Agentic process automation companies: top 10 for 2026

Ten agentic process automation partners scored on process results, method, fit with existing automation, compliance and capacity, with sources.

Nicholas Berryman Content Lead and Market Analyst February 11, 2026 Updated November 5, 2026 17 min read

Agentic process automation companies: top 10 for 2026

This ranking scores 10 agentic process automation companies: firms that build AI agents into a client's business processes and run them there. An agent in this sense reads an incoming case, such as an invoice, a claim or an order, checks it against the systems the process already uses, resolves it or prepares it for a person, and records what it did. Each firm is scored from 0 to 4 on five public criteria: process evidence, design and evaluation method, fit with existing automation and deployment, certification and capacity. Research was done on October 30, 2026, from each firm's own pages and filings.

The candidate pool starts from the 26 providers in Everest Group's first Agentic Process Automation Solutions assessment (2026). Many of them are better known as intelligent process automation companies, and several of the firms that score well here come from business process services rather than software engineering. Vstorm publishes this ranking and is one of the 10 firms. It is scored only from vstorm.co, by the same rules as everyone else. The previous version of this page, published February 11, 2026, ranked Vstorm first; this version scores it like any other firm.

Agentic process automation companies at a glance #

Each criterion is scored from 0 to 4, and the score is the weighted total out of 100, rounded to the nearest whole number. Firms with equal scores share a place and are listed alphabetically.

Rank	Firm	Suits	Evidence (30)	Method (20)	Estate and deployment (20)	Compliance (15)	Capacity (15)	Score
1	Accenture	Agentic operations inside large transformation programs	4	4	3	4	4	95
2	Cognizant	Regulated processes at enterprise scale	3	4	3	3	4	84
2	IBM Consulting	Enterprises running IBM watsonx	3	4	3	3	4	84
4	Deloitte	Accounts payable agents run as a service	3	3	4	2	4	80
5	EXL	Finance and insurance operations with agents in existing workflows	4	2	3	2	4	78
5	Genpact	Accounts payable at very high volume	4	2	3	2	4	78
7	Vstorm	One process taken to production and handed over	4	4	3	1	0	69
8	PwC	Invoice validation with a documented method	3	3	2	1	4	66
8	Sagility	Health plan claims	3	3	2	1	4	66
10	Roboyo	Agents on Appian and UiPath estates	3	1	3	3	2	61

What an agentic process automation partner does #

Most back-office processes already run on some automation. Rule-based bots and RPA handle the fixed path: copy a field, post an invoice, file a status check. Every case that leaves that path, such as a reformatted invoice, a missing purchase order or an unusual claim, lands in a queue that a person clears by hand. Before agents, the accounts payable team at one of Genpact's clients, a distributor processing close to 3.5 million invoices a year, went through invoices by hand to find the urgent ones and resolved exceptions over email.

An agentic process automation partner adds a layer for those cases. The agent reads the document or message, checks it against the ERP, the contract or the policy, and either resolves the case or passes it to a person with the work already prepared. The bots that run the fixed path usually stay. Vstorm's page on agentic process automation describes the same split: the agent takes the exceptions that existing automation routes to a person, and a person stays the decision-maker where an error is costly.

A partner is different from a platform. Platforms such as UiPath, Appian or IBM watsonx Orchestrate sell the software to build and run agents. A partner maps the process, builds the agents on a platform or in code, connects them to the client's systems and answers for the result. This ranking scores partners. The Vstorm glossary defines the term itself.

How the 10 firms were scored #

The pool started from the 26 agentic automation service providers in the Everest Group assessment and added the service firms named on the previous version of this page, Vstorm as publisher, and firms found while searching for published agentic process cases. Twenty firms were checked. To be ranked, a firm needed at least one public case in which agents act inside a business process and a process measure is reported. A process measure is a number that describes the process: cycle time, the share of cases handled without a person, exception rate, error rate or cost per case.

Nine firms did not meet that rule on the pages reviewed. Capgemini, HCLTech, TCS, Ashling Partners, Infosys and Wipro publish agentic offerings, analyst positions or partner awards, but no client case with agents and a process measure. Ashling's Arnott's case and Centric Consulting's published work use RPA with document extraction. HGS does not say whether agents or RPA bots run its order-to-cash case, and The Hackett Group publishes no case. Accelirate qualified and scored 50, below the cut. The other 13 providers in the Everest assessment were not checked for this edition.

Criterion	Weight	Level 4	Level 3	Level 2	Level 1
Process evidence	30	Two or more process cases with a process measure, one for a named client	One process case with a process measure, or several with no named client	An agent case with no process measure, or a process case run on rules or RPA only	Capability described, no case
Design and evaluation method	20	Level 3, plus a technical guide or an in-depth write-up	Agent design (exception routing, human checkpoints) and an evaluation method both described	One of the two described	Method mentioned without detail
Estate and deployment	20	A published case adds agents to running RPA, BPM or ERP flows, and on-premises deployment is stated	One of the two	Private cloud, client cloud or the client's own environment stated	Public cloud only
Security and compliance	15	Three or more audited standards stated	Two audited standards stated	One audited standard stated	Compliance described, no certification
Delivery capacity	15	10,000 or more people	1,000 to 9,999 people	250 to 999 people	50 to 249 people

Level 0 means nothing was found, or that the firm has fewer than 50 people. ISO 22301 covers business continuity, not security, so it does not count. A certificate counts only when a current public page states it. Every result below is reported by the firm or its client; none was audited for this page.

The score measures what a firm publishes. It says nothing about the quality of any project. Vstorm publishes this comparison and sells agentic process automation, so readers should treat it as an interested party. Each fact is linked, so any row can be re-scored, and firms that find an error can send a source to the author. Buyers whose need is one agent rather than one process can use the sister ranking of AI agent development companies, which uses the same scale with different definitions for evidence and deployment.

Each profile below gives the firm's size, its strongest process case, its method, where its agents run, its certifications and one limitation.

1. Accenture, for agentic operations inside large transformation programs #

Accenture reported approximately 814,000 people at the end of fiscal 2026 (fact sheet). For Google Cloud, its multi-agent lead enrichment system on Vertex AI and Gemini processes 25,000 records in four hours instead of nearly a week, 30 times faster, and cuts cost-to-serve by up to 80% (case study). For YouTube, a Gemini Enterprise agent cut average handle time by 37% during the NFL Sunday Ticket surge (release).

Its AI Refinery platform includes an evaluation agent that scores agents against rubric metrics, and it runs on-premises, including air-gapped environments. Accenture states ISO 27001, ISO 27701 and CSA STAR Gold (Information Security at Accenture).

Accenture suits global enterprises that want agents as one part of a large transformation. Neither case shows agents added to automation the client already runs, and Google is as much a partner as a client.

2. Cognizant, for regulated processes at enterprise scale #

Cognizant reported 351,600 employees at the end of 2025 (results). On its Agent Foundry page, regulated content approvals at an unnamed UK financial services firm went from four weeks to four minutes, with first-time approvals up from 20% to 80%. For Signature Performance, Cognizant configured the ClaimsXM platform on TriZetto Facets, and auto-adjudication rose from about 60% to over 94% (case study). That system runs on configured business rules, so it counts here as automation and does not count as agent evidence.

Cognizant's AI Lab published a write-up on agents testing agents in its open-source neuro-san framework. Through its partnership with Domyn, it offers models on-premises or in a private cloud. It states ISO 42001 and SOC 2 Type II for TriZetto.

Cognizant suits insurers, health plans and banks that want an industrialized build-and-run model. Its agent process result does not name the client.

2. IBM Consulting, for enterprises running IBM watsonx #

IBM reported 264,300 employees at the end of 2025 (Form 10-K). For Avid Solutions, watsonx Orchestrate automates onboarding, project management and expenses: onboarding time fell 25% and manual project-management errors fell 10%.

IBM publishes an evaluation framework that simulates user stories, matches tool calls and scores journey success. watsonx Orchestrate runs as SaaS or self-managed, including on-premises. IBM states ISO 27001 for its business units and ISO 42001 for IBM Granite.

IBM Consulting suits enterprises already committed to IBM's stack. The Avid case is a product story that does not name IBM Consulting as the delivery team, and IBM's other agent case, Inter, measures deployment effort rather than a process result.

4. Deloitte, for accounts payable agents run as a service #

Deloitte reports nearly 500,000 people (FY2026 revenue announcement). At an unnamed large consumer products company, the accounts payable team used to research each invoice by hand before processing it, at up to 30 minutes per invoice. Deloitte's Zora AI agents now read invoices, including handwritten and nonstandard ones, code them and route them: 92% are processed touchless, processing time is down 50% to 75% and invoice processing staff fell by half (case study). The agents work beside the company's upgraded ERP and earlier RPA, and invoices still arrive in the existing accounts payable mailboxes.

A confidence score on each field decides whether an invoice needs review, and Deloitte's Operate team verifies the agents' recommendations. Deloitte's AI platforms run on-premises through HPE Private Cloud AI. Certification is published by member firms: Deloitte Luxembourg holds ISO 27001.

Deloitte suits large enterprises that want finance agents operated as a managed service. Its one process case does not name the client.

5. EXL, for finance and insurance operations with agents in existing workflows #

EXL reported over 65,000 people at the end of 2025 (Form 10-K). For NAES, a global energy services firm, agents extract invoice line items, match them to timesheets and receipts, and propose each invoice with a confidence score and an audit trail. Only exceptions and low-confidence matches go to people. Auto-matching is above 85%, and invoice processing time is down by more than 50%. The system runs on Microsoft Power Platform and the client's existing ERP approval workflow. For an unnamed infrastructure firm, six agents draft engineering reports, cutting report generation time by 70% (case study).

EXL's Governance Hub has built-in guardrails (release), but EXL publishes no agent evaluation method and no deployment options. Its 10-K states HITRUST certification for its healthcare systems.

EXL suits insurers and finance teams that want agents added to workflows they already run. The NAES page gives no baseline and no date.

5. Genpact, for accounts payable at very high volume #

Genpact had over 146,500 employees at the end of 2025 (Form 10-K). For Wesco, a distributor running several ERPs, Genpact AP Suite agents handle data extraction, exception handling, duplicate detection and supplier questions. According to Wesco's CFO, 40% of more than 3 million invoices a year are now processed with no human intervention (release). For an unnamed global beverage company, Genpact reports 80% touchless accounts payable, replacing OCR-based invoice handling.

In Genpact's agentic operations model, human experts supervise outcomes and manage exceptions, but Genpact publishes no method for testing agents. AP Suite runs on Microsoft Azure; no on-premises option is stated. Genpact holds ISO 27001, valid to October 2027 according to BSI's directory.

Genpact suits finance teams with millions of invoices a year across several ERPs. It does not publish how agents are tested before release.

7. Vstorm, for one process taken to production and handed over #

Vstorm is an agentic AI engineering company from Wrocław, Poland, founded in 2017, with 40+ people, including 25+ AI engineers (About). For Mixam, an online print company, a three-agent advisor with 15 tools completes print orders through Mixam's product API: the workflow succeeds in 95.4% of cases against an 80% target, and conversion to a final sale rose from about 20% to about 40% (case study). For an unnamed US telecom, an orchestrator with five sub-agents automates 98% of device activations. A monitoring agent inside the client's existing service platform cut daily error analysis from 330 to 30 minutes (case study).

Vstorm builds evaluation suites from the client's own cases and agrees the success threshold before the build (Evaluation and observability). It also publishes an open, rerunnable benchmark: a clinical triage agent for Schmitt-Thompson Clinical Content reached 98% on 50 scenarios, against 42% for the raw model (case study). Agents run in the client's environment, including on-premises infrastructure (Deployment and infrastructure). Its agentic process automation services end with code, evaluation suite and runbook handed to the client.

Vstorm suits a team that wants one process taken from proof of value to production and then run by its own people. It states no audited security certification, has fewer than 50 people, and neither case adds agents to RPA, BPM or ERP flows already running.

8. PwC, for invoice validation with a documented method #

The PwC network had over 360,000 people in its 2026 fiscal year (PwC at a glance). A hyperscaler client of PwC India validated invoices from a large vendor base with OCR, rules and spreadsheet macros, which handled standard invoices but not unusual ones. PwC split validation into small, single-purpose checks run by specialized agents through an orchestration layer, each returning a confidence score; low-confidence cases go to people. Manual processing per invoice fell from two to three days to under 10 minutes, automation leveled off at about 95%, and accounts payable operating costs fell by up to 60% (case study).

The same write-up tracks validation accuracy during rollout, reaching 85% by the end of week three. PwC India's agentic automation brochure describes a governance framework and deploying agents in the client's target environment. No current security certification was found for the firms involved.

PwC suits enterprises that want a consulting firm to rebuild one control-heavy finance process. Its case does not name the client.

8. Sagility, for health plan claims #

Sagility had 46,860 employees on March 31, 2026 (results). For an unnamed regional health plan with 350,000 to 400,000 members, claims reviewers used to work every claim queue by hand. Agents now read claim documents with OCR and a client-approved GPT model, validate the data and apply versioned rules. About 80% of in-scope claim queues run without manual handling, cycle times are 10% to 15% shorter, and accuracy is above 95%, rising above 99% with human review. The work went from proof of concept to production in five months (case study).

Reviewers accept, edit or pend each claim in an interface that links every field to its source, and dashboards track exceptions and rule hits. The system plugs into the client's existing IT through APIs. No security certification was found on the pages reviewed.

Sagility suits US health plans that want claims operations and agents from one provider. Its case does not name the client.

10. Roboyo, for agents on Appian and UiPath estates #

Roboyo was founded in 2015 in Germany and operates in 14 countries (About us); a third-party jobs profile puts it at about 719 people (Simplify). For the National Injury Insurance Scheme Queensland (NIISQ), invoice processing took more than a week each month. Roboyo deployed Appian's AI agents on the scheme's Appian platform: processing time fell from five days to under 24 hours and manual effort by up to 90% (techpartner.news). Everest Group named Roboyo a Star Performer in its 2026 assessment.

Roboyo states ISO 27001 for several entities and Cyber Essentials Plus (information security statement). It publishes no agent design or evaluation method, and no deployment options.

Roboyo suits companies that already run Appian or UiPath and want agents added to that estate. The NIISQ agents extract invoice data, which is closer to document processing than to agents acting across systems.

Which firm fits which process #

The firms' published cases cluster in a few processes. A firm with a case in your process has at least met that process's exceptions once.

Process	Firms with a published case	Strongest published result
Accounts payable and invoice validation	Deloitte, Genpact, PwC, Roboyo	92% of invoices touchless (Deloitte)
Billing and order-to-cash	EXL	Auto-matching above 85% (EXL, NAES)
Claims	Sagility	About 80% of claim queues automated
Order completion, activation and onboarding	Vstorm, IBM Consulting	98% of device activations automated (Vstorm)
Sales and regulated approvals	Accenture, Cognizant	Approvals from four weeks to four minutes (Cognizant)

If you need a named reference you can call, the shortlist is shorter. Six firms name the client next to a process result.

Firm	Client	Result
Accenture	Google Cloud	Lead processing 30 times faster, cost-to-serve down by up to 80%
EXL	NAES	Auto-matching above 85%, invoice processing time down by more than 50%
Genpact	Wesco	40% of more than 3 million invoices a year with no human intervention
IBM Consulting	Avid Solutions	Onboarding time down 25%, project-management errors down 10%
Roboyo	NIISQ	Invoice processing from five days to under 24 hours
Vstorm	Mixam	95.4% workflow success, conversion from about 20% to about 40%

Questions to ask before you sign #

A ranking shows only what firms make public. Six questions close the gap.

Which exceptions does the agent resolve alone? Ask for the exception types from a live case and the share of cases each one takes. The process measures above only mean something next to that list.
Who approves what? Ask where the confidence threshold sits and who reviews the cases below it. Deloitte, EXL and PwC publish this for their cases.
How is the agent tested before each release? Seven of the 10 firms describe both agent design and an evaluation method. EXL, Genpact and Roboyo do not.
What happens to the bots you already run? Only four firms show agents working inside an existing ERP, BPM platform or approval workflow. If your process already runs on RPA, ask for a case where it stayed.
Where do the models run? Five firms state on-premises deployment. If a regulator or a customer requires your own infrastructure, ask for a project that already runs that way.
Who owns the result? Ask who owns the code, the rules, the prompts and the evaluation suite when the project ends, and whether the agents depend on a platform license.

Platforms these partners build on #

The previous version of this page ranked UiPath, Automation Anywhere, Salesforce Agentforce, IBM watsonx Orchestrate, Beam AI and Lindy AI alongside service firms. They are software, so this version names them without scoring them. Several ranked partners build on them:

Appian: Roboyo's NIISQ agents run on the scheme's Appian platform.
UiPath: Roboyo runs UiPath estates as a managed service (OPmobility), and Accelirate, which scored below the cut, builds its agentic cases on UiPath.
Microsoft: EXL's NAES agents run with Power Platform, and Genpact AP Suite runs on Azure.
IBM watsonx Orchestrate: the platform behind IBM Consulting's Avid Solutions case.
Google Gemini Enterprise: behind Accenture's YouTube agent.

Automation Anywhere and Salesforce Agentforce do not appear in any ranked firm's process case reviewed here.

A platform configured in-house can be enough when the workflow is standard, the systems are common SaaS tools and the rules rarely change. A partner earns its cost when agents must act in your own systems, follow your own rules or meet a regulator's requirements. Vstorm's buy versus build framework works through that decision one workflow at a time.

Frequently asked questions #

What is the difference between an agentic process automation company and an RPA vendor?

An RPA vendor sells software that repeats a fixed sequence of steps. An agentic process automation company builds agents that handle the cases the fixed sequence cannot, such as a reformatted document or a missing record, and connects them to the client's systems. Some firms in this ranking do both: Roboyo, for example, also runs UiPath estates for clients.

Do we have to replace the bots we already run?

No. In the four cases where agents joined existing automation, the existing systems stayed: Deloitte's client kept its ERP and earlier RPA, EXL built on NAES's ERP approval workflow, Genpact worked across Wesco's ERPs, and Roboyo used NIISQ's Appian platform.

How long does a first process take to reach production?

Few firms publish timelines. Sagility moved from proof of concept to production in five months. PwC reports 1.5 weeks from pilot to production for its invoice validation agents. Vstorm states three to six weeks for a scoped proof of value on one process (agentic process automation).

What does agentic process automation cost?

None of the case studies reviewed states what the project cost. Some report what it saved: PwC's client cut accounts payable operating costs by up to 60%, and Accenture reports cost-to-serve down by up to 80% for Google Cloud's lead processing.

How do we judge a partner's case studies?

Look for four things: a named client, a process measure, a baseline and a date. Every result in this ranking is reported by the firm or its client, and only six firms name the client next to a process result.

Is this ranking independent?

No. Vstorm publishes it and is one of the 10 firms. Vstorm is scored only from vstorm.co, every other firm from its own pages, and each fact is linked so that any row can be checked and re-scored.

References #

All pages were read on October 29 or 30, 2026.