The hardest part of improving a business is often not the work itself. It is deciding what deserves attention, finding someone who can handle it, giving that person enough context, reviewing the result, and making sure the learning carries into the next task.
An agency, an in-house team, and a business operating system solve different parts of that problem. None is automatically best. The right model depends on the work, the frequency of decisions, the risk of error, and how much coordination the owner can absorb.
The three models
Agency
An agency is an external team contracted to own a defined area such as advertising, search, web development, brand, or communications. Strong agencies bring specialist experience, operating processes, and production capacity without requiring the business to hire every role.
The tradeoff is distance. The client must transfer context, approve work, and judge whether agency activity is creating business value. Broad retainers can also make it difficult to see which task consumed the budget.
In-house team
Employees can develop deep knowledge of the business, stay close to customers, and coordinate daily with leadership. The model provides direct control and continuity when there is enough recurring work to justify the roles.
The cost extends beyond salary: recruiting, management, payroll obligations, tools, benefits, training, and coverage all matter. The SBA hiring guide outlines the operational steps involved in employing people in the United States; requirements vary by jurisdiction.
Business operating system with execution
In this model, software maintains business context, turns findings and owner requests into a shared backlog, prioritizes tasks, and routes eligible work to AI, people, or a hybrid workflow. The owner sees the evidence, decides what to activate, and approves sensitive changes.
Its strength is coordination across many small, recurring improvements. Its limits depend on product coverage, data quality, integrations, execution capacity, and the human support available. Complex strategy, regulated decisions, bespoke creative projects, or major technical transformations may still require specialists.
Compare the operating questions
| Question | Agency | In-house | Operating system + execution |
|---|---|---|---|
| Who finds the work? | Usually within the contracted scope | Managers and employees | Monitoring, review, owner requests, and support staff |
| Who prioritizes it? | Agency and client | Internal leadership | Shared task logic with owner visibility |
| Where does context live? | Briefs, meetings, agency tools | Team knowledge and internal systems | Business profile, evidence, tasks, and approvals |
| How does work move? | Retainer, project, or statement of work | Assigned through normal management | Activated within plan coverage and capacity |
| What needs owner time? | Briefing and approvals | Hiring, management, and decisions | Input, access, activation, and sensitive approvals |
| Best fit | Defined specialty or campaign | Core, continuous capability | Coordinating recurring cross-business improvements |
The table is a starting point, not a verdict. A capable business may use all three.
Calculate the coordination cost
Compare more than the invoice or salary. For each model, estimate:
- time spent finding and briefing people;
- the effort to provide access and assets;
- review and approval time;
- rework caused by missing context;
- the cost of work waiting between owners;
- management and administrative overhead; and
- the cost of a wrong or delayed decision.
A low hourly rate can become expensive when the owner must define every task and repair every handoff. A higher-cost specialist can be efficient for a high-stakes, bounded problem. An operating system can reduce repeated coordination, but only if its task quality and execution coverage are trustworthy.
Match the model to the work
Use an agency when
- the business needs a discipline it cannot justify hiring internally;
- the outcome has a defined campaign or project scope;
- external pattern recognition matters; and
- the business can provide a decisive owner and timely feedback.
Build in-house when
- the capability is central to competitive advantage;
- work is continuous and context-heavy;
- decisions must happen close to customers or operations; and
- the organization can recruit, manage, and retain the role.
Use an operating-system model when
- opportunities arrive across several business areas;
- the owner needs help deciding what matters next;
- many tasks are focused, recurring, and evidence-led;
- AI can prepare work but important actions still need human judgment; and
- the business benefits from one backlog, approval trail, and completion record.
Design a hybrid on purpose
A practical hybrid might look like this:
- the owner sets goals, budget, risk tolerance, and approval rights;
- the operating system reviews signals and maintains the backlog;
- AI prepares research, drafts, and routine analysis;
- internal staff contribute customer and operational context;
- specialists handle regulated, high-risk, or bespoke work; and
- the owner approves material public changes.
Do not let the hybrid become ambiguous. Every task still needs one accountable owner, one definition of done, and one place where evidence and approval are recorded.
Protect worker and vendor boundaries
Calling someone a contractor does not necessarily determine their legal status. Control, economic dependence, and the real working relationship matter. Review applicable employment and tax rules; in the United States, the Department of Labor provides current misclassification guidance. Other countries and provinces use their own tests.
An AI system also needs governance. NIST’s AI Risk Management Framework provides a voluntary structure for mapping, measuring, managing, and governing AI risk. The presence of automation does not remove the business’s responsibility for decisions or published work.
Use a 30-day evaluation
Choose one important workflow and compare models against the same criteria:
- Was the right task identified?
- Did the recommendation include enough evidence?
- How much owner coordination was required?
- Was the work completed to the agreed standard?
- Could the owner verify the result?
- Did the learning improve the next task?
- What was the full cost, including internal time?
The best model is the one that repeatedly produces accountable progress at a cost and level of owner involvement the business can sustain.