How Long Does an AI Project Actually Take?

"How long will this take?" It's the first question every client asks. It's also the question most consultancies answer dishonestly — either too optimistically to win the deal or too vaguely to be held accountable.

Here's our honest framework, based on dozens of delivered projects.

The Four Phases

Every AI project moves through four distinct phases. Each has a predictable range:

Discovery (1-2 weeks). Understand the problem, map the data landscape, define success criteria. This is where you figure out whether AI is actually the right solution — and sometimes it isn't. Skipping discovery is the single most expensive mistake in AI projects.

Prototype (1-2 weeks). Build a working proof of concept that demonstrates the core value proposition. Not a slide deck. Not a mockup. A functional system that processes real inputs and produces real outputs. This is where you validate assumptions and kill bad ideas early.

Production hardening (2-4 weeks). Transform the prototype into something reliable. Error handling, edge cases, monitoring, security, performance optimization. This phase is where most of the actual engineering happens — and where most timelines blow up if the prototype was sloppy.

Integration (1-2 weeks). Connect the system to your existing infrastructure. Authentication, data pipelines, APIs, deployment. The technical work is usually straightforward. The organizational coordination is usually not.

Total: 6-10 weeks for a typical project.

Where the Time Actually Goes

Here's the uncomfortable truth: most "six-month AI projects" are actually two-month projects wrapped in four months of organizational friction.

We've tracked where time goes across our engagements, and the breakdown is consistent:

Actual engineering work: 30-40% of elapsed time. This includes design, implementation, testing, and deployment. AI-augmented development has compressed this dramatically — what used to take months now takes weeks.

Waiting for decisions: 20-30%. Stakeholder reviews, approval chains, priority conflicts. A prototype sits finished for two weeks while the VP is on vacation.

Data access and quality: 15-25%. Getting access to the right data, discovering the data isn't what anyone thought it was, cleaning and transforming it into something usable.

Scope changes: 10-20%. "Can we also add..." is the most expensive sentence in project management.

The engineering is rarely the bottleneck. The organization is the bottleneck.

Our Approach: Time-Boxed Delivery

We structure every engagement around two-week delivery cycles. At the end of every cycle, there's working software that a real user can interact with.

This does three things:

First, it forces prioritization. When you only have two weeks, you build the most important thing first — not the most interesting.

Second, it creates natural checkpoints. If the project is going sideways, you know within two weeks, not six months.

Third, it keeps organizational momentum. Stakeholders stay engaged when they see progress every two weeks. They disengage when they hear "we're still in the design phase" for the third month running.

A Real Timeline

Here's a recent project — an automated document processing pipeline for a legal services firm:

• Week 1-2: Discovery. Mapped 12 document types, identified extraction requirements, defined accuracy thresholds.
• Week 3-4: Prototype. Built end-to-end pipeline for the 3 highest-volume document types. Hit 94% accuracy on extraction.
• Week 5-8: Hardening. Extended to all 12 types, built error handling and human review queue, load tested at 10x expected volume.
• Week 9-10: Integration. Connected to their document management system and case tracking software.

Ten weeks. Working in production. Processing real documents.

What AI Changes (And What It Doesn't)

AI eliminates the easy parts — writing code, generating boilerplate, implementing standard patterns. That's genuinely valuable, and it's why the engineering phases are so compressed.

But AI doesn't eliminate the hard parts. Data quality issues still need human investigation. Integration points still need careful coordination. Change management still requires patience and empathy.

The best thing AI does for project timelines isn't making the coding faster. It's freeing up engineering time to focus on the hard problems that actually determine whether a project succeeds or fails.

When someone asks how long your AI project will take, the honest answer is: the engineering is the easy part. The question is how fast your organization can make decisions, provide access, and commit to a scope.