MILabs
Tools

Granite AI models for small business workflows

What IBM's Granite models are, what small businesses use them for, and how a Sydney AI agency sets one up for your team.

Updated 2026-07-30

What Granite is

Granite is a family of language models from IBM. A language model reads and writes text, so it can answer questions, summarise documents, draft replies, sort incoming messages and pull fields out of forms. IBM offers Granite through its watsonx platform, and publishes model files that a business can run on infrastructure it controls. That second option suits owners who want closer control over where their information goes. A model sits at the centre of an AI system, producing text or a decision while other parts decide what it can see and what happens to its output. Connections to an inbox, a calendar or an accounting file are built around the model by whoever sets the system up. For a small business, the choice of model usually comes down to cost, data handling and how much of the stack the owner wants to manage. Granite is one option among several, and it deserves a fair comparison with other model providers before a decision is made.

IBM: official site

Granite in plain language

Most business software follows rules someone wrote in advance. A language model works differently. It takes a block of text, including instructions and any supporting material, and returns new text in response. Because the output is generated rather than looked up, a fluent reply can be right or wrong, and the text reads the same either way.

Two questions shape most decisions. The first is where the model runs. A hosted service sends each request to the provider's infrastructure, which is simple to start with. A published model can instead be installed on a laptop, an office server or a private cloud account, which trades per-use charges for hardware and upkeep.

The second question is what surrounds it. The model keeps no memory of your business between requests. Anything it needs, such as a price list or a policy document, has to be retrieved and placed in front of it when the question is asked. Supplying the right source material at the right moment usually decides most of an assistant's quality.

Smaller models usually need less hardware, while larger ones tend to handle long documents and tangled instructions more reliably. Testing two or three candidates on real examples from your own inbox is a more dependable guide than published benchmark scores.

What businesses use it for

  • Replying to common customer questions from a published set of approved answers
  • Preparing replies to enquiries, quotes and follow-up emails for staff to review
  • Condensing long documents such as contracts, tender packs and supplier terms
  • Labelling incoming messages so the right person sees them first
  • Pulling names, dates and amounts from forms, invoices and intake sheets

Who it suits

  • Owners who want the model to run on infrastructure they control
  • Teams handling large volumes of documents to read and summarise
  • Businesses comparing model providers before committing to one

How MILabs sets it up

MILabs starts with the owner or team lead to decide where the model will run. Some businesses prefer a hosted option through IBM's platform, while others want the model on a machine or server they control. We then map the systems the business already uses, such as the email account, shared drives, booking software and accounting file, and work out which the model needs to read and which it needs to write to. Access is set first. Each connection gets its own permissions, staff see only what their role requires, and the actions the model can take are limited to those the owner approves. Finally, the owner and MILabs agree who looks after the setup day to day and who reviews the answers.

Builds it fits

Industries it fits

Granite in practice

In a working office, a Granite-based assistant usually sits beside a task rather than replacing one. Picture a shared mailbox where an incoming message arrives, the assistant reads it against the business's own approved wording, and it places a suggested reply in a draft folder. A staff member opens the draft, changes what needs changing and sends it under their own name. The model never reaches the customer directly, and the person who sends the message stays accountable for it.

Most of the design effort goes into the boundaries of that loop. Someone decides which folders the assistant can read, where it may write drafts, and what it must never touch, such as payment records. Source material lives in a short collection of documents that a named person updates when prices or policies change.

Over time the team builds a small set of reference cases, such as a typical enquiry, a tricky one and a message that should go straight to a person. Rerunning them after any change to the model or its source documents shows regressions before a customer does.

Routine work such as sorting and tagging is a natural place to begin. A setup that keeps a log of what the assistant read, drafted and escalated lets the owner review patterns.

good questions, straight answers

Frequently asked questions

When is Granite the wrong choice?
A different model or an existing product can serve better when a business needs the strongest possible reasoning on very complex, open-ended tasks, or when a tool the team already uses covers the job well. MILabs will talk through those options during the first conversation, including cases where building is not the answer. The right answer is found together, so the owner leaves with a clear view.
Can my team use Granite without training?
Yes, staff can use a Granite-based assistant without learning how the model works, because the interface they see is built around their task. A short walkthrough covers what the assistant does well, which tasks to hand it first and how to steer it toward a different answer. MILabs covers that handover during setup, and the owner decides how much time the team wants. Most teams feel comfortable after one walkthrough.
Can MILabs build on top of Granite?
Yes. MILabs can build workflows and assistants on top of Granite where it suits the business, connecting the model to the email, documents and software the team already uses. Each build follows the business's own rules, such as which records the model may read and which actions need approval first. Whether Granite is the right model depends on the task, the data involved and where the owner wants it to run.
What does a Granite setup need from me?
A Granite setup needs an owner or team lead who can decide where the model runs, which systems it can reach and who checks its answers. You grant access to the accounts and files involved and name one person to approve changes. Clear examples of good and bad answers help most, as do the real documents the model should rely on. No knowledge of machine learning is needed.
What does Granite not do?
A model like Granite produces text from what it is given. Prices, stock and client history are known to it only when those records are supplied, and it can state something wrong with confidence. Any action in your systems, such as sending an email or updating a record, is built and permitted separately. A person reviews results before they reach a customer.
let's make something happen

Workflow automation: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.