Services · Outcome-based managed service

Keystone AI: build, govern, operate, optimize.

An enterprise AI governance and factory managed service. From zero to governed, production-ready AI in six months, with continuous assurance after that.

Book a discovery call All services

AI GovernanceAI Asset LifecycleAI FactoryAI Observability

The problem

AI in BFSI is scaling faster than its controls.

€35M

Maximum EU AI Act fine, or 7% of global turnover if higher

15 days

EU AI Act deadline to report a serious AI incident

Recommendations in RBI's FREE-AI framework (2025)

No single inventory

Models, GenAI apps, agents and MCP tools run with no register of owner, purpose or risk tier.

Approvals without gates

Reviews live in email and spreadsheets; nothing stops an unapproved model going live.

Every team builds differently

No shared RAG, agent or guardrail patterns; every use case starts from zero.

No runtime evidence

Drift, prompt injection and incidents surface at audit, with no trail back to the approval.

The result: AI delivery slows, audit findings pile up and regulatory exposure grows with every new model.

Source: EU AI Act (Regulation 2024/1689) Art. 73 and 99; RBI Committee report on the FREE-AI framework (Aug 2025).

What Keystone AI delivers

Four layers, one accountable managed service.

Keystone AI = consulting + ControlPlaneX platform + managed operations.

AI Governance

Policies enforced, not filed

AI Constitution and policies

Mapped to RBI, SEBI, DPDP, EU AI Act, ISO 42001

Risk tiers and review board

ControlPlaneX: constitution, rulebook

AI Asset Lifecycle

Every asset owned and gated

Inventory of models, agents, MCP servers

Gated intake to retirement

Re-review on drift or scope change

ControlPlaneX: registers, approvals

AI Factory

Governed use cases, shipped faster

RAG, agent and MCP patterns

Templates with guardrails built in

Evaluation and red-team in CI/CD

ControlPlaneX: approved catalog

AI Observability

Runtime proof for auditors

Gateway policy, budgets, kill switch

Drift, guardrail and incident alerts

Audit packs and board posture

ControlPlaneX: telemetry, audit packs

Outcome: governed, production-ready AI in six months, with continuous assurance after that.

Why Keystone AI

One partner, one contract, one result.

	Consulting	Tool only
Policy and operating model		—
Controls enforced at runtime	—
Integrated with your estate	Partly	Partly
Audit evidence by default	—	Partly
Run and reported monthly	—	—
Fees tied to outcomes	—	—

What changes for you

Policies become runtime controls, not documents

Evidence is a by-product of daily work

One team owns design, build and run

Part of our fee is at risk against your KPIs

We build ControlPlaneX and run the service, so accountability never splits.

ControlPlaneX · Functional architecture

One governed lifecycle, one system of record.

Propose

Asset intake, rules and risk tiering.

Validate and review

Model validation, asset validation and AI review.

Approve

Board and quorum, a decision gate and conditions.

Deploy

Production onboarding, provisioning and policy push.

Assure

Monitoring, re-certification, obligations and audit packs.

Five-pillar AI review: agents advise, a named human decides, hard gates block. Approval compiles to gateway policy; drift and incidents trigger re-review.

ControlPlaneX · Solution architecture

Every AI call is governed and leaves audit evidence.

Clients reach LLM providers, MCP servers and enterprise tools only through the AI and MCP gateway, so only approved models, tools and servers are reachable.

Gateway controls

Policy and guardrails

AuthN/Z and identity

Routing and orchestration

MCP proxy and tools

Observability and audit

Central registers

Use caseAI systemModelDatasetProviderPromptsAI SBOMAudit log

After go-live

Drift and scope creep flagged automatically

Incidents tracked against reporting windows; contain in one click

Risk across the AI estate rolled up in one board posture view

AI Factory

The AI Factory turns governance into a fast lane.

Each pattern is built once, approved once in ControlPlaneX and reused by every team.

Use cases

Credit decisioningFraud detectionCustomer servicingOps copilots

Factory patterns

RAGAgentic workflowsMCP tool serversEvaluation harnessGuardrails

Platform foundation

Model catalogAI gatewayIdentityCI/CDObservability

Pick a pattern

From the approved catalog.

Build from template

Guardrails and logging built in.

Evaluate in CI/CD

Quality, bias and red-team suites.

Pass the ControlPlaneX gate

Dossier pre-scored by AI agents.

Deploy via gateway

Policy, budgets and kill switch applied.

Plus enablement: playbooks, training and pairing.

How we deliver

ControlPlaneX is the platform; we make it work.

Design

AI policy, risk tiers, review board and RACI, built with your risk and compliance teams.

Approved operating model

Deploy

Install ControlPlaneX in your environment; configure roles, SSO and data residency.

Production ControlPlaneX tenant

Integrate

Connect the AI gateway, model registries, IAM, CI/CD, ITSM and SIEM.

Policy enforced at runtime

Instrument

Instrument apps and agents; stream runtime events to ControlPlaneX as evidence.

Live telemetry and alerts

Operate

Run reviews, monitor controls, handle incidents and report to the board.

Monthly assurance pack

No hand-offs between vendor and integrator: one team designs, deploys and runs it.

Commercial model and outcomes

Four stages, paid on acceptance, measured by KPIs.

Stage 1 · Weeks 0-6

Establish

OUTCOME

Target architecture, AI inventory, risk taxonomy, governance model, roadmap

ACCEPTED WHEN

CIO and CRO sign-off

Stage 2 · Weeks 6-16

Build

OUTCOME

ControlPlaneX deployed, gateway integrated, factory templates and controls live

ACCEPTED WHEN

Environment passes UAT

Stage 3 · Weeks 14-24

Activate

OUTCOME

First wave of use cases onboarded with monitoring and evidence

ACCEPTED WHEN

Agreed use cases live

Stage 4 · Month 6 onward

Operate

OUTCOME

Governance reviews, control reporting, incident handling, tuning

ACCEPTED WHEN

Monthly KPI scorecard

Month-six KPI targets

100%

AI inventory

In-scope assets registered

Ungated deployments

Live assets not approved

10 days

Review cycle

Standard-risk decisions

100%

Runtime coverage

Live traffic via gateway

24 hrs

Incident triage

Reportable AI incidents

Monthly

Assurance pack

To the risk committee

Timelines and KPI targets are indicative and fixed in the statement of work; we do not contract to eliminate all AI risk.

Commercial model

Fees that follow outcomes.

Establish · Build · Activate

Milestone fees

Fixed price per stage, invoiced on acceptance of the stage outcome.

Operate

Managed service fee

Monthly fee by tier, sized by the number of AI assets under management.

Operate

Outcome-linked fee

10-15% of the monthly fee at risk against agreed KPIs, earned on meeting targets.

Platform

ControlPlaneX licence

Annual subscription, priced by tier and deployment model.

Pricing is shared in discovery and fixed in the statement of work. Gateway, cloud and model-provider charges are outside Keystone AI fees.

Next steps

Start with a three-week discovery.

Week 1

Workshops

With CIO, CISO, CRO, model risk and engineering leads to agree scope and priorities.

Week 2

AI estate baseline

Inventory sample and gap assessment against RBI, SEBI and DPDP expectations.

Week 3

Stage 1 proposal

Fixed outcomes, acceptance criteria, KPIs and price for Establish.

You walk away with

AI estate baseline report

Regulatory gap heatmap

Target operating model outline

Stage 1 statement of work with fixed price

Decision needed: name an executive sponsor and confirm discovery dates.

Book a discovery call founder@StratifyIT.ai