Copilot frustration on Reddit: “not embedded” and overpromised results
A punchy thread on Reddit argues that Microsoft Copilot is being pushed into enterprises with unrealistic claims and thin integration.
“Can we all agree COPILOT is crap”
“It’s not even embedded… in Excel, SharePoint, Power BI… so people don’t understand why it can’t do anything.”
It’s a familiar story: a vendor or partner pitches sweeping efficiency gains, but teams open their everyday tools and don’t see anything different. Expectations sky-high, outcomes ambiguous, and staff left wondering what they’ve paid for.
Let’s unpack why this happens, what’s fair criticism, and how UK organisations can make sensible, measurable progress without buying into hype.
Why Copilot can feel underwhelming in enterprise rollouts
It is not “embedded by default” in Excel, SharePoint, or Power BI
The Reddit post’s core point is fair. If users expect Copilot to light up across Excel, SharePoint, and Power BI from day one, they’ll be disappointed. In practice, people have to know where to find it, how to invoke it, and what it can and cannot access. Without clear enablement, it feels invisible.
Marketing promises can outrun the operational reality
“100% efficiency everywhere” is a recipe for disillusionment. AI tools are assistive. They can draft, summarise, and accelerate routine tasks, but they don’t fix broken processes or dodgy data. If your SharePoint is a permissions maze or your spreadsheets aren’t consistently structured, Copilot won’t save the day.
Change management and discovery are non-trivial
If people don’t know when to use Copilot or what good prompts look like, adoption stalls. Many rollouts front-load licences but under-invest in training, guidance, and simple examples tailored to each team. The result: “It can’t do anything” becomes a self-fulfilling perception.
What the tech actually needs to be useful
Even without getting into product minutiae, three practical truths apply to Copilot and most workplace AI:
- Access follows permissions. If the AI doesn’t have a clean line of sight to the documents, emails, or data you expect it to use, it won’t be helpful.
- Structure matters. Unstructured knowledge sprawled across random sites, folders, and chats limits retrieval. Better information architecture pays dividends.
- Specific workflows beat vague ambition. “Improve productivity” is not a use case. “Draft first-pass client emails from call notes” is.
Pragmatic steps to get ROI from Copilot in UK organisations
1) Start with three concrete, repeatable use cases
- Summarise meetings with action items and owners.
- Draft first versions of documents or emails from bullet points.
- Generate short research briefs pulling from approved internal sources.
Keep the scope narrow and tie each use case to a baseline metric (time spent, lead time, or rework). Measure before and after.
2) Align information access and governance
- Sanity-check SharePoint and Teams permissions. If access is chaotic, clean-up first.
- Label and organise high-value content. Make the “good stuff” discoverable and current.
- Avoid feeding sensitive or regulated data into prompts unless your governance covers it. UK teams should complete a DPIA and consult data protection officers where needed.
3) Enable the right entry points and make them obvious
- Show staff where Copilot lives in the apps they already use. Quick loom-style walkthroughs beat long PDFs.
- Create a short internal “playbook” with two or three prompt templates per team.
4) Train for judgement, not just prompts
- Teach people to verify outputs, cite sources, and avoid over-trusting confident text.
- Have a red/amber/green policy for which content can be used with AI drafting.
5) Measure, iterate, expand
- Run a 6–8 week pilot with a small cohort. Track usage, time saved, error rates, and satisfaction.
- If the numbers don’t stack up, adjust the use cases or pause. Don’t scale failure.
Where Copilot can add value today
Based on the concerns raised, it’s worth stating where many teams do see benefits when expectations are grounded:
- Summarising large documents and meetings to speed up catch-up and delegation.
- Drafting first-pass content that humans then edit, improving throughput for comms-heavy roles.
- Helping non-experts get started with formulas or content structures, reducing blank-page time.
These are “assist” scenarios, not fully automated tasks. They shine when content is clear, sources are accessible, and the team has a habit of reviewing outputs.
When to look elsewhere or supplement Copilot
If your core workflows sit outside Microsoft 365, or you need very specific automations, you might supplement with targeted tools. For example, teams that live in Google Sheets sometimes connect a model directly to a spreadsheet for structured tasks. I’ve written a simple guide on that approach here:
How to connect ChatGPT and Google Sheets (custom GPT)
Equally, some processes are better served by traditional automation (scripts, scheduled jobs, data quality checks) than by a general-purpose assistant. No AI tool will compensate for inconsistent data or unclear ownership.
UK-specific considerations: privacy, cost control, and vendor pressure
- Data protection: ensure lawful basis for processing, document a DPIA, and set clear rules for handling special category data. Review guidance from the UK ICO.
- Contracting and procurement: pressure to “keep up” can lead to premature scaling. Pilot first, buy later. Insist on measurable outcomes in any vendor-led rollout.
- Change impact: unions and staff councils may have views on AI use. Engage early, define boundaries, and focus on augmentation rather than replacement.
Bottom line: the Reddit criticism has a point—outcomes depend on fit and execution
The thread highlights a real gap between sales promises and day-to-day utility. Copilot is not magic, and it is not omnipresent by default. If you take the time to target a few solid workflows, tidy up access, train people, and measure impact, you can get value. If you expect a drop-in 100% efficiency boost, you’ll be disappointed.
For current capabilities and deployment guidance, see Microsoft’s official documentation:
And for context, here’s the original Reddit discussion that sparked this piece:
