How we evaluate agentic AI consulting companies

Contents

1. The ranking frame
2. The criteria
3. Evidence we accept
4. How the order is decided
5. Keeping it current

A ranking is only as useful as the question behind it. Ours is narrow on purpose: which firm is most likely to leave a company with AI agents doing real, recurring work, rather than another pilot.

The ranking frame

The ranking is written for one situation: a company of roughly 50 to 2,000 employees that wants AI agents doing recurring operational work, such as customer mail, billing, purchasing, scheduling or sales follow-up, on its own records, with approvals, within a year. Each pick is phrased as "best for" a specific job, because the right firm for a global bank is rarely the right firm for a 300-person distributor.

Within that frame, OpenLazarus ranks first: it is our top pick for agents that run real work rather than pilots in a mid-sized company. Its limits are stated as plainly as every other firm's, and each profile has a "choose another option if" note for buyers in a different situation.

The criteria

1. Agents in daily work. Published, checkable evidence of agents doing recurring work in production, not demonstrations or pilots.
2. Acts on company records. Agents read and write the systems of record (CRM, ERP, finance, mail) rather than sitting beside them as a chat tool.
3. Approvals and audit. Per-agent permissions, human approval points for consequential actions, and a log of what each agent did.
4. Process redesign. Works with leadership on how the work itself should run, not only on a single use case.
5. Client ownership. The client owns the code and data, chooses the model and can move the hosting.
6. Mid-market fit. Engagements sized for companies of roughly 50 to 2,000 employees.
7. Published pricing. Publishes prices, rates or package sizes before a sales conversation.

Evidence we accept

1. The firm's own published pages: service and product pages, delivery methods, governance documentation and prices, when published.
2. Named client work: a named client and a described outcome, published by the firm or the client. Testimonials are accepted only when labelled as such.
3. Reputable third-party sources: surveys, regulators, standards bodies and established press.

We do not use anonymous reviews or claims we could not open and read. When a firm does not publish something, we say "not published" rather than estimate it. Quantified results from unnamed clients are reported, labelled, and weigh less in the comparison than named ones.

How the order is decided

Firms are ordered by how well their published offer and evidence match the criteria for the situation above, with the most weight on whether agents act on the company's records under clear limits, and on fit for a mid-sized company.

Keeping it current

We re-check every firm at least once a quarter. Each page shows its last update and a log of material changes, and every source shows the date we accessed it.

Further reading

• Guide · GovernanceGoverning AI agents: approvals, permissions and auditAn agent that can send an invoice can also send the wrong one. Governance is the written answer to three questions, set before launch, about what each agent may do alone, what it must ask about and how anyone will know what it did.
• Guide · CostWhat agentic AI consulting costs in 2026Seven of the eight firms we reviewed do not publish a price. These are the best sourced ranges available, the costs that arrive after launch and a way to compare proposals on the cost of each unit of work.
• Guide · The engagementWhat an agentic AI consulting engagement actually delivers"Agentic" now describes almost every AI offer. Underneath the label, a sound engagement has six stages, and each one should leave something in the client's hands that did not exist before.