The side-by-side comparison
| Decision | Independent consultant | Agency | In-house hire |
|---|---|---|---|
| Best scope | One bounded workflow or a short discovery. | Several connected workstreams. | A continuing portfolio of automation work. |
| Who you speak to | The person diagnosing and building. | Usually account, strategy and delivery roles. | Your own employee and internal stakeholders. |
| Delivery capacity | Focused and sequential. | Several specialists can work in parallel. | One role at first, then a team if demand grows. |
| Knowledge transfer | Strong when handover is part of the scope. | Depends on whether support stays managed. | Knowledge remains inside the business. |
| Procurement | Light for most smaller businesses. | Better fit for formal supplier requirements. | Recruitment, employment and management overhead. |
| Support | Defined handover or an agreed support block. | Can provide a staffed managed service. | Available as part of the role. |
| Main risk | One person becomes a capacity constraint. | The build gains handoffs and management layers. | You hire before proving there is enough useful work. |
Choose a consultant when
- You have one valuable process to test.
- You want direct access to the person making technical decisions.
- Your team can own the system after a documented handover.
- The scope is important but does not need several people working at once.
- You need someone to challenge the process before building.
This is the work my AI automation consultant service is built for. I map, build, test and hand over one controlled workflow.
Choose an agency when
- Several departments need different systems at the same time.
- The project needs security, data engineering, product design and change management in parallel.
- Your procurement rules require larger insurance limits, formal certifications or supplier depth.
- You want a staffed managed service rather than an internal owner.
- A single person would clearly become the delivery bottleneck.
More people can deliver a known scope faster. They can also multiply meetings and rework when the business has not agreed the input, rules, owner or desired output.
Hire in-house when
Recruit when automation is becoming an operating capability, not a project. The person needs a continuing backlog, authority to work across systems and somebody senior enough to own risk.
A common sequence is to use a consultant for discovery and the first build. The result proves the value, exposes the maintenance load and produces a better job description. You can then hire for work that exists rather than a title that sounds current.
Advice-only or hands-on?
Consultant does not always mean builder. Ask the question directly: who will create the workflow, connect the systems, run failure tests and write the operating guide?
I prefer a joined model for bounded work. The same person maps the process and builds the system. That removes the gap between a recommendation and the implementation. A larger programme may still need a separate delivery team.
Seven questions to ask any supplier
Who builds?
Meet the person who will make the technical decisions, not only the person selling the work.
What is excluded?
Ask about data cleaning, API access, hosting, software fees, monitoring and post-launch support.
Who owns access?
Your business should control the accounts, credentials, code and documentation it has paid for.
How is failure handled?
The proposal should explain retries, alerts, low-confidence outputs and human approval.
What proves success?
Name the current handling time, error rate or backlog before the project starts.
Who maintains it?
Decide whether your team, the consultant or a managed service owns changes after launch.
How do we leave?
Ask how access, data and documentation move if the supplier relationship ends.
Security is an ownership question
A larger supplier is not automatically safer. The proposal needs named responsibility for data access, testing, logs, incident handling and supplier changes. The NCSC guidance for secure AI systems treats security across the whole system lifecycle. The ICO risk toolkit helps teams examine data protection risks.
My honest fit boundary
If you need five departments transformed at once, continuous staffed support or a supplier that can absorb a large procurement process, use an agency. If you need a permanent queue of builds, recruit. The right answer is the operating model, not the nicest pitch.
Before comparing suppliers, confirm the use case with the guide to business processes AI can automate. Then use the UK cost breakdown to compare how each proposal allocates the work.
