of the agreed retainer, once, if an engagement starts.
If you can name the seat and cannot name a person to fill it, brief a search. If you run these engagements and want the page, list.