Copilot or ChatGPT Enterprise is often enough for individual work: drafting, summarising, searching your own files. You need a custom AI workflow when the work crosses several systems, a person must approve the output and every action is logged, the inputs are Arabic documents, or hosting rules decide where data is processed. Many companies need both, plus training.
#Key takeaways
- General assistants are good at helping one person draft, summarise and find things.
- They wait to be asked; they do not sit inside a workflow and move work from one system to the next.
- Low usage after a rollout is usually a training and workflow problem, not a tool problem.
- Custom workflows fit work that crosses systems, needs approval and a log, reads Arabic or has hosting rules.
- Before expanding the rollout or building anything, ask what your people actually use and for which tasks.
#What are Copilot and ChatGPT Enterprise good at?
Helping one person do their own work faster. Writing a first draft, summarising a long thread, pulling points from a meeting, searching documents the person can already open, rewriting in a different tone. For knowledge workers this is real, everyday value, and both products are built for it.
Both vendors also publish how they handle company data. Microsoft says Copilot only surfaces organizational data to which individual users have at least view permissions (Microsoft Learn, accessed november 2026). OpenAI says business data in ChatGPT Enterprise, including conversations and files, is not used to train its models by default (OpenAI, accessed november 2026).
#Why does adoption stall after a Copilot rollout?
Buyers describe the same pattern: Copilot rolled out to thousands of employees, a small fraction using it, and staff still opening ChatGPT on their own. The most common explanation is that the tool does not insert itself into the workflow. It waits to be asked.
The Work Trend Index found that 75% of knowledge workers use AI at work and 78% of AI users bring their own tools, yet only 39% of AI users had received AI training from their company (Microsoft and LinkedIn Work Trend Index, may 2024). People need to see the tool used on their own tasks. That is a training problem first, which is what AI training at Stennir is for.
#When do you need a custom AI workflow?
- the work crosses systems. an invoice arrives by email, is checked against the ERP, approved by finance and posted. A general assistant helps one step; a workflow runs all of them.
- someone must approve and everything must be logged. regulated or high-stakes output needs a named approver and an audit trail.
- Arabic documents. scanned Arabic invoices, bilingual contracts and Gulf dialect customer messages need testing on your own inputs.
- hosting rules. if your rules say where data must be processed, check whether the assistant meets them for that use.
#What should you ask before deciding?
These are the questions buyers raise most, and the answers depend on your own setup.
- Which tasks do people actually use the assistant for today, and which do they avoid?
- Who in our tenant can already open files they should not, such as HR folders, and would an assistant make that easier to find?
- Where is our data processed for each tool, and does that meet our hosting rules?
- Which of our workflows cross more than one system and need an approver?
We are
Stennir trains teams to use the assistants they already have on their real work, and builds custom workflows only where the work crosses systems, needs approval, reads Arabic or has hosting rules.
We aren't
Stennir does not tell you to replace Copilot or ChatGPT Enterprise, and it does not resell either product.
#What should a CIO or head of operations do next?
Measure current usage before expanding the rollout or building anything. Then pick one workflow that crosses systems and test whether a custom build is needed. See AI implementation, AI training and AI for operations, read what an AI agent development company builds, then book a 30-minute call.
#Common questions
Should we cancel our Copilot or ChatGPT Enterprise rollout?
Usually not. The assistant is useful for individual drafting and search. Low usage is more often a sign that people were never shown how to use it on their own tasks. Train first, measure usage before and after, then decide which workflows need something more.
Can a custom workflow use the same model as our assistant?
Often, yes. A custom workflow usually calls a commercial model through your own account. The difference is not the model but what surrounds it: the connections to your systems, the approval step, the log and where the processing runs.
Does it work in Arabic?
General assistants handle Arabic conversation reasonably well. Arabic documents are a different test: scanned invoices, mixed-language contracts and Gulf dialect messages. Stennir tests on your own documents with a human reviewer before any workflow runs without close supervision. Other languages on request.
What do you need from us to start?
Usage data from your current assistant if you have it, a list of the workflows that take the most time, one owner per workflow who can approve output, and your hosting rules. From that we tell you what to train, what to build and what to leave alone.
