The filter
Before costing anything, answer two questions. How many hours a week does a person currently spend on this? And does the information needed to do it already exist in a system somebody can query?
If the first answer is under five hours, the automation will not pay back within a year at any realistic build cost. If the second answer is no, you do not have an AI project. You have a data project wearing an AI costume, and it should be priced as one.
What pays back, roughly in order
| Project | UAE market range (AED) | Pays back when |
|---|---|---|
| Document extraction and processing | 15,000–80,000 | Someone re-types data for 8+ hrs/week |
| Quote and proposal generation | 25,000–90,000 | Quoting is slow and volume is lost to it |
| Inbox and ticket triage | 20,000–70,000 | Volume is high and routing is manual |
| Internal knowledge assistant | 80,000–200,000 | Staff hunt for answers across systems daily |
| Customer-facing chatbot | 22,000–150,000 | Support volume is high and repetitive |
What usually does not
- Anything justified by "our competitors have AI". That is a marketing budget, and should be argued as one.
- Assistants over documents nobody maintains. The model will faithfully repeat a policy that was superseded in 2023.
- Predictive models on a few hundred rows of history. The honest answer is a spreadsheet.
- Replacing a process that is broken. Automation makes a bad process faster and much harder to change.
How to buy it
Ask for the hours-saved calculation before the quote, not after. Any provider who cannot state how many hours the process consumes today has not looked at your process, and is quoting a technology rather than a result.