01What custom ML development includes
- Assistants on your own data: chat assistants that answer from your documents, policies or product data, and say when they do not know.
- Reading documents: pulling names, amounts and dates out of invoices, forms and PDFs into your system.
- Predictions: models for demand, churn, pricing or risk, trained on your own history.
- Generative AI in your product: text, image or video generation features inside your software.
- Automating enquiries: AI that handles routine questions on chat, WhatsApp or phone and hands the rest to your team.
02Custom or off-the-shelf?
Your data
- An off-the-shelf tool fits when
- The task works on generic or public information
- Custom development fits when
- The value is in your own documents, history or product data
Your workflow
- An off-the-shelf tool fits when
- You can adapt your process to the tool
- Custom development fits when
- The AI has to sit inside your existing software and steps
Privacy
- An off-the-shelf tool fits when
- Sending data to a third-party service is acceptable
- Custom development fits when
- You need data kept in a region, not used for training, or on your own servers
Cost control
- An off-the-shelf tool fits when
- Per-seat pricing suits your usage
- Custom development fits when
- You need limits, monitoring and predictable running costs
Differentiation
- An off-the-shelf tool fits when
- AI is a back-office convenience
- Custom development fits when
- AI is part of what you sell
03How a custom ML project runs
- Pick the use case: find the task where AI saves the most time or money, and check it is realistic.
- Small proof first: a working prototype on your own data, so you can judge the results before spending more.
- Build it in: connect the feature to your product or tools, with limits, logging and fallbacks.
- Watch and improve: track quality and cost after launch and tune as real usage comes in.
04Questions to ask a custom ML development company
Services in this area vary a lot. These questions show whether a team will build something that works for you.
- Will you test two or three models on my data before choosing one?
- Do you start with a small proof, and what will it show me?
- How will we measure whether the output is good enough?
- What will it cost each month to run, and how is that capped?
- Where is my data stored, and is it used to train anyone's model?
- Who owns the code, the prompts and the models at the end?
05When custom machine learning is not the right choice
- An existing tool already does the job well enough for your data and process.
- The task is not clearly defined, so nobody could tell whether the AI is right.
- You have no data or documents for it to work from, and no way to get any.
- The result must be perfect every time, with no human check.
Saying this early saves money. We would rather tell you to use a ready-made tool than build something you do not need.
06What you need to start
- A clear task with a result you can measure, such as hours saved or errors reduced.
- Some of your own data or documents to work from.
- A person who owns the process and can judge whether the output is good.
07Running costs and data privacy
AI features have running costs that grow with use, so they should be estimated before you commit. A good proof stage includes an estimate of the monthly cost to run the feature.
On privacy, use providers and settings that do not train on your data, keep data in the region you need, and consider running open-source models on your own servers when the data is sensitive.
08How we work at MOUNTSMY
We compare two or three models on your own data before choosing one, and we start with a small proof so you can judge results first. We build with models from OpenAI, Anthropic and open-source families such as Llama, and with Python tools such as PyTorch, scikit-learn, LangChain and FastAPI. Most AI features are added to software you already have.
?Frequently asked questions
How do I know if I need custom machine learning?
If an off-the-shelf tool already solves the task with your data and process, you probably do not. Custom is worth considering when your data, workflow, privacy needs or product make a generic tool a poor fit.
Is our data safe?
We use providers and settings that do not train on your data, keep data in the region you need, and can run open-source models on your own servers when required.
Which AI model do you use?
Whichever fits the job and the budget. We often compare two or three on your own data before choosing.
How much does it cost to run?
It depends on how much it is used. We estimate monthly running costs during the proof stage, before you commit.
Can AI fit into our existing software?
Usually, yes. Most AI features are added to software you already have rather than built as a separate product.
What is the difference between machine learning and generative AI?
Machine learning is the wider field of models that learn from data, such as predicting demand or reading documents. Generative AI is one kind, which creates text, images or code. Many projects use both.
How long does a custom machine learning project take?
It depends on the task and your data. We start with a small proof, so you see real results early, and give a timeline and quote in writing after a free call.