How Long Does an AI Project Take? From 2 Hours to 8 Months
A small, well-defined AI job usually finishes in 2 hours to 2 weeks. In a corporate project, Gartner’s data puts the move from prototype to production at about eight months on average. The gap comes less from how complex the AI is and more from scope, how many systems are touched, and how fast approvals arrive. Below I explain which job takes how long, what stretches the timeline, and how I give an estimate.
Which job takes how long?
These durations come from the small and mid-sized jobs I have completed; they are not an industry measurement.
| Job type | Example | Time | Why |
|---|---|---|---|
| One channel, one task | Connecting a Q&A bot to a WhatsApp line; an auto-reply on an inbox; a reporting flow for a single account | 2-4 hours | One system, one integration, testing included |
| A few systems talking | Matching a CRM, calendar and email; a custom assistant fed by company documents; one report from several sources | 1-3 days | Each connection needs its own test and access |
| Scale grows the job | Reporting and automation across many accounts or branches | 1-2 weeks | The technical work is the same, but access, tests and exceptions multiply per account |
The work does not grow linearly with how much AI is in it; it grows with the number of systems and people involved.
What really stretches the timeline: waiting
The technical part is rarely the bottleneck. When a job jumps from 2 days to 2 weeks, it is usually one of these:
- Waiting for approval. If an email is answered three days later, a one-hour job spreads over a week on the calendar. That is the speed of communication, not the length of the work.
- Access. Getting read permission on a system can take longer than building the model on top of it. In organizations this is often the longest step.
- Data quality. If name fields in a CRM are inconsistent or dates in an Excel file are mixed, the data must be cleaned before the automation. That step usually takes longer than expected.
- Integration surprises. A system’s API may not behave as documented. These are unpredictable but not rare.
I state all four risks up front when I give an estimate.
How long does a corporate AI project take?
According to Gartner’s survey published in May 2024 (run in the last quarter of 2023 among 644 organizations in the US, Germany and the UK; CIO Dive summary, Gartner press release), moving from prototype to production takes about eight months on average, and fewer than half of AI projects reach production. This data is almost three years old, so read it as an order of magnitude, not a current average.
There is no single “right duration” here. In the long projects I have seen, what they share is multi-stakeholder approval, security and data-access reviews, and a success measure that was not defined at the start. That is my observation, not a number.
How long does it take to get an AI project approved internally?
There is no general figure; it depends on the organization’s approval chain. I can tell you what shortens it (again my experience, not measured data): one accountable person, a scope that fits on one page, which data will be used and whose access permission it needs, how success will be measured, and what happens in the worst case. If these are ready, the approval conversation gets shorter.
How do I give an estimate?
First we clarify the scope together: which systems are involved, who approves, what data is used. Then I give a range such as “4-6 hours” or “5-8 working days”. Because billing is hourly, the estimate is not a ceiling but a realistic range; you are billed for however long it takes me to finish. For how the rate is set, see AI consulting pricing.
Jobs that do not fit this model
Large integrations that must connect many systems at once, projects that need a model trained from scratch, and very large data migrations do not fit this range; they need a different way of working and a different timeline. If the scope turns out to be large in the call, I say so up front rather than understate the time and stretch it later.
You can bring this work to me
The scope and timeline logic above describes the work I do: fully remote, hourly, usually between 2 hours and 2 weeks. Describe your scope and I will give a time and cost range. Details are on the AI consulting page.
Frequently Asked Questions
Does a 2-hour job really finish in 2 hours?
If the scope is truly narrow and single-system, yes; connecting a predefined bot to a WhatsApp line can take 2-4 hours including testing. If the Q&A content is not ready, preparing it is a separate job and is not part of that time.
Which type of job takes the longest?
Multi-account or multi-branch structures. The technical work is not large, but each account means separate access, separate testing and separate exceptions.
What if my job looks like it will take more than 2 weeks?
I say so clearly in the first call and propose a different working plan.