Freelancer, AI agency, or interim consultant: how do you choose?

Compare the operating model, not the sales pitch

The same AI project can fail under any label when nobody owns decisions. Examine who joins meetings, who builds, who handles absence, and what happens after launch.

• Freelancer: direct and flexible, but dependent on one person.
• Agency: broader capacity, but check who actually does the work.
• Interim consultant: close to the team, but needs a clear internal sponsor.
• Project team: strongest for integrations and parallel disciplines.

Put continuity in the agreement

Ask how code, credentials, documentation, monitoring, and handover are managed. A low rate does not compensate for a workflow nobody can maintain.

Compare five decision factors, not only rates

Freelancer, agency and interim consultant are different operating models, not quality levels. Compare them on complexity, speed, internal ownership, continuity and required disciplines. A freelancer may be the strongest specialist for a bounded integration; an agency can deliver parallel workstreams; an interim consultant can organise daily decisions and knowledge transfer. State which risks the client can own and where external cover is genuinely required.

• Scope: stable and narrow, or changing across the organisation.
• Capacity: one specialist, several disciplines or ongoing coordination.
• Continuity: dependency on one person and cover during absence.
• Ownership: who prioritises, accepts and operates after launch.
• Speed: serial delivery or several workstreams in parallel.

Match each model to the right engagement

Use a freelancer for a specific technical task such as an API connector or evaluation framework. Use an agency when design, engineering, data and rollout must happen together under one delivery owner. Interim support fits moving priorities, cross-functional coordination and the creation of lasting internal capability. A project team is usually stronger than one consultant for business-critical platforms involving several integrations.

• Freelancer: specialist task with clear inputs and acceptance criteria.
• Agency: bounded multidisciplinary project with delivery accountability.
• Interim: temporary embedded role with mandate, rhythm and knowledge transfer.
• Project team: parallel engineering, security, data, integration and adoption.

Calculate the total cost of collaboration

A lower day rate can disappear through extra coordination, slow delivery or rework. Include the internal management each model requires, the cost of absence and the support available after launch. Agencies carry overhead but may provide continuity and several disciplines. Interim consultants work closely with the organisation, but their time must create delivery as well as meetings. Compare cost per completed phase and operational result rather than per hour alone.

• Include selection, onboarding, internal guidance and quality assurance.
• Price the delay created when one specialist is unavailable.
• Compare the actual scope of documentation, monitoring and support.
• Separate recurring licences and operations from project delivery.

Combine models without fragmenting accountability

Many SME projects benefit from an interim lead protecting scope and internal decisions, supported by freelancers or an agency for specialist delivery. This is flexible only when one person remains accountable for integration and acceptance. Use a shared backlog, common architecture principles and one definition of done. Otherwise each supplier may optimise its component while user experience, data flow and operation fall between contracts.

• Name one product owner and one technical authority.
• Use client-controlled repositories, cloud accounts and documentation.
• Demonstrate complete business workflows rather than isolated components.
• Define escalation and replacement for every critical role.

Use a short procurement checklist before signing

Ask who will actually perform the work, how much time they have and how absence is covered. Check whether cases match the process risk and integration complexity, not merely the industry. Agree intellectual property, data use, subcontractors, security and exit terms. Where possible, begin with a paid phase that creates standalone value. You can then judge working relationship, quality and speed before committing the full budget.

• Names, roles, seniority and availability of the delivery team.
• Deliverables, exclusions, acceptance criteria and gates by phase.
• Ownership of code, prompts, documentation, accounts and generated data.
• Support, response times, handover and termination arrangements.

Make the choice with a weighted score rather than instinct

Score each model on specialist depth, speed, continuity, internal embedding, scalability and total cost. Weight the factors for the engagement: continuity matters more for a critical integration, while individual expertise may dominate a technical audit. Ask two stakeholders to score independently and then discuss differences. Include a no-buy option, such as establishing internal ownership first. This prevents the best-known supplier or lowest rate winning when its operating model does not fit the work.

• Use identical project context and assumptions for every option.
• Record the evidence or risk supporting each score.
• Test the preferred model through a small paid first phase.
• Revisit the model when scope, risk or required disciplines change.