Sierra × Airtable — Proposal

An AI‑native support layer for 500,000 organizations

A proposal to replace Airtable’s deflection-oriented help experience with a conversational agent that resolves the majority of inbound support itself.

Prepared for Airtable Customer Experience · Valid 30 days

80% Resolution rate reached

Four in five inbound conversations closed by the agent without a human touch — the target we commit to at steady state, measured on your own conversation logs, not on a benchmark.

Where support stands today

Airtable’s current support experience is organized around documentation. A customer asks a question and receives a link to a help center article. The answer exists, but the customer still has to find it, read it, and translate it into their own base.

That model breaks down exactly where Airtable is strongest: configurable, customer-specific workspaces. There is no single article that describes your automation, your field types, your permissions. The result is a support queue growing in step with adoption, and a help center measured on deflection rather than on whether anyone was actually helped.

What we propose

Answers, not articles

The agent reads the customer’s workspace context and returns step-by-step instructions for their base — not a link to a page that describes the general case.

One agent, every channel

Chat first, then email on the same agent and the same knowledge. A customer who switches channel mid-issue does not start over.

Context that survives the turn

Topic shifts, multi-turn email threads, and follow-ups days later all resolve against the same conversation history.

Rollout

  1. Discovery and agent build

    We ingest the help center, past ticket history, and your escalation policy, then build and evaluate the agent against replayed real conversations.

    Weeks 1–3
  2. Chat goes live

    The conversational interface ships to production traffic behind a staged rollout, with human handoff on every path from day one.

    Week 4
  3. Email channel added

    The same agent takes over the inbound support inbox, including multi-turn threads it can carry autonomously.

    +7 days
  4. Tune to target

    Weekly review of unresolved conversations, feeding new skills and guardrails until resolution rate holds at target.

    Ongoing

Projected outcomes at steady state

AI resolution rate

80%

Of all inbound conversations, chat and email combined.

Workload absorbed

95 FTE

Equivalent full-time support agents’ worth of volume handled by the agent.

Time to first channel live

4 weeks

From kickoff to chat in production, with email seven days behind it.

Commercial terms

Outcome-based pricing. You pay per resolved conversation, not per seat and not per message — if the agent does not resolve it, it is not billed.

Line itemTermBasis
Platform12 monthsAnnual
Resolved conversationsCommitted tierPer resolution
ImplementationIncluded
Additional channelsIncluded
Resolution rate floor80%Reviewed quarterly

Next steps

Send us a redacted export of one month of support conversations. We will replay them against a configured agent and return the measured resolution rate before anything is signed — the number above should be something you verify, not something you take on faith.