The Agency Engagement Models Guide: projects, retainers, embedded, and hiring in-house

There are four ways to buy digital expertise: a project (a defined thing, built once), a retainer (ongoing work from outside your walls), an embedded model (senior agency people working inside your team), and an in-house hire (a colleague on payroll). Each is the right answer to a different situation, each fails predictably when applied to the wrong one, and the industry's quiet problem is that agencies tend to recommend whichever model they happen to sell. This guide treats all four fairly, including the two we don't offer.

The decision rule that organises everything below: work out what you're buying. An outcome, a stream of work, a colleague's judgement, or a permanent capability. Those map to project, retainer, embedded, and hire respectively, and most engagement disasters are a mismatch between what was bought and what was needed, discovered around month six.

Buy an outcome: project. A stream of work: retainer. Judgement in the room: embedded. A permanent capability: hire.

A project buys a defined outcome: a brand, a website, an AI build, scoped, priced, shipped, done. It's the right model when the thing has edges: you can describe what finished looks like, the work has a natural end, and ongoing ownership can transfer to your team afterwards. It's how most agency relationships should start, because it lets both sides judge the fit on something real before anyone signs an ongoing arrangement.

For orientation, our project pricing: brand refreshes from $5,000, full brand builds $25,000 to $100,000+, websites $10,000 to $50,000, AI builds from $5,000. The failure mode to watch is the project that never ends: scope creeping into a de facto retainer without anyone renegotiating, which serves the agency's revenue and nobody's clarity. A good project has a finish line both sides can see, and a good agency points at it.

When projects are the wrong model: work with no natural end. Search is the canonical example; "do our SEO as a project" is buying a haircut that promises to last forever, and any agency selling one-off SEO projects at meaningful prices is selling against the nature of the work.

A retainer buys an ongoing stream of work from outside your walls: a monthly commitment, a defined scope, results reported on a rhythm. It's the right model for continuous disciplines where the work compounds: search is the obvious one, and it's why our Search + AI Visibility service is a retainer at $5,000 to $10,000 a month rather than a project. The agency carries the expertise and the tooling; you carry the context across in briefings and reviews; the report is the interface.

The model's known disease is retainer theatre: activity reports where results should be, hours consumed as the deliverable, and a monthly meeting whose real function is justifying the next month. The defence is simple and worth writing into any retainer you sign: outcomes named upfront, a report you can read without a specialist, and an exit that doesn't require a lawyer. A retainer you can leave easily is a retainer the agency has to keep earning, which is the only healthy kind.

When retainers are the wrong model: when the bottleneck is context rather than capacity. If the agency keeps missing not because the work is bad but because they don't know what you know, no amount of better briefing documents fixes it, because the briefing document is the problem. That's the boundary where the next model starts.

The embedded model buys judgement inside your rituals: senior agency people who join your planning, work in your tools, see the context firsthand, and make calls in the room, with the agency's wider bench behind them for the disciplines they don't personally carry. It exists because most agency work that misses does so from lack of context, not lack of skill, and no briefing document carries context the way sitting in the Monday meeting does.

Draw the line carefully against its lookalike, because this is where the market blurs on purpose. Staff augmentation rents capacity you direct: a contractor in a seat, valuable when you already know what needs doing. Embedded supplies the direction itself: the senior person is there precisely because you don't have that judgement in-house yet. Confusing them costs real money in both directions; augmentation priced as embedded is overpaying for hands, and embedded staffed like augmentation is a junior in a senior's chair.

An embedded seat without a hard cap is a retainer with better marketing.

The structural truth of the model is the capacity cap. A person can be genuinely embedded in very few businesses at once; the whole value is presence and context, and both divide badly. So the real version of this model always comes with a stated, small, hard limit on concurrent clients, and the absence of one is the tell that you're being sold the word rather than the arrangement.

Ours, on the record: $15,000 a month, three-month minimum, capped at three concurrent embedded clients. The minimum exists because month one is context-building, and judging the model on it is judging a hire on their first fortnight. The cap exists because the cap is the model. When the three seats are full, the answer to the fourth enquiry is a waitlist or a different model, not a quiet stretch.

When embedded is the wrong model: when your need is one discipline at full-time load (hire), when the thing has edges (project), or when you have strong internal direction and just need execution (augmentation, or a retainer). It's a specific tool for a specific situation: senior, multi-discipline need, now, without a full-time role's worth of any single discipline.

Sometimes the right answer is no agency at all, and a guide from an agency that can't say so isn't a guide. Hire in-house when a discipline has become permanent and full-time: enough work in one lane, every week, to fill a role, with enough organisational gravity to keep a good person growing. A full-time need deserves a full-time colleague, and an agency arguing against a justified hire is protecting revenue, not advising you.

Cost the hire properly, though, because the salary is not the price. Add superannuation, tools and subscriptions, management time, onboarding months, and the recruitment risk of getting it wrong, and the fully loaded cost of genuine seniority comfortably exceeds most retainers. That's not an argument against hiring; it's an argument for comparing real numbers to real numbers. The common mistake runs the other way too: hiring one mid-level generalist to cover brand, web, search, and AI, which buys you four disciplines at the depth of none.

The pattern that works for most growing businesses is sequencing: agency models while the need is varied and forming, then hire into the lane that's proven permanent, with the agency narrowing to the disciplines that stay episodic. A good agency helps you make that transition and shrinks its own scope accordingly. It's also how you tell the good ones from the rest.

The decision, compressed into four questions. Does the work have edges you can describe? Project. Is it continuous, compounding, and directable from outside? Retainer. Is the missing ingredient senior judgement inside your context, across disciplines? Embedded. Has one discipline become permanent and full-time? Hire. Answer straight and the model usually picks itself; the hard cases are transitions, where yesterday's right answer has quietly become today's wrong one.

Mixing models is normal and healthy: a project to start, a retainer for the compounding work, embedded for a growth phase, a hire when a lane matures. What matters is that each piece of work sits in the model that fits it, and that someone revisits the fit yearly, because businesses outgrow engagement models the way they outgrow brands.

Whichever direction you lean, three questions expose whether an agency is recommending a model or selling one. Which of your models is wrong for me, and why? (An agency that can't name one thinks every problem is a nail.) What does the exit look like? (Healthy models survive easy exits.) And for anything called embedded: what's the cap, and where are you against it right now? A straight answer to all three is rarer than it should be, and it's worth more than any case study. For the AI visibility version of this screen, see DIY or Hire It Out, the closing chapter of our book.

Our own answers, for the record: we sell projects, the search retainer, and the embedded model; we don't sell staff augmentation, and we'll tell you to hire when hiring is the answer. The first call is free, there's no follow-up sales sequence, and the recommendation you'll get is the model that fits, including the ones that aren't ours.