What an AI Automation Company Should Put in the Contract: Scope, Support and Handover

By the time an operations manager at a UAE SME is comparing proposals from more than one AI automation company, the sales conversations are over and the differences have narrowed to price and promises. The documents that actually decide whether the engagement works are the ones most people skim: the scope section, the support terms, the ownership clause and the exit clause.

This post is about those documents. It assumes you have already shortlisted vendors; our guide on how to choose an AI automation agency covers the selection checklist and we will not repeat it. Here the question is narrower: once you have picked a company in Dubai, Abu Dhabi or Sharjah, what should the contract say so that you still have a working system, and the ability to change supplier, two years from now?

What an AI automation company typically delivers

It helps to name the services before you negotiate them, because proposals often blur them into one line. A typical engagement includes some or all of the following:

• Discovery and scoping. Mapping the workflow as it runs today, the systems it touches, the volumes, and the exceptions. Output: a written specification.
• Build. Connecting your existing systems (CRM, accounting or ERP, inventory, email, messaging platforms) and adding AI steps where documents or free text need reading, extracting or drafting.
• Testing with your data. Running the workflow in shadow mode on real cases before it touches a customer or a ledger.
• Training and documentation. Teaching the people who will run it and writing down how it works.
• Monitoring and maintenance. Watching for failures, fixing breaks when a connected system changes, adjusting rules and prompts as your business changes.
• Advice. Which systems to keep, which to replace, what to automate next.

Each of these should appear as a separate line in the proposal with a deliverable attached. If "build" and "maintenance" are bundled into one monthly figure, you cannot tell what you are paying for when a workflow is finished.

Engagement models: fixed-scope pilot versus retainer

Most companies offer two models and will happily sell you the second first.

Fixed-scope pilot. One workflow, a written specification, a fixed price, a defined period, a success measure. You pay for an outcome. The risk sits with the company, which is why good ones scope carefully before quoting. For a first engagement this is the model to insist on.

Retainer. A monthly fee for a defined amount of capacity or a defined set of responsibilities. Appropriate once several workflows are live and need monitoring, small changes and occasional new additions. Inappropriate before anything is running, because you are paying for availability with no output to judge it against.

Some proposals mix the two: a pilot price plus a mandatory monitoring retainer from day one. That can be reasonable if the retainer is small and clearly tied to monitoring, but ask what happens to it if the pilot fails. The cleanest structure is pilot first, retainer later, with the retainer scoped against the workflows that actually exist by then.

What the scope section must say

A scope section that reads "automation of lead management and related processes" will cost you money in month three. For each workflow in scope, the contract should state:

1. The trigger. What starts the workflow (an email arriving, a form submission, a status change in the ERP).
2. Inputs and systems. Which systems are read from and written to, by name and version.
3. The steps. In plain language, including which steps use AI and which are rules.
4. The exception rule. What the workflow does when it cannot proceed, and who is notified.
5. The human review point. Where a person checks the output before it has an effect, especially for anything that commits money or contacts a customer.
6. The success measure and baseline. The number you will compare before and after, and when you will compare it.
7. What is out of scope. Named explicitly, so that "can you also make it do X" is a change request, not an argument.

Ask the AI automation company to write this section. How they write it tells you how they will build it.

Support after go-live: the questions to ask

Automations break for ordinary reasons: a connected system changes its interface, a template is rejected by a messaging platform, a supplier starts sending documents in a new layout. What matters is who notices and how fast. The support section should answer:

• Who is the named contact, and who covers when they are away?
• What is the response time for a failure that stops work, and for one that does not?
• Who monitors? If the answer is "you tell us when it breaks", that is not monitoring.
• What happens when a connected system changes? Is adapting to a CRM or ERP update included, or billed as a change?
• What is reported, and how often? A monthly summary of runs, failures and fixes is a reasonable minimum.
• What are the working hours of support, in UAE time, and what happens at a weekend?

Put a number on each. "Prompt response" is not a term.

Ownership: code, prompts, accounts and data

This is the section most often missing from proposals and most expensive to fix later.

Code and configuration. Everything built for you (workflow definitions, scripts, prompts, templates, integration settings) should be owned by your company or licensed to you perpetually, with working copies delivered at each milestone. If the company uses its own platform, ask what you can export and in what form.

Accounts and credentials. Every account created for your workflows (messaging platform, model provider, automation tool, hosting) should be in your company's name with your staff as the admin, and the company's staff added as users. If an account is in the company's name, you do not control the workflow that depends on it.

Data. The contract should state where your data is stored and processed, which third-party providers see it, and what happens to it at the end. The UAE Personal Data Protection Law generally requires consent for processing personal data and sets requirements for cross-border transfers, so the company should be able to show the basis for each location. Where a model provider is involved, ask for its data terms; the business tiers of major providers generally state that customer data is not used for training by default, and OpenAI's enterprise privacy page, for example, says business data is not used to train its models.

Run costs. Usage-based costs (model calls, per-message fees on the WhatsApp Business Platform, hosting) are usually passed through. Ask whether they are billed at cost or marked up, and ask for the official pricing pages so you can check.

Handover and exit

A good AI automation company expects to be replaceable and builds accordingly. The exit clause should require:

• Documentation that a competent third party could follow to run, change and restart each workflow.
• Delivery of all code, configuration, prompts and templates in editable form.
• Confirmation that all accounts are in your name, with admin access verified.
• A recorded handover session with your internal owner as part of any termination.
• A notice period long enough for you to find and brief a replacement.
• Deletion of your data from the company's environments, confirmed in writing.

Then test it. Our automation provider exit drill checklist describes a yearly rehearsal in which someone who did not build the workflow restarts it from the documents alone. If the drill fails, the handover clause is worth less than the paper.

As an illustrative example, a Sharjah distributor could specify an order-intake pilot around one shared inbox and an approved product catalogue. The agreement would name who checks ambiguous quantities and who authorises an ERP entry. Handover would include the account list, exception queue and instructions for stopping the workflow. The acceptance test would include a replacement operator following those instructions without the original builder.

How to decide: a contract checklist

Before you sign, confirm each line is true:

1. Each workflow in scope has a trigger, systems, steps, exception rule, review point and measure written down.
2. The first engagement is a fixed-scope pilot with a fixed price.
3. Support has a named contact, response times and a monitoring arrangement.
4. You own the code, configuration and prompts, or hold a perpetual licence with working copies.
5. Every account is in your company's name.
6. Data locations and providers are listed, with the PDPL basis for each.
7. Run costs are itemised and billed at cost or with a stated margin.
8. The exit clause requires documentation, delivery, account confirmation and a handover session.

Two gaps can be negotiated. More than that, and the AI automation company is asking you to trust rather than verify.

Next step

Take the proposal you have in front of you and mark each of the eight lines above as present, missing or vague. Send the marked copy back and ask for a revision before you discuss price. Our Dubai AI automation agency page describes how we scope and hand over, and you can talk to us about your workflow if you would like a second pair of eyes on a contract.