A founder asked me this last month over coffee. They'd decided AI was a priority. Budget was allocated. The question was: do we hire a full-time AI/ML engineer, or do we bring in a consultant to get us started?
My answer was "it depends" which he rightly told me was a shit answer. So here's the real answer.
The hiring reality right now
Good AI engineers are expensive and hard to find. In the UK, you're looking at £90k-£140k for someone with genuine production experience. Not someone who did a Coursera course and put "AI/ML" on their LinkedIn. Someone who's actually built and deployed models or agents that real people use.
The hiring timeline is brutal too. I've spoken to three companies in the last two months who've had an AI engineer role open for 4+ months with no decent candidates. The talent pool is small and everyone's fishing in it.
And here's the thing nobody talks about: even when you hire someone good, they need 2-3 months to understand your business, your data, your systems, and your team before they're producing real value. That's not a criticism of them. That's just how it works. Domain knowledge takes time.
So the realistic timeline from "let's hire an AI engineer" to "we've shipped something useful" is 6-9 months. Sometimes longer.
When hiring makes sense
Hire when AI is going to be a continuous, core part of your product or operations for the foreseeable future. When you need someone embedded in the team, building institutional knowledge, iterating on models week after week.
If you're a company whose product is AI-powered (a recommendation engine, a prediction tool, a data analysis platform), you need the person in-house. Full stop. You can't outsource your core competency.
Also hire when you've already figured out what to build and you need someone to maintain and improve it over time. The exploration phase is done. You know what works. Now you need sustained engineering effort.
When a consultant makes sense
Bring in a consultant when you're still figuring things out.
When you don't know where AI fits. When you don't know what's possible. When you need someone who's seen this problem at 10 other companies and can shortcut the learning curve.
A good consultant brings pattern recognition. I've run AI readiness assessments for companies across fintech, e-commerce, health tech, and logistics. The problems are different but the patterns repeat. The data access issues. The team resistance. The tendency to start with the hardest use case instead of the easiest one. I've seen all of it before and I can save you the 6 months of figuring it out yourself.
Consultants also make sense for defined projects with a clear end point. Build an agent for this workflow. Assess our AI readiness. Set up our AI infrastructure and hand it to the team. These are time-boxed pieces of work that don't justify a permanent hire.
The cost comparison
Let me just lay out the numbers.
Full-time AI engineer: £120k salary (mid-range UK), plus about 20% employer costs (NI, pension, benefits), so roughly £144k per year. Plus recruiter fee, probably 15-20% of salary, so another £18k-£24k. First-year total cost: approximately £165k. And you won't see production value for 3-6 months.
Consultant engagement: depends massively on scope, but a typical 3-month engagement to assess, prototype, and ship the first AI use case might run £30k-£60k. You'll have something running in production by the end of month 2.
For the cost of hiring one AI engineer and waiting 6 months for results, you could run a consulting engagement, ship 2-3 working AI implementations, and use what you learn to write a much better job description for the eventual hire.
The approach I actually recommend
Start with a consultant. Get your first wins. Figure out what kind of AI work your company actually needs on an ongoing basis. Then hire based on real knowledge instead of guesswork.
I can't tell you how many job descriptions I've read for "AI Engineer" roles that are basically a wish list. "Experience with LLMs, computer vision, NLP, reinforcement learning, MLOps, and data engineering." That's five different specialisms. No one person does all of that well.
After a 3-month consulting engagement, you'll know whether you need an ML engineer, a data engineer, a prompt engineer, or a full-stack developer who's good with APIs. That specificity is the difference between hiring someone who thrives and hiring someone who flounders because the job wasn't what anyone expected.
The hybrid model
Some of the best setups I've seen work like this: a consultant comes in for the first 3-6 months, builds the initial AI capabilities, and overlaps with a newly hired AI engineer for the last month or two. The consultant transfers knowledge, the new hire ramps up with real working systems to learn from instead of a blank slate, and the consultant steps away once the handoff is solid.
It's more expensive in the short term but it massively reduces the risk of the hire failing. The new person isn't walking into an empty room being told to "figure out AI for us." They're walking into a room with working systems, documented decisions, and a clear direction.
This is essentially how our embedded AI leadership works. Come in, build, hand over, leave. Not a retainer. Not a permanent dependency. The goal is always to make myself unnecessary.
If you're sitting on this decision right now, don't overthink it. Get the help, ship something, then figure out the long-term plan from a position of knowledge rather than uncertainty.