The short answer
We do not publish prices: "an AI assistant for our staff" can mean light configuration or a substantial build. Drivers:
- Use-case clarity, teams touched
- Data readiness
- Systems to connect
- Accuracy needed, cost of a mistake
- Data sensitivity
- Metered usage
Two buckets: one-time work to assess, prove, build and launch; ongoing costs to run, monitor and support. A trustworthy proposal shows both, states its assumptions and lets you stop between phases.
Where the one-time cost goes
| Component | Covers | Bigger when | Smaller when |
|---|---|---|---|
| Discovery and readiness | Use cases, data, systems, permissions, risks | Many teams, unclear goals, regulated | One clear problem, a deciding sponsor |
| Proof of concept | One use case on your data, agreed thresholds | Vague measures, scattered data | Real examples, one source |
| Build and integration | ERP, Microsoft 365 or CRM connections; interface; permissions | Several systems, no APIs, write-back | Read-only, one documented system |
| Data preparation | Removing outdated and duplicate content, structuring | Conflicting versions, scans, no owners | Current sources, named owners |
| Evaluation and testing | Test sets, accuracy, permission and security tests | High targets, outputs affecting people | Narrow, easy-to-check task |
| Security and governance | Risk and data protection review (a DPIA where required), logging, prompt injection tests, policy | Sensitive data, data sent abroad, agents that act | Internal, non-sensitive content |
| Change management | Training, champions, procedures | Many users, new ways of working | Small team, tool fits current work |
The AI model barely appears. Most one-time effort sits in data, integration, testing and controls: exactly where the cheapest quotes cut corners.
Ongoing costs, and how they scale with volume
Model usage
Metered per token, request, page or document; rates vary by model and deployment, and change often.
Spending limits
OWASP lists "unbounded consumption" as a risk; every production design needs limits and alerts.
Cloud hosting
Search indexes, databases, storage, hosting, logs; a deployed fine-tuned model bills even unused.
Licences
Per user, not volume: Copilot is an add-on; Copilot Chat comes with eligible plans.
Exchange rates
Most AI and cloud subscriptions are dollar-priced, so naira running costs move monthly: review regularly.
Monitoring and support
Smaller than the build, never zero: accuracy sampling, source updates, re-testing, incidents, user help.
Self-hosted models
Fixed GPU cost plus upkeep; pays off at high, steady volume or strict data limits.
Who gets a Copilot licence first is a budget decision, so our Copilot service starts with a pilot. Servers or per-token fees? See self-hosted models versus AI APIs.
Vendor pricing details and sources
- Microsoft: the billing unit and rate can vary by model, deployment type and meter; AI service charges are only part of total solution costs, and a deployed fine-tuned model incurs hosting charges even when unused (Microsoft Learn).
- Anthropic, for example, publishes separate per-million-token rates for input and output, with a discount for batch work that can wait (Anthropic). Check each vendor's current pricing.
- Microsoft 365 Copilot is generally sold as an add-on licence for users on an eligible Microsoft 365 plan; web-grounded Copilot Chat is included with eligible subscriptions (Microsoft Learn).
- Unbounded consumption is OWASP's term for excessive or malicious use that runs up the bill (OWASP).
- Support covers re-testing when a vendor retires a model. Self-hosting means paying for servers busy or idle, plus the effort to patch, monitor and update them, instead of per token; it suits steady, high volumes or data that must not leave your environment.
Budgeting a pilot versus production
| Proof of concept | Production | |
|---|---|---|
| Scope | One use case, small user group | All intended users and sources |
| Data | Approved sample, de-identified where possible | Live, permissions enforced |
| Integration | Minimal or read-only | Supported, approvals for write-back |
| Usage cost | Small, measured per transaction | Measured cost × realistic volume, peak headroom |
| Result | Go or no-go against thresholds | A supported service with a runbook |
Budget the proof of concept as one decision and production as a second, informed by its measurements. Microsoft's guidance agrees: test a small representative workload and compare actual charges with your estimate before rollout. Thresholds: see running an AI proof of concept.
Hidden costs to plan for
Data clean-up
Three versions of one procedure produce confident, wrong answers.
Permissions
AI surfaces overshared files; fixing access is often the first Copilot rollout cost.
Integration
API access can depend on subscription tier; older systems may need work first.
People's time
Experts write tests and judge answers, an owner decides, pilot users join. Never invoiced.
Re-testing and exit
New models and documents mean re-tests; contract for ownership of prompts, tests and integrations.
Rather talk it through? If you need a realistic AI budget for leadership, we can scope your use case and give you phased estimates in naira (or US dollars where a licence or service is priced in dollars), with one-time and ongoing costs separated. Talk to a Promatics specialist
How Promatics structures an AI proposal
A written scope
Use case, users, data sources, systems, success measures, assumptions and exclusions.
Phases with gates
Assessment, proof of concept with thresholds agreed first, then build; stop at any gate.
Costs separated
Build effort apart from usage, hosting, licences and support; usage estimated from measured results.
Estimates in naira
Or US dollars where a licence or service is priced in dollars; vendor charges listed separately and marked.
Support on clear terms
Service agreement with written response targets; 24/7 monitoring and support for managed-service clients.
Each phase is estimated separately; the assessment is yours to use with any provider. Vendor subscriptions and usage are listed apart because you pay them to the vendor.
When to bring in help
A Copilot Chat trial or an AI feature you already pay for needs no consultant: read the licence terms, check pricing, set a data policy. Bring in help when AI touches personal data, financial records or decisions about people, sends data to an AI service abroad, connects to ERP, CRM or Microsoft 365 data, needs a board-ready budget, or your quotes cannot be compared. Our AI strategy and readiness service gives a scored use-case shortlist with build and running cost ranges. Comparing providers? See choosing an AI partner.
Before you approve an AI budget
- Use case, users, success measure, monthly and peak volumes written down.
- Each data source has an owner, and its quality is known.
- Systems to connect are listed, read-only or write-back.
- The accuracy threshold, and who judges it, are agreed.
- Personal and sensitive data identified; data protection review (DPIA where required) planned.
- One-time and ongoing costs are shown separately, with dollar-priced items marked.
- Cost per transaction comes from a measured test, with spending alerts planned.
- Internal time from experts, owners and pilot users is budgeted.
- Support, monitoring and re-testing after launch are priced.
Sources and further reading
Product capabilities and guidance change. These are the primary sources this article relies on, checked on the review date above.
- Plan and manage costs for Microsoft Foundry, Microsoft Learn
- License options for Microsoft Copilot, Microsoft Learn
- Pricing, Anthropic (Claude Platform Docs)
- LLM10:2025 Unbounded Consumption, OWASP Gen AI Security Project
This article is general information, not legal, accounting or security advice for your specific situation. Examples are hypothetical unless stated otherwise.