SAMSOFT builds FinOpsAI — one view of Google Cloud, AWS and Azure spend for finance and engineering teams: anomalies, commitments, waste, rightsizing and CFO-grade reporting, plus Kubernetes, carbon and AI cost intelligence, and an A2A agent that explains what it finds and fixes it with your approval. Customer data stays resident in India.
finopsai.app is one workspace, one identity, one team and one billing relationship, with two products on top of the same billing data: Cloud Cost Intelligence, available now, and AI Cost Intelligence, in preview.
Google Cloud, AWS and Azure spend in one view: dashboards, anomalies, budgets, commitments, waste, Kubernetes, licences, carbon, executive views and reports — with an A2A agent that explains and, with approval, remediates.
What your AI spend buys: Vertex AI and Gemini usage by model, token and caching mode; direct LLM API spend from vendors; seats and gateways; allocation to teams and products; unit economics; GPU hardware; and savings plans that stay honest about what is known.
How it connects: the customer grants a per-workspace reader identity on their Google Cloud billing export; AWS through a CloudFormation reader role on the Cost & Usage Report; Azure through a consented sign-in to the cost exports. No service-account keys, no stored credentials, nothing installed in the customer's environment. Figures are shown as billed, after credits, in each billing currency; a figure that cannot be known yet is shown as "—", never as zero.
The dashboard shows what was billed, after credits, in each billing currency. Windows of 7, 30 and 90 days or month-to-date; split by provider, project or service; forecast from run-rate once there is enough history.
A daily audit of each cloud finds the spikes that matter and explains each one in plain language: which service, which project, since when, and what changed. Assign, close, or reopen an anomaly with regression detection.
GKE, EKS and AKS cost by cluster and node pool — control-plane fees, node compute, GPU nodes, persistent volumes and load balancers — with agentless requests-versus-usage so you can see the slack before buying more capacity.
Illustrative figures.
What FinOpsAI reads, where it sends findings, and what is on the roadmap. Each item is marked live, in preview, or planned.
Roadmap — API usage
Roadmap — seats & subscriptions
Roadmap vendors are metered through usage exports and seat billing as each vendor exposes them; waste and savings rules (idle seats, caching, batch, model right-sizing) apply uniformly once a vendor is live.
Logos: official partner and product marks are added under each vendor's brand guidelines once the respective partner programme assets are in place.
What runs where. Customer billing data stays in the customer's own cloud; FinOpsAI reads it with a granted identity, aggregates in BigQuery in asia-south1, and serves the product from Cloud Run.
The services and products FinOpsAI runs on, reads from, or integrates with.
Grouped the way the product is: monitor, optimise, govern, report.
Billed cost after credits by provider, project or account, and service; 7-, 30-, 90-day and month-to-date windows; day or month grain; tax and refunds toggle; CSV export. Forecast from the last seven complete days once there is enough history. Data-freshness indicator per cloud.
Per-project and per-account detail for all three clouds: spend trend, top services, resources, commitments in scope, and the account's connection health and export table.
A daily anomaly audit per cloud with floor resolution so small projects do not drown in noise. Each anomaly is explained, assigned, closed or reopened with regression detection. An ensemble model runs in shadow and is promoted only when it beats the baseline on your data.
Budgets per workspace, project or account with threshold and forecast alerts; generic alert rules across modules; delivery by e-mail, webhook, Microsoft Teams or Jira.
Natural-language questions over your own cost data — "what drove the 18 % rise in May?" — answered from the ledger with the figures cited, never from a guess.
What FinOpsAI can see for each connected cloud: which exports, recommenders and telemetry are granted, and which features are therefore live, partial or unavailable — stated, not assumed.
Committed-use discounts, reserved instances and savings plans across clouds: coverage, utilisation, discount rate and effective savings rate; expiry calendar; buy / hold / let-lapse recommendations with break-even utilisation; realised-savings close that measures reductions actually achieved.
Billing-derived rules plus each cloud's native recommender: idle and orphaned disks, addresses, snapshots, load balancers and network attachments; non-production resources running out of hours; telemetry-backed idle compute where metrics are granted. Each finding carries its quantified saving and the rule that produced it.
GKE, EKS and AKS. Cost by cluster and node pool — control-plane fees, node compute, GPU nodes, persistent volumes, load balancers. Overhead ratio, GPU share, fee dilution and capacity variation. Agentless requests-vs-usage-vs-allocatable at cluster and node-pool grain; pricing levers (committed, spot, machine families); storage-class choices; efficiency scoring and recommendations K-1 to K-6. Namespace allocation and shared-cost rules are the next tier.
Rightsizing candidates from recommender data and telemetry, with the saving computed on one documented formula set and the confidence stated; remediation hands off to the agent with approval.
Seat and subscription waste for the SaaS and AWS Marketplace subscriptions on your bill: unused seats, duplicate subscriptions, renewals due — priced on the bill basis.
Every optimisation FinOpsAI can identify, on one page, with its data requirement and whether it is live, partial or not yet verified for your clouds.
CEO, CFO and CTO views on the same numbers: spend under management, savings history, commitment position, risk, and a Savings Audit Report that shows its arithmetic. A CFO pitch and payback generator for business cases.
The emissions each cloud publishes for your own usage — Google Cloud Carbon Footprint, AWS Customer Carbon Footprint Tool and Azure emissions exports — ingested, never estimated, with the method named (location- or market-based). A BRSR financial-year table (April to March) with CSV export, with purchased cloud noted under Scope 3 category 1. "Not connected" and "awaiting data" are shown as such.
Monthly report e-mails per workspace, the Savings Audit Report, ledger exports, and realised-savings statements suitable for finance review.
Company workspaces with invitations and domain join; four roles; per-cloud scoping; ownership transfer; SAML single sign-on and SCIM provisioning for enterprise identity providers (in live test with Okta).
Residency posture per workspace, retention and deletion controls, an audit event stream of every material action, and a compliance page that states what applies to your data.
Workspace API keys and a stable public API (v1) over the same handlers the app uses; outbound webhooks with signed envelopes; Microsoft Teams and Jira delivery; the A2A agent card for agent-to-agent use.
AI spend is the fastest-growing and least-explained line on the cloud bill. AI Cost Intelligence treats it as a product of its own, on the same workspace and billing data.
Vertex AI and Gemini SKUs classified by model, modality and serving mode — on-demand, provisioned throughput, batch — with token counts and context-caching where the billing export carries them, and an honest coverage statement where it does not.
Usage pulled from LLM vendors' own usage endpoints for the APIs you call directly, so the AI picture is complete rather than Google-Cloud-only.
AI cost allocated to teams, products and features; cost per request, per thousand tokens, per active user; trend against volume.
Seats and gateways, budgets and alerts per team, parameter overrides for your own prices and assumptions, and who changed what.
Rule-based recommendations: caching candidates, batch-eligible workloads, model right-sizing flagged "needs quality check" until you validate a trial, and GPU hardware utilisation.
A savings plan built from the recommendations you accept, and commitment options for provisioned capacity, each with its break-even stated.
Not promised: vendor rate-limit headroom (not exposed by any vendor), inference quality (never claimed), or prompt content (usage only, in line with DPDP).
FinOpsAI publishes an agent card at /.well-known/agent.json following the A2A protocol, so other agents and tools can ask it questions. A separate remediation agent runs inside the customer's own cloud identity and applies approved actions.
Read-only skills over the same figures the app shows: what changed, why, and what FinOpsAI can and cannot see for each cloud. No new numbers are invented for the agent path.
Recommendations become proposed actions; a named person approves; the remediation agent executes under the tenant's identity with OIDC audience checks; the result is verified (for example, the instance is confirmed terminated) and written to the audit trail.
No autonomous production change, ever. Every infrastructure action has a human approver, a reason, and a reversible record — the pattern regulated customers already require.
The Co-Sell workbench lets a seller or the customer drop an AWS Cost & Usage Report or an Azure cost export into the page and get an anonymised Google Cloud target projection: equivalent machine shapes, pricing, commitment opportunity and sovereignty surfaces. Nothing is uploaded; there is no IAM role, no storage grant and no network call on that path. It removes the security review that usually sits between a seller and a first conversation about numbers.
Available on Google Cloud Marketplace; every FinOpsAI subscription bought there counts toward the customer's Google Cloud commitment.
One control set, applied to every vertical. We state plainly what is in place today and what is still being built.
| Area | In place today | Planned |
|---|---|---|
| Data residency | Production in asia-south1 (Mumbai). Customer billing data stays in the customer's own cloud project; FinOpsAI reads, aggregates and stores derived figures. | Regional deployments for EU and GCC customers with the Regulated-Global edition. |
| Access | Keyless by design: a per-workspace reader identity the customer grants on their dataset; AWS via a CloudFormation reader role; Azure via consented sign-in. No service-account keys or stored credentials. | SAML SSO and SCIM for enterprise identity providers (in test). |
| Change control | The remediation agent executes approved actions only, with an audit trail; human approval on every infrastructure change. | Maker-checker roles for regulated editions. |
| Privacy | Designed for India's DPDP Act 2023: billing data, not personal data; no personal identifiers kept from Marketplace signups; retention and deletion controls per workspace. | Independent security assessment (VAPT) before the regulated editions. |
| Platform security | Cloud Armor in front of the API, OIDC-verified messaging, tenant isolation on every route, secrets in Secret Manager. | Formal certifications are not yet held; we will not claim them until they are. |
For teams that want help getting to the numbers, our architects work alongside the product.
A two-to-three-week review of your Google Cloud, AWS and Azure spend: savings plan, commitment strategy and waste, with FinOpsAI as the working tool and the evidence you keep.
Target architecture and landing zone for workloads moving to Google Cloud — including SAP, Oracle and VMware estates — with a cost model and a commitment plan from day one.
Vertex AI and Gemini adoption with cost governance built in; A2A agent integration for your own tooling.
Essential edition, monthly. Every plan includes all three clouds, the A2A agent and a 14-day free trial. Prices in INR for India (GST applies) and USD elsewhere.
Up to 3 cloud projects or accounts, 3 users, 500 AI queries a month, 6-hour data refresh.
Up to 8 projects or accounts, 8 users, 2,000 AI queries a month, 90-day history, Teams and Jira delivery.
Unlimited projects, SSO and SCIM, VPC isolation options, 99.99 % SLA, custom terms. Regulated editions are quoted separately.
Full plan details at finopsai.app/pricing. Trial data is destroyed at the end of an unconverted trial.
SAMSOFT AI Technologies is a private limited company incorporated in Chennai in 2026, with an 18-member team across Chennai, Bengaluru and Hyderabad in India and New Jersey in the United States. Our leadership team is below.
Leads the company: business operations, finance, legal and compliance, and company strategy. Point of contact for customers and partners.
[email protected]Over 35 years building cloud-scale enterprise systems across compute, network and storage infrastructure. Most recently Senior Director of Research at VMware (Broadcom), leading research at the forefront of enterprise infrastructure. Brings that systems depth to SAMSOFT's technical direction.
LinkedIn →Enterprise architect with 23 years across BFSI, pharma, manufacturing and global institutions — most recently Enterprise Architect at Lloyds Technology Centre, earlier Associate Director (Journey to Cloud Advisory) at KPMG and architecture roles at Wipro Digital, Accenture and HCL. Led the modernisation and cloud migration of 530+ applications for a global investment bank. Certified on Google Cloud, AWS and Azure (Professional Architect), Kubernetes (CKA), Terraform and TOGAF. Clients have included UBS, CRISIL, Pfizer, State Bank of India and the World Bank / IFC. Builds FinOpsAI.
LinkedIn →Start a free 14-day FinOpsAI trial, ask about an assessment, or reach us as a Google Cloud or AWS partner.