The advice most enterprises get about AI teams goes something like this: hire a data scientist, an ML engineer, a data engineer, a product manager, and a change lead. That is five full-time roles before you've written a line of code. For most New Zealand enterprises, that's not a staffing plan. It's a fantasy.
What You Need to Know
- Most NZ enterprises don't need a full-time AI team. They need AI capability applied to specific problems at specific times. The fractional model matches capability to demand.
- A fractional AI team combines internal domain expertise with external technical capability. The enterprise owns the problems and the outcomes. The fractional team provides the engineering and architecture.
- The model works because AI delivery is inherently cyclical. Discovery is intensive for 2-4 weeks. Build is intensive for 8-16 weeks. Then it's maintenance and iteration, which requires a fraction of the build capacity.
- The risk is not "we can't attract AI talent." The risk is "we hire AI talent with nothing for them to do." A fractional model avoids both problems.
Why Full-Time Doesn't Fit
The traditional enterprise response to a new technology capability is to hire a team. It worked for digital. It worked for cloud. It does not work well for AI, and the reason is structural.
AI capability building is front-loaded. The discovery phase requires concentrated expertise: understanding the problem, evaluating feasibility, designing the architecture, selecting models, building data pipelines. This phase demands experienced people working intensively.
Then you build. This is also intensive, but it's a defined period. A typical enterprise AI capability takes 8 to 16 weeks to build and deploy.
Then you operate. And operating an AI system requires a fraction of the build capacity. Monitoring, tuning, handling edge cases, responding to model drift. Important work, but not full-time-team work.
If you hire a full-time AI team sized for the build phase, they're underutilised during operation. If you size for operation, you can't build. And if you try to keep the team busy by finding new AI projects to fill their time, you end up with AI solutions looking for problems, which is how you get the "we built an AI but nobody uses it" pattern.
60-70%
of AI team capacity is typically underutilised between major build phases in enterprises that maintain full-time teams
Source: RIVER, advisory engagement data, 2023-2024
The Fractional Model
The fractional AI team model matches capability to demand. Here's how we structure it:
What Stays Internal
Domain expertise. Nobody external knows your business like your people do. The claims handlers, the underwriters, the compliance officers, the operations managers. They define the problem, validate the solutions, and drive adoption. This is non-negotiable internal capability.
Strategic ownership. The AI strategy, the prioritisation of use cases, the governance framework. These belong to the enterprise. A fractional model does not mean outsourcing your AI strategy. It means outsourcing the engineering of your AI strategy's execution.
Change management. Adoption is an internal function. The people who drive adoption need organisational relationships, cultural understanding, and the authority to make changes. External teams can support this, but they can't own it.
What Goes Fractional
AI engineering. The people who build data pipelines, engineer prompts, integrate models, and architect AI systems. These skills are scarce, expensive, and not needed full-time. A fractional model gives you access to senior AI engineers for the 10-20 weeks you need them, without carrying the cost for the 32-42 weeks you don't.
Architecture and design. AI system architecture requires experience across multiple deployments. A fractional architect who has built 10 enterprise AI systems brings pattern recognition that a first-time internal hire cannot match. This experience is most valuable during discovery and early build, then tapers.
Specialist skills. ML model evaluation. Security review. Performance optimisation. These are skills you need periodically, not permanently. The fractional model gives you access when you need them.
The Engagement Pattern
A typical fractional engagement follows this rhythm:
Discovery (2-4 weeks, high intensity). Full fractional team engaged. Deep dive into the problem, data, and feasibility. Output: validated use case, architecture design, implementation plan.
Build (8-16 weeks, high intensity). Core fractional team building. Regular collaboration with internal domain experts. Iterative delivery with weekly demos and feedback.
Launch and stabilise (2-4 weeks, medium intensity). Deployment support, monitoring setup, initial tuning. Knowledge transfer to internal team.
Operate and iterate (ongoing, low intensity). Fractional team on retainer for monitoring, tuning, and incremental improvements. Internal team handles day-to-day operations.
Next capability (when ready). Fractional team scales back up for the next build. Faster this time, because the foundation exists.
The question isn't whether you can afford a fractional AI team. For most NZ enterprises, the maths is straightforward.
Tim Hatherley-Greene
Chief Operating Officer
Common Objections
"We'll lose our IP"
This is the most common concern, and it's legitimate. The answer is contractual and architectural. All code, models, and data remain the enterprise's property. The fractional team works on the enterprise's infrastructure, in the enterprise's repositories, under the enterprise's governance. When the engagement ends, everything stays.
"We need people who know our business"
You do. That's why domain expertise stays internal. The fractional model works precisely because it combines deep domain knowledge (internal) with deep technical knowledge (fractional). Neither alone is sufficient.
"What about continuity?"
Fractional doesn't mean disposable. The same core team works with you across multiple capability builds. They accumulate context over time. The retainer model during operation ensures continuity between build phases.
"Isn't this just outsourcing?"
No. Outsourcing is "here's a specification, build it, deliver it." Fractional is "here's a problem, let's solve it together, and your team learns how along the way." The collaboration model is fundamentally different. Internal capability builds over time, reducing fractional dependency.
Who This Works For
The fractional model fits best for enterprises that:
- Have clear business problems that AI can address
- Have domain experts who can partner with technical teams
- Don't have (and shouldn't build) a large permanent AI engineering team
- Value speed to capability over building internal empires
- Are willing to invest in the right engagement model, not just the cheapest
In New Zealand, that describes most enterprises. We're not a market where every company needs a 20-person AI team. We're a market where every company needs access to AI capability. The fractional model bridges that gap.
Getting Started
If you're considering a fractional AI approach:
- Identify your first AI use case. Not the most ambitious one. The one with the clearest data, the most willing users, and the most measurable outcome.
- Designate your internal team. Domain expert, executive sponsor, adoption owner. These people will be the fractional team's partners.
- Start with discovery. A 2-4 week discovery sprint will tell you more about what's feasible than six months of strategy documents.
- Measure outcomes, not activity. The fractional model succeeds when it delivers business outcomes, not when it fills a bench.
The AI talent shortage is real. The fractional model is how most NZ enterprises will navigate it. Not by competing for scarce permanent hires, but by accessing capability when they need it, in the form that actually fits.