Score your context
Rate each dimension from 1 (low) to 5 (high) based on your organisation's current situation. The calculator weights all factors and produces a tailored recommendation.
How to read the sliders: Each dimension is worded so that a high score (5) represents a stronger need in that direction. For example, "Urgency" at 5 means time-to-value is critical — which favours Copilot.
Microsoft Copilot
—
Bespoke LLM app
—
Adjust the sliders above to see your recommendation
Decision signals
Quick heuristics for identifying which path fits — before running the full scoring model.
Choose Copilot when…
Primary use is productivity in M365 apps (Teams, Word, Excel, Outlook)
Workforce is large and AI literacy varies widely
Time-to-value is critical — weeks, not months
Data remains comfortably within Microsoft's trust boundary
Budget is per-seat opex rather than capex project
No dedicated ML/AI engineering capacity in-house
Use cases are general: drafting, summarising, searching
Compliance needs are met by Microsoft's BAA / DPA
Build bespoke when…
Use case is domain-specific or proprietary (e.g., legal review, clinical triage)
Workflow requires deep integrations with proprietary systems or databases
Fine-tuning or RAG on a private corpus is needed
Organisation has AI/ML engineers or a capable partner to engage
Competitive differentiation depends on unique AI capability
Regulatory constraints require custom data handling or on-prem deployment
User experience must be embedded in existing internal tools
Scale warrants capex investment over per-seat licensing
Hybrid path: Many organisations deploy Copilot for broad productivity and build a small number of bespoke agents for high-value, domain-specific workflows. These are not mutually exclusive — the question is often sequencing, not choice.
Dimension matrix
A side-by-side comparison across the key decision dimensions.
| Dimension | Microsoft Copilot | Bespoke LLM app |
|---|---|---|
| Time to deploy | Days–weeks Licence, configure, roll out via M365 admin | Months Design, build, test, deploy, and iterate |
| Upfront cost | Low Per-seat licence (~£25–30/user/month) | High Engineering, infra, and ongoing maintenance |
| Customisation | Limited Plugins, connectors, and prompt configuration only | Full Any model, any UX, any logic, any data source |
| Data control | Managed Stays within Microsoft's commercial cloud boundary | Full control Choose cloud provider, on-prem, or hybrid |
| Integration depth | Deep (M365) Shallow for non-Microsoft systems | Any system Via APIs, webhooks, or custom connectors |
| Model choice | Locked GPT-4o via Azure OpenAI — no alternatives | Open Any provider (Anthropic, Google, Meta) or open-source |
| Maintenance burden | Low Microsoft owns infrastructure, updates, and uptime | High Your team owns infra, model versioning, and bugs |
| Competitive moat | None Same capability as every other Copilot customer | High Proprietary IP, logic, and trained capability |
| Governance & audit | Moderate Logs available; limited guardrail configuration | Full control Custom guardrails, audit trails, and RLHF policies |
| User adoption | Easier Familiar M365 surface; low friction onboarding | Requires effort Change management and training needed |
| At-scale economics | Linear cost Grows with headcount indefinitely | Improves at scale Fixed infra cost amortised over usage |
Feature comparison
High-level profile of each option at a glance.
Microsoft Copilot
Best for broad productivity
Fastest path to AI-assisted work across the whole organisation
Deployment speedDays–weeks
Cost modelPer seat (opex)
Customisation ceilingLimited
Data residencyMicrosoft cloud
Engineering requiredLow
Model lock-inYes
Competitive moatNone
M365 integrationNative & deep
Bespoke LLM app
Best for domain-specific value
Purpose-built workflows, proprietary data, and competitive differentiation
Deployment speedMonths
Cost modelCapex + opex
Customisation ceilingUnlimited
Data residencyYour choice
Engineering requiredHigh
Model lock-inNo
Competitive moatHigh potential
M365 integrationCustom build needed
Guidance notes
Practical context for each scoring dimension — what it means and why it matters.
Urgency / time-to-value
How quickly does the organisation need demonstrable AI capability? Copilot can be live in days post-licensing; bespoke builds rarely deliver production capability in under three months. If leadership urgency is high, Copilot buys breathing room while a bespoke strategy is designed.
Domain specificity of the use case
Generic productivity tasks — drafting emails, summarising documents, generating first drafts — are well-served by Copilot. The moment a use case requires specialised reasoning (e.g., reading a pathology report, interpreting a legal clause, analysing a structural design), generic models hit a ceiling and domain-specific fine-tuning or RAG becomes necessary.
Need for deep system integrations
Copilot integrates natively with M365 and has a growing connector ecosystem. But if the AI needs to read from or write to a proprietary ERP, a legacy database, or a custom internal platform, a bespoke build gives you full control over the integration layer. This is often underestimated at the outset.
M365 ecosystem dependency
Organisations that live and work inside Teams, SharePoint, Outlook, and Office benefit disproportionately from Copilot because the AI is embedded directly in the tools people already use. Organisations with non-Microsoft productivity stacks capture far less value from the licence.
In-house AI/ML capability
A bespoke build without engineering capacity is a liability, not an asset. Before committing to a custom route, honestly assess whether the organisation has (or can engage) the MLOps, prompt engineering, and software engineering talent to build, deploy, monitor, and maintain the system over a multi-year horizon. A failed bespoke project often poisons future AI investment appetite.
Available capex budget
Bespoke builds typically require a six-figure minimum investment for anything production-grade. Copilot is opex, predictable, and approvable through procurement cycles. At large scale (thousands of users), bespoke economics can eventually surpass per-seat licensing — but only after the break-even horizon, which is typically 18–36 months post-launch.
Data sensitivity and sovereignty
Highly sensitive data categories — patient records, legally privileged documents, trade secrets, regulated financial data — may require you to control exactly where data is processed and stored. Copilot processes data within Microsoft's commercial cloud under their DPA; bespoke allows you to choose your own cloud region, on-premises infrastructure, or a hybrid model.
Need for competitive moat
If an AI capability is core to your organisation's value proposition or competitive strategy, licensing the same platform as every competitor provides no advantage. The only path to proprietary AI capability — trained on your data, fine-tuned for your domain, embedded in your product — is a bespoke build. This dimension should carry significant weight in strategic planning.
The most common mistake
Organisations either over-index on Copilot — "quick wins, low risk, just roll it out" — and find the ceiling quickly when real workflows demand depth; or they leap to bespoke builds without the engineering capacity to sustain them.
The robust default is a hybrid approach: deploy Copilot for broad organisational productivity, and build bespoke for the one or two strategic workflows where domain-specific AI creates genuine competitive or operational leverage. These paths are not in competition — the question is usually sequencing and prioritisation, not either/or.