25%
of respondents have moved 40% or more of their AI pilots into production.
Deloitte, 21 January 2026Survey of 3,235 leaders in 24 countries, August to September 2025
Resources
A guide for business buyers choosing a first AI project, published by REIT Limited, a Bangladesh-based AI services and technical-delivery company. It explains what makes a project suitable and what a scoped engagement should deliver.
The short answer
A suitable first AI project is one repeated process with stable inputs, a clear owner and a result that can be checked. A scoped engagement should give you a written scope, acceptance checks agreed in advance, evidence from a pilot, and a clear statement of who operates the system afterwards.
Published surveys point to the same causes: pilots that never reach production, data that is not ready, and agent projects started without clear value or controls. The seven steps of this guide address each one.
25%
of respondents have moved 40% or more of their AI pilots into production.
Deloitte, 21 January 2026Survey of 3,235 leaders in 24 countries, August to September 2025
63%
of organisations either do not have, or are unsure whether they have, the right data management practices for AI.
Gartner, 26 February 2025Survey of 1,203 data management leaders, July 2024
40%+
of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls, Gartner predicts.
Gartner, 25 June 2025Analyst prediction, not a measured result
Not every problem needs an AI agent. Work through the options in order and stop at the first that solves the problem.
| Option | Choose it when | Related page |
|---|---|---|
| Improve the process | Steps are unclear or unnecessary. Fix the process before automating it. | How We Work |
| Configure existing software | A feature in a tool you own already does the job. | None |
| Integrate systems | The need is to move data reliably between two systems. | AI & Digital Orchestration |
| Automate a workflow | The steps are known and repeat. | AI Workflow Automation |
| Build an agent | The task must choose among permitted tools case by case. | Agentic AI Development |
| Orchestrate | Several of these parts must work as one process. | AI & Digital Orchestration |
Most projects depend on connecting to systems you already use. Before a pilot, establish for each system:
A provider's claims fall into four kinds: a company fact, a proposed scope, an illustrative example and a measured result. Ask which kind each statement is. A measured result should come with the period, the baseline, the method and the client's permission to share it. Where a provider has no published case for your type of project, a labelled demonstration and a pilot on your own data are the appropriate evidence.
An acceptance check is a written, measurable condition. Examples:
Checks are agreed at design, tested at evaluation, and decide whether the work goes to production.
Decide who will run the system after launch: your team, the provider, or both. The proposal should name the operating owner, the person who approves changes, and the support channel. It should also state who owns and can export the custom code, workflows, prompts and documentation, and what third-party licences apply.
| Stage | What you are paying for | Type of cost |
|---|---|---|
| Assessment | A decision on feasibility and a recommendation | One-off |
| Pilot | One bounded use case, tested against agreed checks | One-off, plus tool usage during the pilot |
| Launch | The production build, release and handover | One-off |
| Maintenance | Monitoring, incidents and controlled changes | Recurring |
| Third-party usage | Model, automation-tool and hosting charges from vendors | Recurring, billed by the vendor |
Keep implementation and recurring costs separate when comparing proposals. Time released by automation is not automatically money saved; count a saving only where the released time is put to a use you can name. For REIT Limited's own stages and indicative prices, see How We Work.
Take the simplest option that works. Fix an unclear process first. Use a feature in software you already own if one exists. Integrate when the need is only to move data. Automate a workflow when steps are known. Build an agent only when the task needs case-by-case choices among permitted tools. Compare access, review effort, exceptions and running cost before deciding.
Ask for a labelled demonstration, a written scope and acceptance checks, and a pilot on your own data with a go or no-go decision at the end. These show how the provider works and limit your commitment. Treat general claims of results without a method and baseline as unproven.
Use the checklist to describe one workflow, then request an assessment.