Head-to-head · updated 13 September 2026

DataShield vs Profisee: whose master data can prove what the agent did?

Profisee is a good MDM product. The mid-market likes it. You get golden records for customers, products, suppliers and locations. You also get a stewardship screen your business users will really open. Then there is the Microsoft story. Nobody else in master data management can match it. A native Fabric workload. A certified Microsoft 365 Copilot agent. A Power Platform connector. Marketplace billing against your Azure spend. They are a Leader in the 2026 Gartner Magic Quadrant. Since March 2026 they also ship an MCP server. So "we have MCP and they don't" is a line we are not going to sell you.

We build master data too, and we build it agent-first. An agent talks to your business users. It builds a spec, then walks a decision gate with you. One transaction then makes the entity types and hierarchies. It also makes draft match configs, reference bindings, and validation rules. That's 143 of our 447 MCP commands, for mastering alone. Under that sits the part suites tend to skip. A golden record has to earn the right to be read by an agent. We re-check each checked call against the agent's current access. Each decision gets sealed into a hash chain. You can verify it without trusting us. We tokenize datasets at ingest. Agents query tokenized data over MCP. Detokenization is a checked, special step. Here is the honest scorecard, rows they win included.

Get free Demo Center access Get an instant quote

The short version

Pick DataShield when

You want the model built in a conversation, not in a services engagement. An agent talks to your team, builds a spec, works a dependency-ordered decision gate, then makes the entity types, hierarchies, draft match configs, reference bindings, and validation rules in one transaction. See the ontology.
You want a golden record an agent can only read once it has earned it: a composite quality score over six axes, every axis above its floor, and a consent check that fails closed. One predicate runs it in SQL and in code.
An agent is going to read your master data, and someone will later ask you to prove the access log was not edited. Our chain answers with math, not a policy PDF. Run the verifier.
You need to pull an agent's authority mid-session and have the very next tool call fail. Not the next token refresh. The next call. How Auth does it.

Pick Profisee when

You live in the Microsoft stack. Fabric workload, OneLake, Copilot, Power Platform, Azure Marketplace billing. Nobody in MDM has built that seam deeper, and we have not built it at all.
Business stewards, not engineers, will run the day job. Their stewardship UI is better than ours, and I would say that in front of a customer.
You need many domains, many source systems, and a partner bench to implement it. Their site says customers manage 100M+ golden records; we publish no record-count ceiling and won't invent one.
You need SOC 2 and HIPAA attestation on paper today. Their pricing page states both. We are not certified yet.

Bottom line: Profisee sells you a model, a stewardship screen and a bench of people to configure it. We let an agent build the model with you, then refuse to hand any golden record to an agent until it clears a quality and consent gate you can inspect. If your MDM buy is really an AI governance buy in disguise, that difference decides it.

Feature by feature

Competitor cells describe what Profisee's public site and press releases say. This is as of the date above. Did we get something wrong? Email support@myorg.ai. We will correct it, and credit you.

What matters	DataShield	Profisee	Edge
Standing up the model	An agent does it with you. The onboarding engine talks to a business user. It builds a spec. Then it presents a decision gate in dependency order. Each proposal states what it does, why, the impact, and the alternatives. You confirm or override. An override then propagates to each dependent decision. One atomic transaction then makes the entity types, hierarchies, and draft match configs. It also makes reference bindings and validation rules. Each object lands draft, labelled with its gate posture. The generator never promotes a config. It never runs resolution. It never skips the battery. 15 commands, no implementation services.	Aisey, rebuilt in July 2026 on the Microsoft Agent Framework. It orchestrates expert agents for modelling, connections, matching, quality, and stewardship. Profisee says an admin can upload a Word doc of needs. They claim a working setup in about ten minutes. Roughly 90% of setup, they say, is automated. That is their claim, not a number we have tested.	◆
Entity resolution and matching	Fellegi-Sunter probabilistic record linkage. Comparators are Jaro-Winkler, Levenshtein, Soundex, and Double Metaphone. Add LSH blocking, EM-trained m and u parameters, and five survivorship strategies. Add drift monitoring too, on each promoted match config. "Most trusted" reads measured overturn rates in stewardship. It is not a hand-ranked source list. An absolute-identifier tier routes identifier conflicts to review, instead of merging them. No config reaches production without a green paired-gate battery run. That run goes through the real scoring path. Batch and on-demand, not a sub-second stream engine.	Mature matching and survivorship. Plus AI vector matching, added in March 2026, that pairs records by meaning. Profisee states matching at hundreds of millions of records, with real-time monitoring.	◈
Multidomain modelling	An open entity-type vocabulary. You model your domain. We don't ship a template you have to bend. You get three matching modes per type. The People, Places and Things taxonomy is table rows, not a fixed list. Hierarchies are parent-scoped. A child config resolves inside its parent's scope at a base threshold. Across parents, it resolves at a stricter one. Relationships carry bitemporal validity. Each domain has an owner and a compliance rulebook. Validation rules gate publication. What we don't have is a packaged Customer 360.	Unlimited domains and attributes at each tier. A long history of customer, product, supplier, and location models. A bench of partners who have built them before. Breadth and packaging are theirs.	◇
Ingestion and binding	Six on-ramps: inline ingest, URL ingest, batch ingest. Also presigned object storage, a file repo, and a catalog bind. The bind is the mastering-specific one. Point a live PostgreSQL object at an entity type once. Mastering then stays current on a set cadence. Remaster policy is auto, agent review, or steward review. Differential loads mark deleted source records stale. Updated ones get flagged for a second look. A weekly full pass reconciles deletes. That is PostgreSQL today, not your whole estate. We are not an ETL product.	Connect, plus the Microsoft connection surface. Plus a partner network. It has wired the usual source systems many times over. Profisee's YMCA case study cites 29 source systems connected. Broader source coverage than ours, by a wide margin.	◇
Stewardship experience	A stewardship dashboard with a flow canvas and a per-gate workbench. Also a records surface and an alerts queue. It is plainer than theirs. It is MCP-first. So an agent can work the same queue a human does.	The stronger product for a business steward. Aisey now sets up a match rule or a model from a chat prompt. Profisee states its AI quality agents cut manual stewardship work by 30 to 50%.	◇
Earned automation and the trust ladder	Stewards review patterns with evidence. One human decision then disposes a pattern, not just one item. A three-rung trust ladder sets how much runs unattended. 50 approvals can promote a rung, at a 98% rate. That needs zero dissents, though. One dissent demotes it. At the top rung, a fixed sampler diverts 10% of auto-disposed items to human audit. This lets you prove why any item was sampled. A standing rule is born staged. It needs a second, clear release. It auto-suspends on an audit overturn. Each dry run uses the writer's own gate code. A preview can never say yes and then refuse.	Automation via Aisey and their quality agents. Governance guardrails are described, but only in general terms. We found no public method for earned, revocable automation. Nor for sampling what the automation decided. Ask to see one.	◆
Golden records for agents	A golden record has to earn the right to be read. The admission gate wants a composite score. It must be at least 0.70, across six quality axes. Each axis has to clear its own floor. A consent check fails closed. Consent covers all six GDPR Article 6(1) lawful bases across nine purposes. No logged basis means no permission. One predicate runs the gate, in SQL and in TypeScript. Reads come from versioned views that mask unconditionally. The reader role has no grant on the base table. So the mask cannot be bypassed from SQL either. Grounding documents for RAG come only from agent-ready goldens. A record that leaves the gate has its chunks deleted.	Golden records are served to Copilot, Claude, and ChatGPT through the MCP server. We found no data-quality check on an agent read. No consent check either, in their public docs.	◆
Microsoft ecosystem	None. We are not in Fabric. Not in Copilot. Not on Azure Marketplace.	A native MDM workload inside Microsoft Fabric, with OneLake and SSO. A certified Microsoft 365 Copilot agent. A Power Platform connector. An Azure Marketplace listing.	◇
Audit evidence	SHA-256 hash chain with Ed25519-signed checkpoints. The checkpoints are chained too. The check names the failure: tampering, insertion, deletion, or truncation. Inside MDM, this applies to each merge, unmerge, pin, and admin override. Each lands on a fixed record. It shows before and after. That trail sits in the same transaction as the change. Try the verifier.	Transaction audit trails. The 2026.R2 release says the MCP server has built-in login checks and audit. We found no public tamper-evidence method. Ask to see one.	◆
Agent authorization	Every checked tool call passes a scope ceiling. It also passes an access tier, and a revoke check. This happens before dispatch. The call fails closed. Delegation is RFC 8693 token exchange with an enforced scope ceiling.	Login checks on the MCP server. Aisey runs inside what they call governance guardrails. We could not find a per-call authorization decision point in their public docs.	◆
Break-glass	Scoped, time-boxed emergency access for agents. It auto-revokes. It cannot be quietly deleted from the log. Mastering runs have their own brake. A canary verdict runs before any ramp. Scope-keyed pause and kill flags apply, and a global kill stops the fleet. Microbatch rollback goes to a pre-image marker.	Not named in their public docs.	◆
GDPR erasure and consent	Crypto-shred of per-subject key material, plus ISO 27560 consent receipts. Add a consent-validity record per subject and purpose under the golden. Only the derived consent score crosses the masking boundary. The consent facts themselves are not granted to the view reader. The change stream is a value-free CloudEvents outbox. That is why it needs no purge cascade. The audit chain still checks out afterwards.	Governance rules and audit trails. The erasure method is not named. That matters a lot in MDM, where the golden record is the last copy.	◆
Tokenization and classification	We use 129 field classes, covering PII, PHI, financial data, and secrets. They come with checksum checks and reproducible verdicts. Tokens at ingest are fixed, join-preserving, and vault-reversible. We generalize quasi-identifiers too. Dates go to year, decade, or age band. ZIPs go to 3 or 4 digits. Phones, SSNs, and emails get cut down to a part. Each column gets a measured cardinality-reduction score. Derived phonetic and n-gram fields inherit the PII verdict. They take it from the field they came from. So a name cannot be rebuilt from its Soundex. Masking is a switch you turn on. It is not a default.	Not part of the product, as stated. Data quality rules and REGEX checks are not the same job.	◆
MCP and agents	More than 200 MCP tools across Ontology, Auth, Corpus, and Lighthouse. Of those, 143 commands are for mastering alone. That's 99 on mdm, 29 on stewardship, 15 on guided onboarding. Golden records are served behind masked views. Per-call metering ties back to the agent. No REST API for goldens. It is MCP, a masked SQL view, and a DuckDB dimension.	A real MCP server since March 2026. It reaches Copilot, Claude, ChatGPT, and custom agents. It was extended in July 2026 across Matching, Connect, FastApps, Forms, and Presentation Views.	◈
Deployment	Self-hosted, in your own cloud or data center. Or a dedicated single-tenant server we operate. Docker images cover Auth, Ontology, Corpus, and Lighthouse. Guardian checks a signed deploy file. Ed25519 audit-signing keys stay on your instance, in a key folder you control. Kid-keyed rotation fails closed on an unknown key. KMS or HSM custody is a built seam in Auth, not a shipped connection. HMAC tokenization keys sit in your own setup today, not in a KMS.	Managed SaaS on their Azure, with a 99.8% uptime promise. Or a containerized Kubernetes service you run on Azure, AWS, GCP, or on-premises. Genuinely flexible.	◈
Maturity and certifications	Founded recently, small team, live in production. Guardian and Lighthouse have run since April 2026. SOC 2 not yet certified, and we say so. We checked results against roughly 370,000 real public records, across three datasets. We publish no record-count ceiling. Design-partner terms include source escrow.	Founded 2007. A 2026 Gartner Magic Quadrant Leader. A long logo list from Blue Cross Blue Shield to AB InBev. SOC 2 plus HIPAA and HI-TECH stated on their pricing page.	◇
Pricing	Published model, scoped instant quote, no sales wall.	The model is public and fair. Pay by record volume. Unlimited domains and attributes. Bulk rates at scale. The numbers are not public. You still have to call.	◆

◆ DataShield leads◇ Profisee leads◈ comparable

Profisee claims come from profisee.com and Profisee's own press releases. Last checked 13 September 2026. We link them below rather than work from memory.

Three things you get here that you won't get from an MDM suite

An agent builds the model with you

A suite sells you a model and a bench of people to configure it. Here an agent talks to your business users. It builds a spec. It hands you a decision gate in dependency order. Confirm or override each one. Overrides propagate. Then one transaction makes the entity types and hierarchies. It also makes draft match configs, reference bindings, and validation rules. All of it lands draft and gate-labelled. See the ontology.

Golden records earn the right to be read

An agent asks for a customer golden. It gets nothing unless the record clears a composite score. That score spans six quality axes. Each axis must clear its floor. A consent check fails closed if there is no logged lawful basis. One predicate runs that, in SQL and in code. The reader role has no grant on the base table. How we hold the line.

Proof that survives an audit

A log that can be quietly edited proves nothing. Ours is a hash chain with signed checkpoints. The verifier names what broke, not just that something did. That is the property EU AI Act Article 12 and HIPAA §164.312(b) reviewers ask about. Pair it with access that can be pulled mid-session. Try the chain in your browser. No signup.

Where Profisee is genuinely stronger

Let me be direct about the gap. Profisee has shipped MDM since 2007. It shows. Two editions. A clean upgrade path for the Microsoft Master Data Services estate. Unlimited domains at one price. Stewardship screens a data steward can learn in an afternoon. The Microsoft seam is the real moat. A native Fabric workload, built on the Extensibility SDK. A certified Copilot agent. A Power Platform connector. Billing through Azure Marketplace. Say your CIO has an Azure commit to burn. That shortens a buying cycle by months. They also move fast. Aisey shipped in August 2025. An MCP server followed in March 2026. A rebuilt Aisey landed on the Microsoft Agent Framework in July 2026. Profisee says an admin can upload a Word doc of needs. They say a working MDM setup follows ten minutes later.

Our onboarding is a conversation, not a ten-minute upload. Their partner bench is real. Ours is a roadmap.

Here is the push-back. Their 2026.R2 notes say the MCP server ships with built-in login checks and audit. Those are the right words. They are also the easiest words in security to say. A login check happens once, at the start of a session. That is not the same as re-checking access on each call. An audit trail is a table. A chain is different. An outside party can check a chain after the fact. Fast setup is not the same as gated setup. Ask what stops a generated match config from going live. This is before it has been trained and battery-tested. Ask the three questions we ask ourselves. What happens on the tool call after you revoke an agent? Can you prove a row of the audit log was not deleted last quarter?

What quality and consent test must a golden record pass? This is before an agent is allowed to read it. If those answers are thin, you now own a very good golden record. You just have no evidence about who read it.

Questions worth asking both of us

These are the questions we would want answered if we were the ones buying. Ask them on each call, ours included.

Can you cryptographically prove an audit log entry wasn't deleted?

DataShield: yes. Each record commits to the one before it. Checkpoints are signed and chained. The check tells deletion apart from truncation and from tampering. Inside MDM, each merge, unmerge, pin, and override lands on a fixed before-and-after trail. That trail sits in the same transaction as the change. Try it against a sample chain at /verify. Profisee: their material describes transaction audit trails and audit on the MCP server. We found no tamper-evidence method published. Ask them to show one.

What happens to a revoked agent mid-session?

DataShield re-checks access on each checked tool call. So a revoke lands on the next call, and it fails closed. A mastering run has a second brake. Scope-keyed pause and kill flags apply, and a global kill stops the fleet. Add microbatch rollback to a pre-image marker. Profisee's MCP server states built-in login checks. We could not find a mid-session revoke method in their public docs. Ask how long a compromised agent keeps reading golden records. That's after you pull its access.

How does GDPR erasure interact with the audit trail?

This bites harder in MDM than anywhere else. The golden record is often the last surviving copy of a person. DataShield crypto-shreds per-subject key material and issues an ISO 27560 consent receipt. Actor identities in the chain are HMAC-committed. The evidence still checks out once the subject is gone. Under the golden sits a consent record per subject and purpose. It covers all six GDPR Article 6(1) bases. The change stream carries no values at all. So there is nothing downstream to purge. Profisee describes rules and audit trails. Ask for the erasure method, not the workflow.

Profisee ships an MCP server too. What's really different?

Theirs is real. It landed in March 2026. By July it reached Matching, Connect, FastApps, Forms, and Presentation Views. It works from Copilot, Claude, and ChatGPT. Good work. Two differences. Scale of surface: mastering alone is 143 of our 447 MCP commands. That includes a guided onboarding engine. It takes a business user from interview to a generated draft model. Depth under the call: tool tokens carry a scope ceiling. The call gets checked before dispatch. The decision gets sealed into a chain. The golden also has to pass the admission gate before it is returned. If you want an agent to read master data, either will do. Want to prove later what it read, and why that was allowed? Ask both of us for the record.

Aisey can configure an MDM environment in about ten minutes. Can you?

That is their claim, and I would not bet against a demo of it. Ours works differently. The agent talks to your team. It builds a spec. Then it makes you walk a decision gate in dependency order. Each proposal states what it is doing, why, the impact, and the alternatives. There is deliberately no accept-all button. Generation happens in one transaction. Everything lands draft. A match config still needs training. It needs a green paired-gate battery. It needs a clear promotion before it touches a production record. Slower on the clock. But you can show an examiner who decided what. Say Fabric is your data plane and Copilot is your interface. Their seam is worth real money. We cannot match it.

Does DataShield have SOC 2?

Not yet, and we will not imply otherwise. Profisee states SOC 2 and HIPAA on its pricing page. We do not. What we offer instead: a public threat model, and a verifier anyone can run. Also code you can host yourself. Guardian and Lighthouse have run in production since April 2026. Design-partner terms include source escrow. A small vendor is then not a single point of failure. Details on the security page.

Profisee's MCP Server connects governed master data to Microsoft Copilot, Claude, and Gemini. AI vector matching pairs records by meaning. CTO Eric Melcher: "Our MCP Server transforms Profisee from a system that data teams manage into a governed data backbone". - Profisee 2026.R1 release, 26 Mar 2026
Aisey rebuilt on the Microsoft Agent Framework to orchestrate specialist agents. MCP server extended across Matching, Connect, FastApps, Forms, and Presentation Views. It has built-in authentication and audit. Profisee states matching at hundreds of millions of records. They also claim a working setup in roughly ten minutes. - Profisee 2026.R2 release, 8 Jul 2026
Native MDM workload inside Microsoft Fabric, with OneLake integration and SSO. Built on the Fabric Extensibility SDK. - Profisee press release, 26 Feb 2025
Aisey introduced as an always-on MDM assistant with self-running agents. Surfaced in Microsoft Teams, Copilot, and Copilot Studio. - Profisee blog, 11 Aug 2025
Pricing is volume-based and works across domains, with unlimited domains and attributes. The page states customers manage 100M+ golden records. It also states a 99.8% SaaS uptime promise, and SOC 2 plus HIPAA/HI-TECH certification. No dollar figures are public. - profisee.com/pricing, 13 Sep 2026
Homepage claims Leader placement in the 2026 Gartner Magic Quadrant for Master Data Management Solutions. - profisee.com, 13 Sep 2026

Other head-to-heads

MDM

Get free Demo Center access

You've seen the proof

Ready for a number? Scope your deployment and we'll price it against your own economics.

Get your quote →