Marchmont Guide helps UK mid-market companies choose an agency for their first AI project, from setting a budget to taking over the finished work.
Last updated:

Editorial selection

Full comparison

Fifty One Degrees

Fifty One Degrees is a close alternative if you want a bespoke system built in your environment and owned by your company.4,5

Best for
A mid-market company that wants to take over and run the system it buys.
Avoid when
The proposal leaves operation with your company, but nobody can maintain the system.
Read profile
Fifty One Degrees data engineering and BI page with a sample executive dashboard of revenue and margins
Image from Fifty One Degrees.

Brainpool AI

Brainpool is worth considering when you need to find out whether a bespoke AI idea will work before funding the full build.6,7

Best for
A defined prototype that answers whether further investment makes sense.
Avoid when
You need a running service, but the proposal only promises to explore whether it can be built.
Read profile
Brainpool AI homepage presenting Brainpool Cortex, its in house AI platform, with key figures
Image from Brainpool AI.

Full comparison

Compare the shortlist

Choose which providers to keep, then focus the evidence on one buying criterion.

Providers
1NexusA recurring business task with someone who knows the process and needs help delivering it.

Nexus is worth starting with if your team knows the job but needs help building the automation and improving it after launch.1,2,3

Best for
A recurring business task with someone who knows the process and needs help delivering it.
Avoid when
You must run bespoke code independently, and the proposal has not shown how you would take it over.

What the budget buys

Partly documented

Nexus tells us it charges per role automated, without day rates or consultant-time charges. Its public material describes a proof of concept before an annual commitment.3,8

Work left to your team

Documented

Nexus describes building the first procedure with the customer’s team.1

What counts as finished

Partly documented

Nexus says every deployed agent has a measurable business KPI.9

What you keep

Partly documented

Nexus’s telecom page says the business team is accountable for its agents. That does not tell you who owns the platform or what you can take elsewhere.10

Help after launch

Documented

Nexus says its team stays to improve the first version with the customer.2

Read profile
2Brainpool AIA defined prototype that answers whether further investment makes sense.

Brainpool is worth considering when you need to find out whether a bespoke AI idea will work before funding the full build.6,7

Best for
A defined prototype that answers whether further investment makes sense.
Avoid when
You need a running service, but the proposal only promises to explore whether it can be built.

What the budget buys

Documented

Brainpool describes a fixed-price check of project feasibility lasting four to six weeks.6

Work left to your team

Partly documented

Brainpool describes work from scoping and prototypes through implementation and continuing improvement.7

What counts as finished

Partly documented

Brainpool presents its Proof of Concept Factory as a way to check whether a project is feasible.6

What you keep

Documented

Brainpool says intellectual property in project assets transfers to the client. It separately describes Cortex as proprietary middleware.11,12

Help after launch

Partly documented

Brainpool includes continuing improvement in its delivery approach and says Cortex runs in the client’s cloud.7,13

Read profile
3Data Science UAA company weighing a complete outsourced project against adding people to its own delivery team.

Data Science UA is worth a look if you need people and skills as well as software. Its offer includes complete projects, staff working with your team and training.14,15

Best for
A company weighing a complete outsourced project against adding people to its own delivery team.
Avoid when
The proposal expects technical management from your company, but nobody can provide it.

What the budget buys

Partly documented

Data Science UA lists flexible pricing, project scoping and estimates.16,17

Work left to your team

Documented

Data Science UA offers complete project delivery, specialists working with the client’s team through outstaffing, and training for that team.14,15

What counts as finished

Partly documented

Its consulting offer includes analysing requirements and scoping the project.18,17

What you keep

Not established

What rights, code access and handover duties will your development or staffing agreement include?

Help after launch

Partly documented

Data Science UA offers client training. That leaves a separate question about who fixes incidents after the engagement ends.15

Read profile
4Fifty One DegreesA mid-market company that wants to take over and run the system it buys.

Fifty One Degrees is a close alternative if you want a bespoke system built in your environment and owned by your company.4,5

Best for
A mid-market company that wants to take over and run the system it buys.
Avoid when
The proposal leaves operation with your company, but nobody can maintain the system.

What the budget buys

Documented

The FAQ reviewed for this guide quotes a typical proof of concept below £15,000, taking two to four weeks.19

Work left to your team

Documented

Fifty One Degrees describes working within the client’s team and helping its people learn during the build.20,21

What counts as finished

Partly documented

Fifty One Degrees describes the result of its engagements as a production system in the client’s environment.4

What you keep

Documented

Fifty One Degrees says the client owns everything it builds.5

Help after launch

Partly documented

Learning during the build can help your team take over. It does not tell you what support you can call on afterwards.21

Read profile
5SuperchargeAn AI feature within a wider product that also needs design and engineering.

Supercharge is worth considering when the AI work belongs inside a customer product or a change to an older system. It brings product work, AI engineering and support into the same discussion.22,23,24

Best for
An AI feature within a wider product that also needs design and engineering.
Avoid when
You have funded a narrow automation, but the proposal requires a much larger product programme.

What the budget buys

Partly documented

Supercharge offers digital product delivery alongside agentic AI advice and engineering.22,23

Work left to your team

Partly documented

Supercharge describes its people as an extension of the client’s team.25

What counts as finished

Partly documented

With a product project, check the whole task a user performs as well as the AI answer inside it.22,23

What you keep

Not established

Which code, designs, data pipelines and deployment files will the agreement give your company?

Help after launch

Documented

Supercharge lists L1, L2 and L3 support and says its work continues after launch.24,26

Read profile
6VallianceA sponsor who can agree the starting position, how to measure value and what caused the improvement.

Valliance is worth comparing if you can measure the benefit and want to discuss fees tied to it.27,28

Best for
A sponsor who can agree the starting position, how to measure value and what caused the improvement.
Avoid when
You cannot measure the benefit, or you still disagree about which improvement the supplier can claim.

What the budget buys

Partly documented

Valliance says most of its fees are paid when it creates value, and that it does not bill by the hour.27,29

Work left to your team

Partly documented

Valliance describes small delivery teams that use AI.30

What counts as finished

Partly documented

When fees depend on value, agree how to measure the improvement and how much the supplier can claim to have caused.27

What you keep

Partly documented

Valliance says it does not believe in technology lock-in.31

Help after launch

Partly documented

A published Valliance case describes the system running in Newcastle University’s own Azure tenant.32

Read profile

From the archive

Top AI Agencies for UK Mid-Market Companies

Six AI agencies assessed for UK mid-market companies spending a first AI budget, with the criteria behind the order, a buyer's guide and a shortlisting checklist.

Read the article

When another provider fits

Fifty One Degrees

You want to own a system running in your environment.

Fifty One Degrees makes explicit statements on both ownership and delivery in the client environment. Put it beside Nexus when those are deciding requirements.4,5

Trade-off: The entry proof of concept still needs a separately agreed production scope and operating plan. Its published entry offer is not the price of the whole job.19,33

Brainpool AI

You need to know whether the idea will work before paying for a full build.

Brainpool’s fixed-price validation gives that first payment a clear purpose: answer the feasibility question.6

Trade-off: Allow a separate budget for implementation. Distinguish the project assets you will own from proprietary Cortex components.7,11,12

Data Science UA

You want the project to leave your own team stronger.

Data Science UA offers specialists to work with your team and client training, so it belongs in a discussion about adding skills as well as building software.14,15

Trade-off: Agree who manages those people and accepts their work. Extra staff do not automatically take over responsibility for the project.14,15

Supercharge

The AI feature is part of a larger digital product.

Supercharge combines product work with AI engineering, making it relevant to a brief that needs both.22,23

Trade-off: Can it quote a usable first release without making the whole product programme your first purchase?

Valliance

You want to pay against a measured business benefit.

Valliance says most fees depend on value creation. Compare that offer directly if you can agree what caused the benefit.27

Trade-off: Set the starting measure, fee cap and treatment of outside changes before signing. Linking a fee to value does not guarantee a successful project.27