Microsoft Copilot vs Custom AI
Copilot Is Already Inside. That Is the Advantage and the Homework.
If your business runs on Microsoft 365, Copilot arrives with something no custom build can cheaply replicate: it can already reach your mail, your files, your chat and your calendar, with no integration project at all. It works by seeing what the person asking is already allowed to see — which is exactly the right design, and which means the thing that decides how your rollout goes is not the AI. It is whether your file permissions are what you think they are.
Microsoft Copilot vs A custom build
Reach inside your tenancy · Reach you define
- What it can reach on day one
Your mail, files, chat and calendar. No integration work at all.Nothing, until you connect it - How it decides what to show someone
By the access that person already has. Correct by design.By rules you write down - What that depends on being right
Your permissions. Most have not been reviewed in years.The scope, which is written once - Systems outside Microsoft
Limited. Your CRM and your ops tools are mostly not in there.Wherever there is an API - Who it works for
A person, at the moment they ask it something.A process, on a trigger - Setup effort
Licensing and enablement, not building. Genuinely fast.A project, with a scope and a cost - What it will not do
Follow your rule instead of its own general behaviour.Anything you did not specify
Row three is the one to act on before you buy anything. Copilot answering from a document somebody should never have had access to is not the AI leaking data — it is the AI finding a permission that was already wrong, quickly and at scale. Every organisation we have looked at has some of these, usually from a folder shared broadly during a busy month in 2021. Reviewing them is a week of unglamorous work and it is the highest-value thing you can do before a rollout.
If you live in Microsoft 365, start with Copilot
For a business already deep in Microsoft 365, the honest recommendation is to start there and be sceptical of anybody selling you a build instead. Summarising a thread, drafting from a document, finding the file somebody mentioned in a meeting — that reach is real and a custom build cannot approach it at a comparable cost, because you would be paying to rebuild an integration Microsoft has already done.
The limit is not quality, it is direction. Copilot answers the person in front of it about the things Microsoft holds. Most of the work businesses actually want automated crosses out of that: the enquiry that arrives in a web form, the record in a CRM nobody has migrated, the supplier portal, the scheduling system a department bought in 2019. A tool built to be excellent inside one estate is not the thing that joins four of them together, and that is not a criticism — it is a description of what it is for.
The second limit is that it assists rather than runs. It is superb when somebody asks it something and reads the answer. It is not the thing that handles an enquiry at 2am to a rule you wrote, applies your pricing logic, and escalates the case that does not fit. That is a different job, and wanting the second is not a reason to skip the first.
So the sequence we would actually recommend, in order: audit your permissions, roll out Copilot, and only then look at what is still slow. What survives that is a much smaller and much better-specified list than the one you started with — and it is a list where a custom build is genuinely worth the money rather than a way of buying enthusiasm.
Copilot surfacing a document somebody should not have seen is not a leak. It is a permission that was already wrong, found quickly.
Where each one actually belongs
The dividing line is where your work lives, not how advanced you want to be.
The work lives in Microsoft
Mail, documents, meetings, chat. Copilot, and it will be hard to beat. This is a large share of ordinary office work and it is what the product is for.
The work crosses systems
A form, a CRM, a supplier portal, a scheduling tool. Nothing inside one vendor joins those, and that is the territory a build earns.
It has to happen unattended
If nobody is there to ask, an assistant does not help. Something that runs on a trigger and follows your rule is a different category of thing.
The rule is specific to you
Your pricing logic, your escalation thresholds, your definition of an urgent case. General-purpose tools are general on purpose.
Your permissions are unaudited
Do this first regardless of which way you go. A week of tidying before any rollout, and it is worth doing even if you then buy nothing.
Both, in that order
Common and sensible. Roll out Copilot, see what is still slow after three months, and build only that. The list gets much shorter and much clearer.
What people ask about this choice
Can Copilot see files it should not?
It is designed to see only what the person asking already has access to, which is the right design. The practical issue is that in most organisations some of those permissions are wrong — a folder shared broadly years ago and never revisited. Copilot does not create that problem, it finds it quickly. Auditing permissions before a rollout is the single most useful preparation.
Will Copilot automate our processes?
It assists a person who asks it something, which is a different job from running a process unattended. If the thing you want is an enquiry handled at 2am to a rule you wrote, with the odd case escalated, that is not what an assistant does. Wanting that is not a reason to skip Copilot — the two do not compete.
Can it work with our CRM?
Mostly not in the way people hope. Its reach is strongest across the Microsoft estate, and the systems businesses most want joined up — a CRM, a scheduling tool, a supplier portal — usually sit outside it. That gap is where a custom build has a genuine case rather than a sales case.
Should we buy Copilot or build something?
If you are already deep in Microsoft 365, start with Copilot and be sceptical of anyone selling a build instead — you would be paying to rebuild integration Microsoft has already done. Then look at what is still slow after a few months. That much shorter list is where building is worth it.
Is a custom build better than Copilot?
Not better, aimed differently. Copilot is built to be excellent inside one estate and it is. A custom build is for work that crosses estates, runs unattended, or follows a rule specific to your business. Comparing them on quality misses that they answer different questions.
What should we do first?
Audit permissions, roll out Copilot, then look at what is still slow. That order matters: the audit protects the rollout, and the rollout tells you what genuinely needs building rather than what sounded impressive in a meeting.