Alternatives
Why WorkGraph rather than an alternative.
Five design choices that set us apart, and the cases where another tool is enough.
Five design choices
What others do, what we do, and what it changes for the client.
01The client’s own process as the join key
Others
They attribute time or cost to a team, initiative, agent or use case (Larridin, IBM Apptio, ServiceNow, Pay-i, CloudZero). We found none that does it on a process taxonomy.
WorkGraph
Hours, tokens and outcomes are tied to the client’s processes, on an open reference base (APQC).
What it changes
A cost per process can be compared, budgeted and managed. A cost per team does not say what work it pays for.
02Human cost and AI cost on the same line
Others
Some measure humans (Skan, KYP.ai, Soroco), others AI (Langfuse, CloudZero). At IBM Apptio, labour is an input. Revenium counts human review time, at workflow scale.
WorkGraph
One line in euros per process: loaded human cost, AI cost, volume, unit cost.
What it changes
McKinsey publishes that a banking customer-service agent costs 70 to 75% in human oversight, 20 to 25% in tokens. Measuring AI alone misses most of it.
03Nothing on devices, statistics only
Others
Skan, Mimica, KYP.ai and UiPath install an agent that captures screens, clicks or keystrokes. Larridin adds a browser extension. Workhelix reads prompts.
WorkGraph
Human capacity from HR data, systems, volumes and declarations. For AI, uses are classified inside the client: no human reads, only statistics come out, below a threshold of five people.
What it changes
No individual-level tool exists, which simplifies the conversation with employee representatives. Their consultation is still due.
04Every vendor, not one suite
Others
Microsoft Copilot Analytics covers Microsoft surfaces, with a task taxonomy defined by Microsoft. ServiceNow tracks its own workflows, Workday its own agents.
WorkGraph
OpenAI, Anthropic, Microsoft, Google and in-house agents, tied to one taxonomy: the client’s.
What it changes
A real company uses several vendors. A per-suite view leaves gaps exactly where spend is scattered.
05Value against a baseline, not hours × rate
Others
Larridin values in hours × rate. Copilot Analytics applies $72 an hour by default. Workhelix estimates. Firms measure in engagements, one-off.
WorkGraph
The pilot sets the baseline before AI, then measures the gap, and follows value from identified to realised.
What it changes
Hours saved are not cost saved. Only a baseline tells them apart.
Head to head
The six most credible alternatives: where they are strong, what they lack, and when they are enough.
Larridin
AI execution intelligence · US start-up, $17M (a16z)
Where they are strong
Same “human + agent” story, funded. Token spend per team, tool and agent; workflows discovered from application sequences.
What they lack
No client process as key. Humans are covered through a browser extension and desktop agent. Value in hours × rate.
They are enough when
You want to manage AI adoption and spend per team, with no link to the cost of work.
SourcesLarridin
Skan AI
Desktop observation · $63M Series C
Where they are strong
Fine measure of human time per process. Best-documented safeguards in the segment: local blurring, pseudonymisation, a works council agreement cited (Allianz).
What they lack
AI compute per process: not found. Requires an agent on devices.
They are enough when
The goal is to find what to automate in one given process, and desktop observation is accepted.
Celonis and process mining
SAP Signavio, Microsoft Power Automate · segment standard
Where they are strong
De facto standard, object-centric model, ERP and CRM connectors. Exports to OCEL 2.0.
What they lack
Work outside systems (email, spreadsheets, calls), loaded human cost, AI spend.
They are enough when
The process lives almost entirely in the ERP. Otherwise it is a source for us: we import their exports.
Microsoft Copilot Analytics
Vendor-native analytics · included with Microsoft 365
Where they are strong
Native, no integration, with tracking of Microsoft agents, a consumption dashboard and a minimum group size (10, lowerable to 5).
What they lack
Outside Microsoft. Task categories defined by Microsoft. Assisted value: hours × $72 by default.
They are enough when
The client is 100% Microsoft and a generic taxonomy is enough.
SourcesMicrosoft Learn
IBM Apptio and ServiceNow
AI value · incumbents
Where they are strong
CFO relationship (Apptio) and workflow records already held (ServiceNow). ROI per initiative or agent.
What they lack
Measured human capacity of the process: at Apptio, labour is a cost input. Grain: initiative or agent.
They are enough when
The client is already equipped and manages by initiative.
SourcesIBM ApptioServiceNow
Ontora and Workhelix
Declarative discovery · Ontora YC 2026, Workhelix $15M Series A
Where they are strong
Fast entry: interviews run by AI agents (Ontora, which also targets PE funds), roles broken into tasks (Workhelix, founded by Brynjolfsson, McAfee and Rock).
What they lack
Measured volumes and costs. Workhelix works at role level, estimates its savings and reads prompts.
They are enough when
A declarative diagnostic is enough to decide.
Consulting firms
They measure AI value in engagements, per use case, with their own frameworks.
- McKinsey (QuantumBlack)
Five-layer measurement framework (April 2026); guide to the economics of agentic workflows.
Value defined up front, metrics from technical performance to financial impact, attribution by A/B test or staggered rollout, evidence pack with total cost.
- BCG (BCG X)
Three postures: Deploy, Reshape, Invent. 10-20-70 rule: 10% algorithms, 20% technology, 70% people and processes.
Per end-to-end workflow transformation; 30 to 50% efficiency gains claimed for Reshape (the firm’s own claim).
- Bain
OpenAI alliance, Elite Partner, investor in OpenAI Deployment Company (May 2026), which targets private equity firms.
Deployment engagements; no measurement tool found.
- Accenture
Tokenomics (29 July 2026): tooled consulting, four-week diagnostic.
Pivot metric: cost per successful business action. Contracts can be indexed on it.
- Deloitte, PwC, KPMG
Agent orchestration platforms: Zora AI, agent OS, KPMG Workbench.
Delivery-oriented. No per-process economic measurement product found.
Their measurement is an engagement deliverable, one-off and per use case. WorkGraph is a continuous instrument, at process level, that a firm can use in an engagement: a channel as much as a competitor. The risk: a firm tools up its own tracking, as Accenture did with Tokenomics.
What we do not rebuild
The plumbing is commoditised and consolidating: we plug into it.
- Token capture
- Langfuse (acquired by ClickHouse), LiteLLM, Cloudflare AI Gateway, OpenTelemetry
- Process mining
- Celonis, Signavio, Power Automate, through the OCEL and BPMN formats
- Billed amounts
- OpenAI and Anthropic billing APIs, as the reconciliation reference
- Starting taxonomy
- APQC PCF, open licence with attribution
Sources and limits
What remains to be proven
- WorkGraph is at design stage. These choices are the intended approach, to be validated in a pilot.
- The work map without a desktop agent must be accurate enough to decide.
- The share of AI spend attributable to processes is still to be measured with real clients.
- The absence of a direct competitor is observed, not proven: the market saw three major launches between July and August 2026.
- Sources consulted on 28 and 29 September 2026: vendor pages, documentation, press releases, trade press.
- Funding and gain figures are the players’ own, unaudited.
- The McKinsey, BCG and Bain pages were read through search summaries, not in full.
- The “why it is better” reading is ours.
Compare on your own case.
One process, your current tools: we tell you what they already cover and what is missing.
With an expert · no commitment