Giga

What it is

Giga builds AI support agents that take the contact themselves rather than deflect it — voice first, with chat and email running off the same agent definition. The company is San Francisco based, founded by Varun Vummadi and Esha Manideep Dinne, and went through Y Combinator’s Summer 2023 batch as Giga ML, selling fine-tuned on-premise open-source models before repositioning the whole company on enterprise support. It announced a $61M Series A led by Redpoint on 5 November 2025, with Y Combinator and Nexus Venture Partners participating, taking total funding to roughly $65M.

DoorDash is the anchor reference: when a Dasher cannot complete a delivery, Giga holds the live connection with the Dasher, calls the consumer to verify the address, and checks the result against policy in the same session. Other named customers include Toast, Postman, Capital.com, Shiftsmart, Flex, Afriex, and Remedy Meds.

Why it shows up in CX and support stacks

• The Browser Agent takes the integration project off the critical path. Giga’s agent signs into your existing web tools as a support rep would, navigates the interface, completes the workflow, verifies the record actually changed, and logs each action. No API required against the system that holds the answer. For teams whose refund, address-change, or entitlement logic lives in a vendor admin console or a homegrown tool with no usable API, that is the difference between a two-week deployment and a two-quarter one — and it is the capability none of the chat-first platforms sell.
• Voice is where the proof sits. Giga states a 400 millisecond voice response and 99-language support, and published a real-time hallucination-correction result for voice. Its DoorDash number — more than 90% delivery workflow resolution over the 20 September to 20 October 2025 window — covers multi-party coordination with policy checks, which is a harder shape than single-caller FAQ containment. Treat all of these as vendor-reported; none has been independently audited.
• The build surface is a policy canvas, not a prompt box. Agent Canvas is a low-code environment where you define policies, escalation paths, and tooling, run simulations with pass/fail counts, and cut releases. Scout, the in-platform AI developer, reads production conversations and human interventions against a KPI you name, proposes a typed change to policy or knowledge, and tests it on limited traffic first. Giga publishes the shape of a 12-day rollout: baseline policies drafted from your existing documentation on day 1, review-ready versions after manual QA by day 6, a 1% traffic canary on day 12 before scaling to 100%.
• Action risk is tiered in the architecture rather than bolted on afterwards. Giga’s own architecture writing classifies work into low-risk read, low-risk write, customer-confirmed write, human-approved action, and blocked action, with the explicit position that workflows should not collapse into a single all-privileged automation account. If you are handing an agent credentials to your admin tooling, that taxonomy is the thing your security review will ask for, and it exists before you ask.
• The compliance set is enterprise-shaped. The public trust center lists SOC 2, ISO 27001, PCI DSS, HIPAA, and GDPR.

Pricing reality

Nothing published, no self-serve tier, no pricing page. Custom enterprise contracts only. Third-party 2026 analysis puts Giga deal sizes in the six-figure annual range at minimum, against a $50K to $100K annual band for comparable AI support vendors, and puts the practical fit above roughly 50,000 monthly inbound contacts — below about 10,000 a month you are outside the segment it sells to.

The more important gap is that Giga does not state a billing unit publicly. Its competitors do: Intercom Fin bills $0.99 per outcome, Zendesk $1.50 per automated resolution, Parloa per minute. An unstated meter is a negotiable meter, so negotiate it. Get four things in writing before signature: the unit you are billed on, the minimum annual commitment, the overage rate above it, and the written test for what counts as a resolved contact. The fourth one is the one that turns into a quarterly argument if you skip it.

Best for

Enterprise support and CX leaders above roughly 50,000 monthly contacts, weighted to voice, whose highest-volume workflows involve multi-party coordination or run through admin tooling with no API worth integrating — delivery and logistics, marketplaces, fintech operations, hospitality. It is the strongest pick on this list when the blocker on your last AI-agent evaluation was “the agent cannot actually do the thing, because the thing lives in a web console.”

Skip it if your volume is chat-dominant and low-complexity, if you need published pricing you can approve without procurement, or if your contact volume cannot repay a six-figure first-year commitment.

Versus the alternatives

The installed-base incumbents are Zendesk and Intercom — take them when the AI already attached to your help desk clears your containment bar, which is the right answer for most chat-weighted operations and always the cheaper first test. Sierra is the fastest-growing entrant in the segment, at roughly $200M ARR by May 2026; pick Sierra when you want the same transactional ambition with a much longer enterprise reference list behind it. Parloa is the pick when telephony estate integration decides it — Genesys, Five9, NICE, Avaya — and when you want a per-minute meter you can model in advance. Decagon wins on aggressive chat deflection. Giga’s case against all four is narrow and real: it is the one that will drive your UI when there is no API, and the one that publishes an action-risk taxonomy for doing so.

Watch-outs

• Every headline number is vendor-reported. 90%+ resolution, 98% in technology, 400ms response — all Giga’s own figures, none independently audited, and “resolution” is defined by the party billing you. Guard: before signature, require a two-week shadow run on your own traffic and score it with your own definition — closed and not re-contacted within 7 days, not abandoned mid-flow, not silently escalated. Put that definition in the contract as the acceptance test, not in the SOW as an aspiration.
• The Browser Agent is a credentialed robot inside your admin tools. The capability that makes Giga interesting is also the largest blast radius on this page, and browser automation breaks when a vendor ships a UI change you did not schedule. Guard: give each workflow its own service identity rather than one shared automation account, cap it at the read and customer-confirmed-write tiers on day one, route anything with money attached through the human-approved tier, and set an alert on action-failure rate so a silent UI change surfaces as a metric instead of a backlog.
• Scout changes production behaviour, and that is the point. A loop that proposes and tests policy edits against a KPI is a real advantage and also a change-management surface your auditors have not seen before. Guard: require named human approval on every Scout change before it leaves the 1% canary, and export the change log into your own system of record monthly — do not let the vendor hold the only history of what your agent was told to do.
• The company is young and this is its second product. Giga ML sold on-premise models; Giga sells support agents. The pivot is recent, the current product’s public track record is short, and the reference list is concentrated in a few logos. Guard: hold the first term to one year, put contractual export rights on your policies, transcripts, and simulation cases in a usable format, and name the fallback vendor before you sign rather than after an incident.
• Omnichannel is marketed ahead of where the evidence is. Giga sells one agent across voice, chat, email, SMS, and WhatsApp with shared history, but the published proof — customers, metrics, technical writing — is heavily voice. Guard: run your own chat and email evaluation on real tickets during the pilot instead of pricing the non-voice channels off the demo, and if voice is under 40% of your contact mix, evaluate Decagon or your incumbent’s native agent alongside it.