What we doThe Mansions · AI development services

AI products.

Have an idea for an AI product, or want AI inside the software your team already uses? We build it, and everything we build is yours.

Book a free consultationTake the free AI readiness quiz

Not sure your data is ready for AI? The quiz shows you in five minutes.

  1. 01AI inside your team's software
  2. 02Chatbots
  3. 03Algorithms
  4. 04Products built on your own data

What AI development services actually do

AI development services build software that can read, write and work things out from what your business already knows. Sometimes that's a whole new product, but most of the time it's AI integration, putting the AI inside the software your team works in. We build it from three blocks.

  1. Your data

    The first is your data, the documents, records, emails and know-how the AI works from. Without it, the smartest model in the world only knows business in general.

  2. The model

    Then there's the model, the AI underneath that does the reading and the writing. We pick the one that fits the job, and sometimes we use two, one to sort the work and another to write.

  3. The product around it

    And then there's the product around it, which is where most of the work is. That's the screens your team works in, the connections to the tools you already have and the checks a person makes before anything goes out.

So whoever you talk to about custom AI development, ask them how they'll handle all three.

The AI products we build

  1. 01AI inside your team's software
  2. 02Chatbots
  3. 03Algorithms
  4. 04Products built on your own data

What we build comes in four kinds. All four sit on what your business already knows, and in every one of them, part of the work always stays with your people.

Station 01 · AI inside your team's software

For most businesses, the AI they need belongs in the tools their team has open all day. And honestly, putting it there is most of what we do.

We connect the AI to your own accounts, so it can read the message, pull up the file, draft the reply or answer the question right there. And nobody has to learn a new tool, or remember to open one. Built around one role, it becomes a copilot, getting that person's work ready before they need it.

Maybe your team already pays for an AI tool and uses it by copy and paste, a client's message into a chatbot and the answer back out by hand. This is the step after that, where the AI has the client's whole history and your own rules right in front of it.

Stays with your team

The tools you chose, and the yes on anything that changes a record or goes out to a client.

Station 02 · Chatbots

A chatbot is AI that people simply talk to. On your website, it answers the questions your customers ask every day, at any hour. Inside the business, it answers your team, from where a file lives to how a job gets priced.

A chatbot built for your business answers from your documents, your prices and your rules. When a question needs a person, it passes the conversation to one. And when a job should run from start to finish with nobody pushing it, that's an agent, and you'll find those on our automation systems page.

A chatbot can coach, too. It asks the questions a good mentor would ask, helps someone find what's really going on and walks them through the fix, whether it's nine in the morning or late the night before a hard meeting.

Stays with your team

What it's allowed to answer, and the person it passes a conversation to.

Station 03 · Algorithms

An algorithm is the maths that turns your data into an answer you can act on. It can score your leads by how likely they are to buy, forecast next month's orders, suggest a price or spot the customers who are about to leave.

Some go further and let you rehearse a decision before you make it. A simulation builds a model of the people a decision touches from real data and lets them react to it and to each other. Then it shows you how big the backlash could get and the one change that would soften it most.

Can you imagine getting that read on a price rise before a single customer hears about it?

Stays with your team

The decision itself, and how much weight you give what it shows you.

Station 04 · Products built on your own data

This one sits in the middle of everything above. Your documents, your records, your emails and the way you do things become what the AI works from.

Think about the gap between an AI that writes the way any business would and one that writes the way your best person does. The second one has read your past work, and it can keep learning from what comes back, like which messages get a reply.

That's also what makes it yours. The model underneath can change, but what the AI knows about your business is your own data, in your own accounts. And if that data is spread across a handful of places, it gets sorted before the AI leans on it, which is our data architecture work.

Stays with your team

Your data, in your own accounts, and who gets to see what.

AI we've built, for ourselves and for clients

Some of these are our own products, and the rest are real builds for real businesses. The names are left out, and the numbers come from our own records.

Built

Real rates built into the AI assistant the owner already used

An owner-run contractor

The owner was already asking an AI assistant to write the company's quotes, and it was working the prices out in its head. We kept that assistant and the quote layout the owner already sends, and built in the company's own labour and material rates, with a real calculator underneath to do the maths. It lays each quote out ready to check.

Built

A copilot built around an account manager's week

A digital marketing agency

We built one copilot for an account manager, made of 30 agents around their daily and weekly routines, from sorting the inbox to watching each client's hours and getting the next client meeting ready. Each agent puts its work on the account manager's desk, and a person decides what goes out.

30 agents

Our own product

VAL

An AI mentor for project managers

A project manager describes what's going wrong in plain words. VAL works out the real problem underneath, picks the framework that fits it and walks them through the fix, ending with options for them to choose from. And it remembers every session, so nobody has to explain their situation twice.

Read about VAL

Our own product

Foresay

A prediction engine

You describe a decision in plain English, like a price rise. Foresay builds a population of simulated customers from real demographic and economic data and lets them react to it and sway each other, round after round. Then it hands you a read you can argue with, before you make the call.

Read about Foresay

Our own product

Alma

A self improving sales engine

Alma finds the leads, researches each one, writes the outreach and learns from the replies. It writes from the positioning, the case studies and the tone it's given, and it gives the last name on a long list the same depth of research as the first.

Read about Alma

How we work

Whether it's a new product or AI in your software, it always starts with our free consultation. We study how your business really runs, then hand you the plan, with every problem priced and every fix designed. From there, the decision on what to build is yours, and once something's live, we stay on it and keep improving it.

See how the free consultation works

What you own, and what drives the cost

You own all of it

  • Your data, tools and builds. What we build is yours, so nothing stops working if we stop working together.

  • The report. Before anything is built, you hold the report, with every problem priced and the fixes put in the order to build them.

What drives the cost

We don't publish prices, because every business is wired differently. Four things move the cost of a custom AI solution:

  • The logic. How many rules and exceptions the AI has to handle, and how many working sessions with you it takes to get them right.

  • Your data. How much of it the AI works from, and how tidy it is before we start.

  • The product around it. The screens, the logins and the connections to your tools, and whether those tools let us in.

  • The scale. A tool for a few people on your team, or a product used by thousands of people or more.

Then there's what it costs to run, like the AI usage it needs each month, which grows with the number of people using it, the hosting and any software subscriptions. The report prices every build, to build and to run, next to what it gives back in hours or money, with the working shown.

Two ways to work with us

  1. Project by project.

    You bring the idea or pick one from the report, we build it, and you pay for that project.

  2. An AI subscription.

    Monthly access to the whole team, building your AI one piece at a time and sharpening its logic as people use it.

Is AI product development what you need?

Not every problem in a business needs an AI product built for it. Find the line that sounds most like you, and you'll see what you actually need.

If this sounds like youWhat you need
You have an idea for an AI product, or you want AI working inside the software your team relies on.
An AI product built around your business, and yours to own.
You want work to move from one step to the next on its own, between tools that don't talk to each other.
Automation. It's a separate service of ours, with its own page under What we do.
Your team wants to write and research faster, and nothing has to connect to your systems.
A good AI assistant and a little practice. It's a much smaller job than a build, and worth trying first.
Your data lives in a handful of places, and nobody fully trusts the numbers yet.
Your data sorted first, because an AI built on numbers you can't defend will plan against them. That's our data architecture work.

And if you book a free consultation and we're not the right fit, we'll tell you on the call, and you'll still leave with pointers to get organised and move forward.

Frequently asked questions

The big AI labs build the models underneath, and teams like ours build the software on top of them, connected to a business's own data and tools. Among our own products, VAL is an AI mentor for project managers, Foresay is a prediction engine and Alma is a self improving sales engine. For clients, we've built AI into the tools they already use, from a quoting assistant on the owner's own rates to 30 agents around an account manager's week.

Often, yes, and for writing and research we'd start there. A general assistant can't see your documents, your records or your tools, though, so when it needs them, it guesses. The contractor on this page had an assistant working its prices out in its head, so we built the owner's own rates and a real calculator into it. A build is worth it when the AI has to know your business or live inside your software.

AI integration services put AI inside the software you already run, working with your own data. An AI platform is a tool you buy, and an AI integration company does the work of connecting it to your accounts, setting its rules and checking what it does. It's most of what we build, like the quoting assistant and the account manager's copilot on this page.

Yes. A first version, what people call an MVP, does the one job that matters most, in the hands of the people who'll use it. Then it grows one piece at a time from there. You own that first version and everything after it. And the software around the AI, like the screens and the logins, is our custom software work.

Not usually. We build on the strong models that already exist and give them your own data and rules to work from, which is what most jobs need. People call that LLM application development, where LLM means a large language model. RAG is when the AI looks things up in your own documents before it answers. We pick the model for the job and sometimes mix them, like VAL, which uses one model to route the conversation and another to write.

It stays in your own accounts, and the model only sees what the job needs. In one of our builds, tax IDs, bank details, phone numbers and emails came out before any text reached a model. The link to the client's drive was read only, and the system runs on their own database account.

They shouldn't have to take it on trust, so we build the checks in. A person approves anything that matters before it goes out, a real calculator does the maths instead of the AI, and a figure the system worked out for itself is marked as such on screen.

  1. 01AI inside your team's software
  2. 02Chatbots
  3. 03Algorithms
  4. 04Products built on your own data

So what's the idea you keep coming back to?

Book a free consultationTake the free AI readiness quiz