MARMATIONDiscuss a project
MARMATION

Start with the job.Then choose the architecture.

Discuss a project

AI Consulting & Product Engineering

AI Consulting & Product EngineeringArchitecture follows purpose.

The next generation
of software is intelligent.
We build it.

AI products, useful company knowledge and connected workflows for established small and mid-market businesses.

Discuss a project
How the parts fit togetherProduct design, AI and existing systems need to work together. Clear access rules, testing and human review turn a promising feature into useful software.
Product thinking. Systems engineering.
01 / Three places to make a difference

Where could better software make a difference?

A better experience for customers. Answers your team can find. Work that moves without another spreadsheet. Start with the change that matters.

01 / For your customers

AI products

Build something people choose to use.

Turn your expertise into a client portal or intelligent feature. Help customers compare options, review a suggestion and act.

  1. AI suggestsA useful starting point
  2. Your customer decidesReview, change or reject
  3. The product remembersA saved decision
An example interaction, with the customer in control.

Worth measuringCan customers finish the task, and do they come back?

Explore AI products
02 / For your team

Company knowledge

Put the right answer within reach.

Make policies, project files and service knowledge easier to use. Give your team supported answers from current sources they are allowed to see.

“Which policy applies here?”

Find
Current, permitted sources
Explain
An answer linked to the evidence
Know the limit
Say when the answer is missing
An example question, answered from approved information.

Worth measuringCan people find a supported answer without asking around?

Explore company knowledge
03 / For the business

Business workflows

Make the whole process work.

Connect a service request from sales to delivery, with the information, approvals and next action in the tools you already use.

  1. Request received
  2. Person or rule approves
  3. Next team can act

When something stops: show what is waiting and who can resolve it.

An example handoff, with a route for unfinished work.

Worth measuringDoes work finish with fewer handoffs and less rework?

Explore business workflows

These are examples of possible work, not client case studies. The first step is to establish what would be useful in your business.

A practical fit

Your software.
A practical first step.

Build around a task that matters to your business, without needing to assemble a permanent software team. Agree the source code, accounts, licensed components and handover before the build.

Bring one recurring task, the tools you use today and the result you want to improve. The first conversation should establish whether custom software is worth owning and what a useful first step would be.

Discuss a focused project See ways to work together

If the right move is still unclear, start with AI Consulting & Architecture. Existing software teams can also explore AI-Native Engineering Enablement.

02 / Try it for yourself

Good ideas deserve
a working test.

A product needs to handle mistakes. A knowledge tool needs to respect access. A workflow needs to recover when a connection fails. Test those decisions before expanding the build.

The Lab includes one working order-review example with sample data and fixed rules, alongside proposed product and knowledge designs. Use it to inspect how a small part of a system behaves.

Explore the Lab
Inspect the working order-review example
Document operationsInteractive browser example

Change the order. Check it again.

Example order / PO-2001 / sample data

  1. 01
    Change quantity: 12 → 13Approval cleared
  2. 02
    Try saving before approving the changeNot saved · fresh approval needed
  3. 03
    Approve, save, then save againOne order saved · no duplicate

Changing the quantity means someone must approve the new version. Saving it twice still creates just one order.

This example uses sample orders and fixed rules in your browser. No AI model or customer system is connected.
See what the example checks
03 / Before you invest

Five questions worth answering first.

Before choosing a model or a supplier, agree what should improve, how you will check it and who will look after the result.

You may already own software that can do the job. Compare buying, adding to an existing tool, connecting systems and building something new.

Compare the four routes How we move from decision to delivery
Value

Which change earns the investment?

Choose one task and agree what better would look like: less rework, a faster review or a more useful customer experience. Measure today’s process first, including the cost of changing it and the reasons to stop.

What to agree

A picture of today’s process, a measure of success and one useful first release.

Control

What is the system allowed to do?

Decide what the software may see, suggest and change. Follow company information through the connected tools and AI service. Agree who can access it, how long it is kept and which actions need a person’s approval.

What to agree

A clear map of where information goes, who may use it and what needs approval.

Quality

How will we know it works?

Choose real kinds of work to test, including missing information and mistakes that would be costly. Use the same examples to check each release. Agree when the software should stop and ask a person.

What to agree

Example tasks, agreed standards and a clear way to involve the right person.

Economics

What will a useful outcome cost?

Include the AI service, hosting, licences, connections, human review and support. Look at the cost of finishing a useful task as well as the monthly bill. A cheap answer is not good value if someone has to redo the work.

What to agree

An estimate based on expected use, review time and support, rather than just the price of the AI model.

Ownership

What happens if it works?

Decide who will run the software, who will improve it and what the team will do with the time it saves. Agree the access, documentation and support they need to take over confidently.

What to agree

A named owner, practical operating instructions and a plan to help people use it.

04 / Go deeper into the architecture

How the pieces work together.

Explore the decisions behind a product, company search tool or internal workflow: who can use it, what it can access and what happens when something goes wrong.

Systems Atlas3 types of system · 18 design decisions

Systems Atlas

Change the business context. Open a layer. Inspect the decision, what can fail, and the evidence worth asking for.

Intelligence inside a product.

Follow a customer’s task from the first request to a checked result. See where privacy, corrections and approval belong.

Example system designLocal / not connected
  2. Read this decision
  • The customer’s task to Customer privacy: Check the user and customer account.
  • Customer privacy to Relevant information: Limit access to permitted records.
  • Relevant information to The AI suggestion: Provide relevant permitted information.
  • Customer privacy to The AI suggestion: Limit what the AI may do.
  • The AI suggestion to Permitted actions: Validate the proposal.
  • Permitted actions to Check each release: Observe the task result.
  • Check each release to The customer’s task: Improve the product task.
DependencyEvidenceFeedback / recoveryChoose a step to learn more

Customer privacy

AUTH
The decision to make
Check which customer account and user are making the request before looking up information or taking action. Carry those access rules through searches, stored answers and connected tools.
What could go wrong
One customer sees another customer’s records through a search, a stored answer or a connected tool.
What to ask the team to show
  • Tests that try to cross between customer accounts.
  • A list of what each user role may view or change.
Return to the diagram

A model may propose an action. The product must decide whether it is allowed.

Example system design.
Not a customer deployment.
05 / From the Marmation Review

The decision behind the technology.

Browse all perspectives
What actually
needs to change?
Buy
An existing product meets the important requirements.
Extend
A platform fits; a focused addition closes the gap.
Integrate
The value comes from making existing systems work together.
Build
A distinctive product or workflow warrants a system of its own.

Four options to evaluate. Requirements decide the route.

Decisions / 5 min

Build, buy, extend or integrate? The AI decision that comes before the technology.

Read when: The roadmap needs a build-or-buy decision.

Read the article

Bring us the
difficult part.

A product idea, a broken workflow, scattered company knowledge or an integration others keep working around. Tell us what needs to change and what has already been tried.

Discuss a project