Nobody on our team understands AI: the real structure of the problem and three ways to solve it
When the leadership of a small or mid-sized business says “nobody here understands AI,” the reflexive answer is to hire someone who does. That reflex is usually wrong. The capability people call “understanding AI” is actually three distinct layers — using the tools, applying them to your business, and selecting the right technology — and trying to cover all three with a single full-time hire is the most expensive and failure-prone option available to an SMB. In most cases, a combination of upskilling internal staff and engaging external embedded support is more realistic on every axis that matters: cost, time to impact, and whether the capability actually stays in the company. This article breaks the problem down, compares the three solution patterns, and offers a decision flow for choosing among them.
What “understanding AI” actually means: three layers
Job postings that ask for someone who “understands AI” conflate skills that behave very differently in the labor market and in your organization. Separating them is the prerequisite for any sensible plan.
Layer 1: Using the tools
This is fluency with generative AI assistants — ChatGPT, Claude, and their peers — including writing effective prompts (the instructions you give the model). It requires no special background. Any employee with real work experience can reach competence in weeks to a few months of practice. Hiring a specialist to cover this layer alone is over-investment; it is a training problem, not a recruiting problem.
Layer 2: Applying AI to your business
This is the ability to look at your own workflows, identify where AI produces real leverage, and embed it into how work actually gets done. The critical observation: more than half of the required knowledge here is knowledge of your business, not knowledge of AI. How quotes get assembled, the unwritten rules of customer communication, which process changes the front line will quietly sabotage — an outside hire arrives with none of this. Your existing operations people hold the scarce asset in this layer.
Layer 3: Selecting the technology
This is the ability to compare tools and architectures and decide what to adopt — weighing cost, data security, vendor lock-in, and integration with existing systems. It is the most technically demanding layer, and the one where genuine specialist judgment matters. But note the frequency: in a typical SMB, decisions at this layer arise a handful of times per quarter, not daily. That gap between required expertise and required hours is the central reason a full-time specialist hire so often fails to pay off.
The implication is direct: layer 1 is solved by broad training, layer 2 is best owned by insiders who know the business, and a full-time hire is justified only when layer 3 decisions arrive frequently enough to fill a job.
Comparing the three solution patterns
With the layers in view, here is how the options compare. Cost figures vary widely by market and company size, so treat them as rough orders of magnitude rather than benchmarks.
| Axis | Hiring a specialist | Upskilling internal staff | External embedded support |
|---|---|---|---|
| Cost (rough guide) | Full specialist salary plus recruiting costs | Training fees and learning time | A monthly retainer, typically well below one full-time salary |
| Time to impact | Slow — the search alone can take six months or more | Slow — six months to a year before results | Fast — work can start as soon as the engagement does |
| Retention of capability | Leaves when the person leaves | Best — stays in the company | Stays only if handover is designed in |
| Layers covered | Layers 2–3, if the hire works out | Layers 1–2 | Layers 2–3 |
| Main risks | Mismatch, early departure, not enough work | Slow, ceiling of self-study | Dependency, outsourcing without learning |
Why hiring is the hardest option for an SMB
Hiring looks like the “real” fix, but three structural disadvantages compound against smaller companies.
First, competition. People with genuine layer-3 skills are being courted by large enterprises and well-funded startups. An SMB with no AI track record starts the negotiation behind on both compensation and the intrinsic appeal of the work.
Second, the evaluation loop. You want to hire because nobody in-house understands AI — which means nobody in-house can evaluate whether a candidate actually does. Interview claims go untested, and the mismatch surfaces only after the start date, when it is expensive.
Third, workload. As noted above, layer-3 decisions are intermittent in most SMBs. A full-time specialist who ends up spending most of the week on layer-1 helpdesk questions will worry about their own career trajectory and leave early, taking the recruiting investment with them. “Is there truly a full-time job’s worth of specialist work here?” is the question to answer before opening a requisition. The structure closely parallels the debate about when a startup should hire a CTO, which we examine in when startups should hire a CTO.
Where upskilling wins, and where it hits a ceiling
Internal upskilling dominates at layer 2. Teaching AI to someone who knows the business is usually faster than teaching the business to someone who knows AI, and nothing beats it for retention. The limits are equally clear: self-study rarely reaches layer 3, because choosing what to learn is itself a specialist decision, and results take six months to a year. Betting the company’s entire AI timeline on internal learning alone is a real risk.
Where embedded support fits
Embedded support is a model in which an external expert works as a continuing member of your team — executing alongside your staff and building their capability — rather than delivering a report and leaving. It lets you buy layer-3 judgment only when and in the quantity you need it, sidestepping the full-time workload problem entirely, and it starts fast. A monthly engagement such as embedded AI support for SMBs can cover layers 2–3 at a fraction of the cost of a specialist hire. The failure mode is dependency: if everything is delegated and nothing is transferred, adoption stops the day the contract ends. A useful screening question for any provider is whether “handover to your internal team” appears in the engagement as an explicit deliverable.
A decision flow for choosing your pattern
The starting question is not “can we afford a hire?” but two others: does layer-3 work arrive at full-time volume, and do you have a credible internal candidate to develop?
The flow makes one thing explicit: a standalone hire is the right answer only when both the workload and the evaluation capacity exist. Everywhere else, embedded support anchors the plan and is combined with upskilling or with a later, better-informed hire.
Combining patterns: launch with embedded support, hand over to the team
The three patterns are stages on a timeline, not mutually exclusive choices. A typical sequence runs in three phases.
- Launch (roughly the first six months). The embedded partner covers layers 2–3 and ships AI into the workflows with the clearest payoff. Internal candidates are pulled into the projects from day one — learning inside a live implementation beats classroom training for retention.
- Transition (roughly months six through twelve). Layer-2 decisions shift progressively to the internal candidate, while the external role narrows to layer-3 selection and review. If this shift is not happening, the engagement has quietly become outsourcing, and that is the signal to renegotiate it.
- Self-sufficiency. Day-to-day operation runs in-house; the external expert is consulted only for new technology decisions. Only at this point can a decision to hire a full-time specialist be made with actual workload data behind it, rather than hope.
The organizing principle: define the success of external support as “the company runs without it.” Before signing, ask concretely what will remain inside the company when the engagement ends.
FAQ
Q1. Where should we start?
A. Before opening a job posting, identify which of the three layers you are actually missing. If it is only layer 1, company-wide tool training solves it with no outside talent at all. Only a gap at layers 2–3 justifies comparing upskilling, embedded support, and hiring.
Q2. How is embedded support different from traditional consulting?
A. Traditional consulting delivers analysis and recommendations — a report. Embedded support includes execution and the transfer of capability to your team within its scope. We cover the model in depth in what is embedded consulting.
Q3. Who should we pick as the internal candidate to develop?
A. Prioritize cross-functional knowledge of your operations and low resistance to changing how work is done, over raw IT skill. The value in layer 2 comes from business knowledge, not technical background.
Q4. If we do hire, what should we watch for?
A. Scope the role tightly around layer 3, and verify beforehand that a full-time volume of that work exists. If it does not, fractional arrangements — a part-time specialist or a technical advisor — are legitimate alternatives to a full-time hire.
Summary
The problem “nobody here understands AI” is half solved the moment you decompose the capability: layer 1 belongs to company-wide training, layer 2 to insiders who know the business, and layer 3 to external expertise — or to a specialist hire, but only when the workload justifies one. For most SMBs the realistic path is to launch with embedded support and hand over to internal staff. For the broader picture of AI adoption in smaller companies, see where SMBs should start with AI; for how to make sound technology decisions without a specialist on staff, see technical decision-making without a CTO.