Buyer guide · tool selection by use case
Five tool types, a use-case matrix with prerequisites and supervision needs, a scoring rubric and a governance self-test. Vendors are named only as examples of a type. ZeroTwo publishes this page and says where it does not fit.
For each job: the tool type, what must already be true, which systems it reaches, how much supervision it needs and who owns it. Choosing the platform underneath comes first or in parallel; see how to evaluate enterprise AI platform governance and portability.
| Use case | Tool type | Prerequisites | Systems reached | Supervision | Owner | Where ZeroTwo fits |
|---|---|---|---|---|---|---|
| Customer service deflection | Vertical agent platformse.g. Sierra, Decagon, Moveworks, Aisera | A clean knowledge base, defined escalation paths and ticketing integration. | Helpdesk or contact-centre software, CRM, order systems. | High: review transcripts, sample resolutions, audit escalations. | Head of support | Drafting macros and help articles for the support team. Not a deflection agent. |
| Internal knowledge retrieval | Knowledge copilotse.g. Glean, Dust | Connectors to your document stores with permissions mirrored, and content hygiene. | Wikis, drives, chat, ticketing. | Medium: spot-check answers and test for permission leakage. | IT or knowledge management | Cross-functional drafting alongside a retrieval tool. |
| Cross-functional drafting and research | Unified model workspacese.g. ZeroTwo, ChatGPT Enterprise, Claude's enterprise plans | An acceptable-use policy, an identity decision and a named pilot group. | Whatever staff connect or upload; the open web. | Low to medium: a person reviews anything before it leaves the company. | IT or the AI program lead | The primary use case. |
| Custom agent inside a cloud perimeter | Build-your-own on cloud primitivese.g. AWS Bedrock AgentCore, Vertex AI Agent Builder, Microsoft Copilot Studio | Platform engineers, evaluation and logging tooling, cloud data governance. | Your cloud data and internal APIs. | High: you own evaluations, regression tests and incident response. | Platform engineering | A place to draft and test prompts before you build. |
| Back-office automation | Workflow automation layerse.g. n8n, Tray.ai, Lindy | A documented process, API access to the systems and an exception queue. | ERP, CRM and email through connectors or webhooks. | Medium: handle exceptions and re-test prompts after any change. | Operations or the automation lead | A place to draft and test the prompts those workflows call. |
Tool names are examples of each type for orientation. Confirm each vendor's current scope and plans on its own site.
Pick the type first. Within a type, vendors compete on the same terms and the rubric below is easier to apply.
One interface and one bill for several providers' models, with shared projects for teams.
Agents built around one operational function, with supervision and integrations for that function.
Agent frameworks inside a hyperscaler, composed by your own engineers.
Search and answers across your own documents, chat and ticket systems.
Model calls as steps inside business-process automation, not the main interface.
For each pillar, score your evidence from 0 to 1 against its criteria, multiply by the weight, and add the five results. The weights are our editorial default; change them to match your risk.
Total score = Σ (pillar score from 0 to 1 × weight). Maximum 100.
These questions are about your side of the table, not the vendor's. Every box you cannot tick is a gap to close before you scale past a pilot. The list is our editorial judgment, not a standard.
The earlier version of this page quoted a range whose low end matched none of the scenarios shown. We removed it and no longer publish competitor totals: the inputs are quote-specific. For a worksheet with live formulas, use the one on the enterprise AI platform guide.
Plan prices and volume discounts vary. Take each figure from the vendor's current pricing page or your quote.
The engineer line usually decides whether this option wins. Count it honestly.
Seat licences only, at an illustrative headcount. Extra credits are separate; see the pricing page.
Self-reported survey results from one primary source. They are context for your plan, not a forecast for your company.
Source: McKinsey, The state of AI in 2026: On the road to ROI, 25 August 2026; online survey of 1,719 participants in 97 nations, fielded 4 May to 8 June 2026. Read on McKinsey's page on 5 October 2026.
A single-vendor workspace is a legitimate answer for some organizations. The question is whether one provider covers what your teams need, and whether you want to keep model choice open.
Choose the pattern before you sign the contract and baseline it before the pilot starts.
AI resolves work that would have reached a human queue.
cases auto-resolved per week × loaded cost per case
The same headcount produces more drafts, briefs, code or analyses.
output volume vs. baseline, plus time saved per task
The same output arrives sooner: RFP turnaround, contract review, onboarding.
median and 95th-percentile days end to end vs. baseline
AI creates or accelerates revenue through enrichment, coaching or expansion plays.
pipeline created, win-rate change, expansion linked to AI-assisted plays
A unified model workspace with 60+ models, a Free plan and a team plan. It is not a deflection agent or an embedded agent for a system of record.
$35.99 per seat per month with annual billing. Feature list from the pricing catalog:
The Free plan ($0) starts with 100 free credits, adds 20 bonus credits each day you log in, has limited model selection and needs no card. Plus is $14.99 and Pro is $29.99 per month for individuals.
Compare ZeroTwo plansA tool earns the word when four things hold at once: documented data residency and retention terms, identity and access that inherit from your existing SSO or SCIM setup, an audit trail your security team can query, and integrations that reach the systems where work happens. A bigger pricing tier on a consumer app does not meet that bar by itself.
Enterprise buyers weigh governance depth, integration into core systems, scale and availability commitments, and the procurement model (annual contracts, security review). Small-business tools usually optimize for self-serve setup and speed to first use.
Pick the tool type first and the vendor second. Classify each use case with the matrix on this page, score two or three vendors inside that type with the 100-point rubric, then pilot the top two on a real internal task and measure the change in cycle time rather than relying on demos.
There is no single figure; it depends on the cost structure you choose. Stacked subscriptions, build-your-own and a unified workspace each have different lines to price, shown on this page. For reference, ZeroTwo Business lists at $39.99 per seat per month, which is $239,940 a year for 500 seats in seat licences at list price with monthly billing, before implementation, admin time or extra credits.
Security depends on both the vendor's evidence and your own operating discipline. Ask each vendor for current audit reports, a sub-processor list and contract terms on training data, then work through the nine-question self-test on this page to find gaps on your side.
For deflection at scale, vertical agent platforms such as Sierra, Decagon, Moveworks or Aisera are built for the supervision and contact-centre integrations that work needs. A general workspace such as ZeroTwo is better suited to the support team's own writing: macros, knowledge articles and internal summaries.
Four patterns are common: deflection, throughput, cycle time and revenue. Pick the one that matches the use case before you sign, and baseline it before the pilot. For context, 37% of respondents in McKinsey's 2026 survey attribute at least some EBIT impact to AI, essentially unchanged from the previous year.
ZeroTwo is a unified model workspace: 60+ models in one product, with Business as the team plan at $39.99 per seat per month and a Free plan that needs no card. It suits cross-functional drafting, research and analysis, and prototyping prompts. It is not a contact-centre deflection agent or an embedded agent for a system of record, so pair it with a purpose-built tool where the work calls for one.
Evaluate platform governance, portability and a cost worksheet before you pick tools.
How agents are built, supervised and run.
The multi-model workspace category in more depth.
Where AI fits inside back-office automation.
How to pick a model for a writing task.
A wider look at platforms across consumer and business use.