Three terms appear constantly in procurement conversations, on LinkedIn profiles, and in introductory sales calls: consultancy, agency, freelancer. They are used almost interchangeably in the UK AI market. They describe genuinely different things — different scopes, different pricing models, different risk profiles, and different IR35 implications. Engaging the wrong model for your situation means you either overpay for capability you don't need, under-deliver on a project your organisation is depending on, or end up owning a system nobody internally can maintain.
This article draws a clear line between all three, explains when each is the right choice, and works through the IR35 considerations that UK buyers often underestimate. If you're an operations director, CTO, or COO evaluating AI support options, this is the decision framework you should work through before you sign anything.
Defining the Three Models
AI Consultancy
A consultancy is a structured firm — typically five to fifty people — that combines strategic advisory with technical delivery. The defining characteristic is that accountability sits with the firm rather than an individual. You engage Mason Bedford, not a named developer who might leave next month.
Consultancies typically offer a defined engagement structure: discovery, build, and handoff. Outputs are contractually specified. Intellectual property arrangements are documented. Post-project support is either included or available under a separate agreement. Fixed-scope engagements with fixed pricing are common, which gives finance and procurement teams predictability.
For regulated sectors — financial services under FCA Consumer Duty, legal firms under SRA AI guidance, businesses processing personal data under UK GDPR — the institutional accountability of a consultancy matters. A firm carries professional indemnity insurance. An individual may not, or may carry inadequate cover.
IR35 does not apply when you engage a consultancy. You are contracting with a firm, not directing an individual's labour.
AI Agency
Agencies in the AI space are typically digital or creative agencies that have added AI service lines to an existing offering. Many have strong capabilities in marketing automation, content generation tooling, and customer-facing conversational tools. Fewer have deep machine learning engineering or large language model integration expertise at the engineering level.
The agency model suits use cases where the output is relatively standard — a customer service chatbot, an email personalisation layer, an AI-assisted content production workflow. Agencies are well-positioned to manage these on an ongoing retainer, iterating and optimising over time. They are less well-suited to novel workflow automation, complex system integrations across enterprise infrastructure, or AI implementations in regulated environments where the decision logic needs to be documented and auditable.
Pricing is commonly retainer-based: a monthly fee covering management, iteration, and reporting. For ongoing AI marketing work, this model is clean. For a one-time operational AI build, it often creates cost drag — you're paying ongoing fees for work that was essentially complete at month three.
IR35 does not apply at the firm level for agency engagements.
AI Freelancer or Contractor
A freelancer is an individual practitioner operating as a sole trader or through their own limited company. The quality range is the widest of any model — from exceptional practitioners with deep domain expertise to generalists overstating their AI capabilities in a market where credentials are not standardised.
Freelancers carry the lowest day rate. They also carry the highest management overhead: you are responsible for evaluating their work, directing their activities, and managing the engagement week to week. Without internal technical capacity to do that well, you're likely to get poor results regardless of how good the individual is.
IR35 is a critical consideration here, and one that UK businesses routinely underestimate.
The IR35 Consideration
IR35 — formally the off-payroll working rules under Chapter 10 of the Income Tax (Earnings and Pensions) Act 2003 — applies when an individual provides services through an intermediary (typically their own limited company) but would, in substance, be considered an employee if engaged directly. Since April 2021, medium and large businesses have been responsible for making the IR35 determination, not the contractor.
For AI freelancer engagements, the risk is real. If the contractor works exclusively or predominantly for your organisation, works under your day-to-day direction, uses your equipment, and cannot substitute another individual to complete the work — they are likely inside IR35. In that case, as the end client, you are responsible for deducting PAYE income tax and employer National Insurance contributions before payment. The effective cost uplift is 30 to 35 per cent on top of the agreed day rate.
Many businesses discover this late. Either they've been paying a contractor as if they were outside IR35 when HMRC would view the arrangement differently, or they belatedly try to structure the engagement correctly and find that the changes required undermine the working relationship they've built.
Engaging an incorporated consultancy or a properly structured agency eliminates this risk entirely. You are buying a service from a firm, not directing a person's labour. This is one reason why, for engagements above a certain value, the IR35-adjusted cost of a freelancer often narrows the gap with a consultancy significantly. See our AI consultancy pricing guide for a fuller breakdown of how total cost of engagement compares across models.
When Each Model Is the Right Choice
When to Use an AI Consultancy
Use a consultancy when the problem is complex, the use case is novel to your organisation, or you need strategy and implementation from the same team. Complexity here means: multiple system integrations, regulated sector requirements, a use case that hasn't been bought off-the-shelf, or a need for the implementation to be understood and maintained by your internal team after handoff.
Consultancies are also the right model when you need institutional accountability. If something goes wrong with an AI system that is processing customer data or making decisions that affect your customers — in financial services under FCA Consumer Duty, or in a legal context under SRA guidance — you want a firm with PI insurance and documented processes, not an individual whose coverage may be capped at £500,000 or non-existent.
Fixed-price discovery is worth seeking explicitly. A well-run consultancy should be able to offer a bounded discovery engagement — typically four to six weeks — that produces a scoped specification and delivery estimate. This protects you from committing to a large build before you know what you're buying. We offer this at Mason Bedford, and it is consistently the most useful entry point for clients who haven't done this before.
Read more on how to evaluate firms in our guide to how to choose an AI consultancy in the UK.
When to Use an AI Agency
Use an agency when the use case is standard, the output is primarily customer-facing or marketing-adjacent, and you want ongoing management rather than a one-time build. If you need a chatbot for your website, an AI-assisted email personalisation layer, or tooling for your marketing team to produce content faster — an agency with a retainer model makes sense.
The caveat is to be clear about the agency's actual AI engineering depth. Many agencies have integrated AI tools (OpenAI APIs, third-party automation platforms) into their service offering without building genuine in-house LLM or ML capability. For simple use cases, this is fine — you're using well-tested tools to solve a common problem. For anything requiring custom model behaviour, fine-tuning, or integration with internal data infrastructure, probe their technical credentials carefully before signing a retainer.
When to Use an AI Freelancer
Use a freelancer when you have internal technical leads who can direct and evaluate the work, the scope is well-defined before engagement begins, budget genuinely constrains a firm engagement, and you have addressed the IR35 position correctly. A detailed technical specification written before the freelancer starts is close to a prerequisite. Without it, scope expands, direction becomes ambiguous, and quality becomes difficult to evaluate.
Freelancers work well embedded in an existing technical team — a senior engineer or data scientist brought in to accelerate a project your team is already leading. They work poorly as the only technical resource on a project that hasn't been properly scoped.
Hybrid Approaches That Work
The models are not mutually exclusive. Several hybrid approaches deliver good outcomes in practice:
- Consultancy for discovery and architecture; freelancers for delivery — the consultancy defines what to build and sets the quality bar; freelancers execute under that oversight. This works when the consultancy remains accountable for the architecture and does not simply disappear after the specification document.
- Agency for marketing AI; consultancy for operational AI — a clean separation by function. The agency manages your customer-facing AI tools on retainer; the consultancy handles the internal workflow automation that affects your operations.
- Consultancy oversight with in-house delivery — useful when you're building internal AI capability. The consultancy provides architectural direction and quality assurance; your team does the build. This is a structured version of what often happens informally and tends to produce much better outcomes when the oversight is genuine rather than nominal.
What to avoid: using an advisory consultancy and a delivery agency simultaneously without a clear integration plan and defined accountability. The advisory output doesn't get translated into delivery requirements. The agency builds something that doesn't match the strategy. Both engagements underdeliver and the business is left with a gap between where they were told they'd end up and where they actually are.
The cleanest test: if your project has never been done before in your organisation and you're not sure what to build — you need a consultancy. If you know exactly what to build and just need hands — a freelancer may work.
Questions to Clarify Which Model You Need
Work through these before approaching any provider:
- Is my use case well-defined, or do I need help scoping it? If undefined, you need advisory input before anything else. That points to consultancy.
- Do I have internal technical capacity to manage delivery? If no one internally can evaluate the quality of AI engineering work, a freelancer engagement is high risk.
- Am I in a regulated sector? FCA, SRA, ICO, or CMA-adjacent work typically requires documented processes, institutional accountability, and PI insurance that only a firm can provide.
- What is my budget? Below £10,000, a freelancer may be the only financially viable option — approach with clear scope and correct IR35 treatment. Above £20,000, a firm engagement is typically cleaner in total cost terms, particularly once IR35 adjustments are factored in.
- Do I need ongoing support or a one-time build? Ongoing management points to agency retainer. A one-time build with handoff points to consultancy.
- Who owns the IP? Freelancers working through their own company may retain IP by default unless the contract explicitly assigns it. Consultancies typically offer clean IP assignment as part of the engagement terms.
Decision Framework
| Scenario | Recommended model | Why |
|---|---|---|
| Novel AI use case, unclear scope, regulated sector | Consultancy | Needs strategy + delivery + accountability from one firm |
| Website chatbot or AI content tooling for marketing team | Agency | Standard use case; ongoing management suits retainer model |
| Defined technical task, strong internal tech lead, budget under £15,000 | Freelancer | Scope is clear; internal capacity to direct and evaluate exists |
| Complex operational AI, multiple system integrations | Consultancy | Integration complexity requires architectural accountability |
| AI marketing automation on ongoing basis | Agency | Retainer model fits continuous optimisation need |
| Freelancer embedded in existing tech team | Freelancer + internal oversight | Works when direction and evaluation capacity exist internally |
| Discovery only — scoping before committing to build | Consultancy (fixed-price discovery) | Produces specification before any build commitment is made |
For a detailed comparison of what AI advisory and implementation actually costs across different engagement types, see our AI consultancy pricing guide. For the broader context of whether you need external support at all versus building in-house, our article on in-house AI versus consulting covers the build-vs-buy decision in detail.
Where Mason Bedford Fits
We are a consultancy. We do both advisory and build, and we don't separate them into different teams or different engagements unless the client's situation specifically calls for it. Our delivery team is in-house — we don't use freelancer subcontractors to deliver client work, which means the people scoping your project are the people building it.
We offer a fixed-price discovery engagement for clients who are not yet certain what they need to build. The output is a scoped specification, a delivery estimate, and an honest assessment of whether the project makes commercial sense. Some clients take that specification and engage a different firm or their own internal team to build. That's a legitimate outcome and we're not in the business of creating dependency where it doesn't serve the client.
Where scope is sufficiently defined, we offer fixed-price build engagements. We work with businesses across the UK and the United States, primarily in professional services, financial services, legal, and logistics — sectors where AI implementations need to be robust, auditable, and maintainable after handoff.
If you are trying to work out which model applies to your situation, the UK AI consultancy guide covers the broader landscape. If you're ready to discuss a specific project, the right next step is a conversation with our team — details at our services page, or reach us directly via the contact page.