OpenAI, Claude or open models? How we choose AI for business work
There is no single best model. We pick per task, weighing quality, cost, speed and where your data is allowed to go. Here is how we think about it.
Clients often ask which AI we use. The honest answer is: it depends on the job. We work across commercial models from providers such as OpenAI and Anthropic, open-weight models we can run privately, and specialist services for voice and video. Choosing well matters more than choosing the newest.
The four questions we ask
- Where is the data allowed to go? Some documents can be processed by a reputable cloud provider under a business agreement. Others should never leave your own environment, which points to models we host privately.
- How good does the output need to be? Drafting a customer reply in your tone needs strong writing. Sorting emails into five categories does not.
- How fast and how often? A chat assistant needs answers in seconds. A nightly batch of documents can take longer and use a cheaper model.
- What does it cost at your volume? A few pence per task is nothing for ten tasks a day and significant for ten thousand.
Commercial models
For writing, reasoning and complex instructions, the leading commercial models are usually the strongest choice. We use business accounts and settings where providers do not train on your data, and we keep prompts and outputs inside your systems.
Open-weight models run privately
When documents are sensitive, or volumes are high and the task is narrow, an open-weight model running in your own cloud account can be the better fit. The trade-off is more setup and usually a little less capability, which careful prompting and validation often make up for.
Voice and video
Speech and video need specialist tools. Services such as ElevenLabs handle natural-sounding voice, and video understanding platforms such as Twelve Labs can search and summarise footage. We bring these in only when a project genuinely needs them.
Designing so you can switch
Models change quickly. We build so the model behind a task can be swapped without rebuilding the workflow around it. That keeps you free to move to something better or cheaper as the market changes, without being locked in.
You don't need to change everything. Just the right thing.
Most first projects go live within four to six weeks, with your team involved from day one.
Book a free call