Case study
How M Squared Replaced Off-the-Shelf Software with a Custom AI-Powered Operating System.
M Squared · Landscape & groundworks
M Squared is a Glasgow-based commercial landscape and groundworks contractor, named in the Financial Times’ list of the 1,000 fastest growing companies in Europe and working on projects from the regeneration of Wembley Park to the Barclays campus in Glasgow. Like every growing construction business, the work that matters does not happen inside one team. A project variation starts with whoever spots it on site, needs commercial input before anyone can price it, goes back to the person who raised it for detail nobody else has, and only then goes for approval. A material request follows a different path again, and a large one follows a different path from a small one.
Construction management software on the market models a form, a status and an owner. That is enough for a task. It is not enough for a process that crosses two teams, reverses direction halfway through, and changes route depending on how much money is involved.
So the pattern is familiar: the business buys the tool, then runs the real process in email and spreadsheets alongside it. The tool holds the record of what was decided. It does not run the deciding. Nobody can answer how many variations are open, which are stuck, or who they are stuck with, because that lives in inboxes.
That is the problem we were asked to solve, and it is the reason the answer was bespoke rather than another subscription.
We built M Squared their own AI-powered operating system: a single platform their teams work in, with eight modules on one spine and an AI layer running across all of it.
Operations runs the processes: project variations, contract variations and material requests. Project planning holds the programme. Accountability carries the scorecards. Meetings runs the weekly rhythm. Recruitment runs the hiring pipeline. Finance and delivery cover the commercial and site side. A management dashboard sits across the top, giving the directors one view of the whole business rather than eight.
Because it is one system rather than eight tools, a variation raised in operations knows which project it belongs to, which client that project is for, and who is on the team. Nothing has to be typed twice, and nothing has to be reconciled afterwards.
This is where bespoke earns its money, so it is worth being specific about what the system does that a generic tool does not.
It refuses to start on incomplete information. A run cannot begin until the detail section is filled in. That single rule removes the most common failure in construction admin, which is a request that travels three steps before someone notices the crucial fact was never captured.
Every step is owned by a team, not just a person. Each step in a process carries the team responsible for it, so work moves from site to commercial to approval as the process itself dictates, rather than because someone remembered to forward an email.
Ownership has to be accepted, not assumed. A step can be proposed to a colleague, and it is not theirs until they take it. Work does not get quietly parked on someone who never agreed to it.
Money changes the route. Steps carry value thresholds. A variation worth a few hundred pounds and one worth forty thousand do not take the same path, and nobody has to remember which approval applies, because the process knows.
Work can go backwards. A step can be sent back to the person who raised it. That sounds trivial and it is the single thing most task tools cannot express, which is why the real process ends up in email.
Every run has a reference and a clock. Each process type issues its own reference numbers, and each carries target days and an SLA, so an open variation has a number to quote and a date it is measured against.
Recurring work runs itself. Templates carry a cadence and a next run date, so the processes that happen every week or every month appear when they are due instead of relying on someone remembering.
Recording work is the easy half. The half that changes how a business runs is knowing, without asking anyone, what is drifting.
Underneath the operating system sits a data warehouse. A scheduled pipeline pulls the operational data out of the boards the business already works in, cleans and tests it, and lands it somewhere built for questions rather than for storage. On top of that sits an AI analyst the directors can simply ask: how many variations are open, what is sitting with commercial, how this month compares with last.
The model never does the arithmetic. It writes the query and the database computes the figure. That distinction matters more than it sounds, because it is the difference between a number you can act on and a number a language model has estimated. Every answer traces back to a row.
It watches for work going rogue. The system flags a step nobody has picked up for five days, a run that should have started and has not, and anything past its target date, then escalates to the manager rather than waiting for someone to notice. Each morning a briefing goes out before anyone opens a laptop.
It builds as well as answers. Describe a new process in plain English and the assistant drafts it as a structured flow, with its steps, owners and gates, for a manager to confirm before it exists. Nothing is created without a person agreeing to it.
The same assistant handles the everyday questions the buttons used to: what is in my queue, what stage it is at, what still needs filling in.
The honest comparison is not bespoke against off-the-shelf on price. It is bespoke against off-the-shelf plus the workarounds.
Every business that buys a platform its processes do not fit pays twice: once for the licence, and again in the hours spent bridging the gap by hand. The second cost never appears on an invoice, which is why it survives for years. It shows up instead as the variation nobody can find, the approval that sat in a holiday inbox, and the monthly report that takes two days to assemble because the answer is spread across four systems.
Building it properly means the software matches how the business already works, rather than the business reshaping itself around a product built for someone else. And because the processes are templates rather than code, changing how the business works is a configuration change, not a development project.
It also means M Squared own it outright. There is no per-seat licence that rises with headcount, and no vendor deciding which feature arrives next.
We did not start with eight modules. We started with the process that hurt most, built it to production standard, and let the rest follow on the same spine. Each module after the first arrives faster, because the projects, the people, the permissions and the audit trail are already there.
That is the same approach we take everywhere: prove it on one process, then compound. It is why the recruitment pipeline could be added as a full workflow rather than a feature request, and why the management dashboard could be built over data the system was already collecting.
The work continues. This is a partnership rather than a delivery, and the platform grows as the business does.
See the recruitment pipeline that proved the approach, or what it does with procurement paperwork, or read how AI sits alongside the systems you already run.
Tell us where your team loses the most time. We will tell you honestly whether AI pays there, what it takes to build, and what we have already delivered for businesses like yours.
We use cookies to measure how the site is used and, if you allow it, for advertising. Nothing is set until you choose. Privacy policy.