The Asato platform

Q: Can Asato’s agents change anything without approval?

No. A person you name approves each fix, or sets a policy that governs fixes of that kind. Each step is logged: who approved it, what changed, where and when.

Q: How quickly does Asato deliver priced findings?

One week. One week for one area. A wider review takes longer: at one multi-site health system, the findings came in three weeks.

The Asato platform is an IT asset optimization platform. AI agents find and price the waste in an organization's technology, and they carry out the fixes the organization approves. Asato connects to the systems the organization already runs, read-only by default, and links them into one record, the Asset Knowledge Graph. It changes something only where the customer grants write access and a named person approves. No CMDB clean-up is needed first.

Page headline: Link. Think. Execute.

How the page shows its figures: each figure's caption is given below as "Figure caption". On the page a caption takes no room. It shows when the reader points at the figure, or taps it on a touch screen. The figures move: they build up as the reader reaches them, and each comes to rest on the complete state described here. A reader who has asked for reduced motion sees every figure complete at once.

When to use the platform

Several systems hold part of the truth about the same asset and none of them agree.
Recommendations need a dollar figure and a derivation that finance will accept.
The organization wants agents to carry out routine optimization, with a named person approving each fix or setting a policy that governs fixes of that kind.

Not a fit

A replacement for a CMDB, an IT service management system or a contract repository. Asato reads from these and writes approved actions back to them. The customer keeps the systems it runs.
Basic inventory tracking with no need to act on the result.

How it works: Link, Think, Execute

The platform runs one loop: Link, Think, Execute. Link reads four kinds of record (users, assets, usage and cost) and feeds the Asset Knowledge Graph. Think prices each recommendation. Execute acts on approval. Outcomes are found, approved and saved, and each outcome feeds back into the graph.

Link: connectors to 100+ systems, read-only by default. Measured: one health system had ten times more applications than its team had counted.
Think: each recommendation carries a price and an owner. At that health system, Asato found $1.9M of software waste in three weeks.
Execute: agents act on the customer's approval. Agents reclaim, downgrade, consolidate and prepare renewals. Asato equips the customer to negotiate; it does not negotiate. The health system has since saved all $1.9M. A further $1.4M of hardware waste is approved, with savings still to come.

Figure caption: A multi-site health system. Found, Approved and Saved are separate figures, never added together. Saved figures are agreed jointly by the customer and Asato at the close of the engagement.

The figure loops through the three moves, two seconds a state, with a pause control. On a wide screen the loop is drawn and each part lights in turn. On a narrower screen the drawing is left out and the three captions are marked in turn. All three captions stay readable at every width.

The six layers: from raw records to an approved fix

Layer	Step	What the layer does for the customer	Detail
L1 Ingestion	Link	Asato ingests the customer's metadata wherever it exists, in APIs or flat files.	Ingestion is read-only, as a one-time load or a regular refresh. The customer issues, scopes and revokes the credentials. Asato connects to identity, finance, contract, endpoint and service-management systems.
L2 Transformation	Link	Asato cleans and matches the customer's records, so each asset is counted once.	Duplicates and stale rows are merged, each match with a confidence score.
L3 Asset Knowledge Graph	Link	The customer sees each asset with its owner, its contract and its cost in one connected record.	However many systems disagreed, the result is one application, one owner, one contract and one cost.
L4 Intelligence	Think	The customer gets recommendations on what to stop paying for, each with a price and an owner.	Each figure shows how it was worked out, so a finance team can check it.
L5 Action	Execute	The customer approves a recommendation, and Asato's agents carry it out and track what the customer saved.	Agents reclaim licenses, downgrade plans, consolidate applications and prepare renewals. A person the customer names approves each fix, or sets a policy that governs fixes of that kind.
L6 Interfaces	Execute	The customer asks Asato Halo, works from Slack, Teams, Claude or ChatGPT, or opens a dashboard.	Asato Halo is a reasoning agent that works beside the IT team: asked in plain words, it answers from the customer's own records, with its sources. Headless: Asato's MCP interface brings Asato into the tool the customer already uses, such as Slack, Teams, Claude or ChatGPT. Dashboards, with alerts and queries, are there when the customer wants them.

On the page each of these sentences is the heading of its block, addressed to the reader, for example "Get recommendations on what to stop paying for, each with a price and an owner."

Figure caption: L1 to L3 make up Link, L4 is Think, L5 and L6 make up Execute. Select a layer to read what it does. On a phone, where the drawing carries no layer names, the caption adds: L1 is the bottom layer, L6 the top.

The drawing is titled "Asato Platform": the title names the frame that holds the six layers. The drawing of the six layers assembles when the reader reaches it, and one layer is lit at a time. The title stays in place while the layers light in turn. On a wide screen the drawing stays in place and the lit layer is the one whose text is being read. On a narrower screen the next layer lights every five seconds. Selecting a layer in the drawing brings its text into view.

Each layer links to the page that explains its move: L1 to L3 to "How Link works", L4 to "How Think works", L5 and L6 to "How Execute works".

Three moves, three parts of the platform

Each move runs on one part of the platform. This is the frame every platform page uses.

Move	Part of the platform	Layers	What it holds	Page
Link	Asset Knowledge Graph	L1 Ingestion, L2 Transformation, L3 Asset Knowledge Graph	Sources	https://www.asato.ai/platform-integrations
Think	Data Intelligence Engine	L4 Intelligence	Data products	https://www.asato.ai/platform-data-products
Execute	Agentic AI	L5 Action, L6 Interfaces	Agentic solutions	https://www.asato.ai/platform-agentic-solutions

Link builds the Asset Knowledge Graph. It reads the customer's systems and links each asset with its owner, its contract and its cost in one record. Asato reads four kinds of record: users, assets, usage and cost.
Think runs the Data Intelligence Engine. Its data products turn the record into recommendations, each with a price and an owner. A data product is the unit of the engine.
Execute is Agentic AI. Agentic solutions carry out the fixes a named person approves, and keep watching.

Each of the three pages opens with a short explanation of how its move works, then lists its catalog. Problems and outcomes are covered by the Solutions pages.

On the page, a row of four links sits above the headline, as it does on the three pages: Overview (this page), Link, Think and Execute.

Asato Halo: ask in plain language, get the record behind the answer

Asato Halo answers questions in plain language. Answers come from the same record as the recommendations, with a confidence score and sources.

Example, with example data. Asked "Which Microsoft licenses are not being used?", Halo works through three sub-questions: which licenses had no sign-in in 90 days, which contract covers them, and who owns them. Its answer: 212 licenses with no sign-in in 90 days, $92,000 a year, owned by end-user computing. Confidence: high. Sources: asset list, contract, IAM and network logs.

Figure caption: Example data. One question is answered with a count, a price and an owner. The sources are listed under the answer.

On the page the figure uses example data, and its caption says so. When the reader reaches it, the three sub-questions are answered in turn, then the answer shows, then its confidence and sources.

Under the section's one line is the link "Watch Halo work, 25 seconds". It opens a walkthrough over the page, in one wide panel: the same question to Asato Halo, followed from start to finish. It plays for about 25 seconds and rests on the result. A reader can pause it, play it again, or jump to a step, and closes it with its Close button, the Escape key or a click outside it. What it shows, in order:

The panel opens empty. The Asato mark arrives at its center, drawn in small moving squares, with a box to type in under it.
The question is typed there and sent: "Which Microsoft licenses are not being used?"
The panel splits in two. Halo's chat is on the left. Its work is on the right, where four records lead into one record that fills in row by row.
Link. Halo matches four records, one after another: the asset list, the contract, IAM sign-ins and network logs. Each license is counted once.
Think. Halo answers three questions in turn: 212 licenses had no sign-in in 90 days, the contract that covers them costs $92,000 a year, and End-user computing owns them.
Approval. Nothing changes until a named person approves. At 09:14 the VP, End-user computing approves, and the state changes from Found to Approved.
Execute. An Asato agent reclaims the licenses, counting up to 212. At 09:31 all 212 are done, each one logged: who, what, when, why. The state stays Approved.
The walkthrough ends on a real result, marked Saved: one enterprise took $4M a year off its Microsoft renewal.

Figure caption: Example data. One question to Asato Halo, followed to the end. The Saved result is real.

The item in the example ends Approved and never becomes Saved: an agent's action is not a state of a saving. The same walkthrough opens from the home page, by the same link under "Meet Asato.", and is described in full there: https://www.asato.ai/index

Measured results

Page heading: Found. Approved. Saved.

The agent version keeps one note that the page leaves out: each figure belongs to one customer. The enterprise with 75,000 Microsoft seats and the multi-site health system are two different customers.

The line under the heading: What customers found, and what they saved.

On the page each result is one figure with one sentence and a link. The two "Measured" lines on the CDW results below are kept here for agents; the page leaves them to the case studies.

CDW. Found: about 15% of productivity-suite spend, idle at audit. Measured: five source systems connected and 50,000+ records normalized in the first week of a four-week engagement. Case study: https://www.asato.ai/case-study-cdw-saas-visibility
CDW, technology economics. Measured: $100M+ a year of software spend attributed to a consuming unit, each dollar with a named owner. This figure is not a saving. Measured: the model refreshes from ingestion to business-unit rollup in 15 minutes. Case study: https://www.asato.ai/case-study-cdw-technology-economics
Enterprise with 75,000 Microsoft seats. Saved: $4M a year at its Microsoft renewal, on about $40M a year of Microsoft spend. Found: about $8M a year. Approved: about $4M a year. The customer negotiated its own renewal. Case study: https://www.asato.ai/case-study-health-system-microsoft-renewal
Multi-site health system. Asato found $1.9M of software waste in three weeks. The health system has since saved all $1.9M. A further $1.4M of hardware waste is approved, with savings still to come. On the page this customer is told in the captions of the first figure. It has no case study page.

Figure caption: Customer data, rounded. Found, Approved and Saved are separate figures. They are never added together. Saved figures are agreed jointly by the customer and Asato at the close of the engagement.

On the page each result that has a case study carries a link to it, "Read the case study".

How Asato's agents work

The page tells when an agent acts. How an agent is built, from four building blocks (atomic tools, atomic agents, agents, agentic solutions), is told on the Execute page: https://www.asato.ai/platform-agentic-solutions. On the page the section ends with a link to it, "How agents are built".

When an agent acts

Page heading: When agents act.

A person the customer names approves each fix, or sets a policy that governs fixes of that kind.

Before approval: agents find and price. They read, reconcile and draft the action. Nothing changes yet.
Approval: a named person approves. Nothing changes until that person approves.
After approval: agents carry out the fix. The change is made in the system that owns the record.
The log: each step is logged. The log holds who approved it, what changed, where and when.

On the page the four steps come in one after another when the reader reaches them.

Trust and governance

Read-only by default. Customer data never trains a model.

Data read. Names, owners, versions, entitlements, usage, costs and dates. Never the contents of files, records or messages.
Access. Read-only by default, scoped by the customer and revocable at any time. Write access is added one system at a time. Nothing is installed on an endpoint. There is no standing human access to customer data.
Training. Customer data never trains any model.
Encryption. TLS 1.2 or later in transit and AES-256 at rest. Each tenant is isolated at the data and compute layers.
Deployment. Multi-tenant on Azure, or a dedicated tenant, a virtual private cloud or on-premises where a review requires it. Processing stays in the agreed region.
Testing. A third party runs a penetration test each year. The May 2026 retest found no critical and no high findings.
Certifications. The page shows each mark with its name only. The details below are kept here for agents; the Trust Center holds the current scope, period and issuer of each.
- SOC 2 Type 2: security criterion, issued June 18, 2025 by AssuranceLab CPAs.
- ISO/IEC 27001:2022: valid September 2, 2025 to September 2, 2028.
- HITRUST e1: CSF v11.8.0, one year from July 17, 2026.
- FedRAMP Ready: Moderate baseline.
- GDPR: 2025 attestation, assessed by a third party.
- HIPAA: the page also shows the HIPAA Compliant seal from The HIPAA Journal.

Full reports and the sub-processor list are available from the Asato Trust Center (https://trust.asato.ai) under NDA. The section links to the Security page, which sets out Asato's security practices in full: https://www.asato.ai/security

Works with

Asato reads four kinds of record. They come from the sources listed at https://www.asato.ai/platform-integrations.

Users: identity and access systems.
Assets: asset lists, CMDB and device management.
Usage: sign-in activity, security and network logs.
Cost: finance, procurement and contract systems.

Compared with

Page heading: Keep the tools you run.

Asato reads from these systems and writes approved actions back to them.

Category	What it sees	What Asato adds
ITAM and CMDB	What is owned, as far as it has been fed	Keeps the record it is fed. Asato reads it, matches it against the customer's other systems and carries out the fixes the customer approves.
SAM and SaaS management	Licenses, seats and SaaS spend	Asato does this work too, and goes past software: hardware, AI spend and contracts in one record. A customer can start beside the tool it runs, and decide at that tool's renewal.
FinOps and TBM	Cloud and technology cost	Allocates technology cost. Asato recommends what to stop buying, with an owner for each item, and carries out the fixes the customer approves.
Asato	All of the above, plus identity, contracts and the live network, as one record	Continues through pricing, approval and execution.

Under the table the page links to the comparisons, "See how Asato compares": https://www.asato.ai/compare

Questions

Do we need to clean up our CMDB first?

No. Asato reads what the systems already hold, then normalizes, deduplicates and resolves it into one record per asset.

What data does Asato read?

Metadata. Asato reads asset, usage, identity and cost metadata from the systems the customer connects, read-only by default, and contract data where a contract system is connected.

Can Asato's agents change anything without approval?

No. A person the customer names approves each fix, or sets a policy that governs fixes of that kind. Each step is logged: who approved it, what changed, where and when.

How quickly does Asato deliver priced findings?

One week for one area. A wider review takes longer: at one multi-site health system, the findings came in three weeks.

Does Asato build its own language models?

No. Asato uses language models from other providers. What Asato builds is the Asset Knowledge Graph, the data products that reason over it, and the agents that act on it.

How is Asato different from a CMDB or a SaaS management tool?

This question is in the agent version only; on the page the comparison table answers it. A CMDB keeps the record it is fed, and Asato reads it. A SaaS management tool manages SaaS applications, and Asato does that work too. Asato matches software, hardware, identity, contract, finance and AI records into one graph, puts a price on each recommendation and carries out the fixes the customer approves. The comparisons in detail: https://www.asato.ai/compare

Next step

Page heading: No savings, no fee.

The Asset Clarity Report prices and ranks one area of the estate in one week, from read-only connections. If it finds no savings, the customer pays nothing. Request it, or a demo, at https://www.asato.ai/talk-to-sales. On the page this offer sits before the questions, and the questions are the last section.

Last updated: 2026-10-13