It support · Production

Smartsheet deploys Claude across customer-facing AI, engineering, and company-wide surfaces in early 2026

The problem

Smartsheet customers faced fragmented data across multiple systems, data warehouses, and project repositories, making a unified view of work difficult to achieve. Inside engineering, existing AI coding tools only supported line-level code completion and could not bridge data silos across Snowflake, service databases, and internal APIs.

First attempt

Existing AI coding tools were limited to line-level code completion and could not work across the data boundaries Smartsheet's engineering team needed.

Workflow diagram · grounded in source
1
User submits request via Claude
trigger
“The Smartsheet MCP Connector for Claude lets users view, create, and manage their Smartsheet work directly from a Claude conversation.”
2
Claude interprets natural language
ai_action
“users discovered 27 distinct capabilities with zero documentation or training, finding them through natural conversation alone”
3
MCP Connector routes to Smartsheet
integration
“The connector supports 36+ operations”
4
Work actions executed in Smartsheet
output
“48% of all connector usage went beyond querying data. Users were not just asking questions about their work. They were creating sheets, updating rows, and flagging risks through Claude.”
5
9-agent SDLC template runs
ai_action
“The team built a 9-agent SDLC template covering architecture planning, implementation, code review, merge request processing, and accessibility remediation.”
6
Token usage linked to engineering output
feedback_loop
“We built dedicated reporting that links token usage to code written, PRs merged, and code deployed—the full cycle. That's how we surface what patterns are working and share them across the org.”
Reported outcome

The Smartsheet MCP Connector reached 1,400+ organizations within 30 days of launch, with 48% of usage going beyond queries into active work creation.
Claude Code users ship 3x more code and merge 31% more pull requests, with cycle time dropping 28%. Helpdesk volume fell 30%, prompt caching cut operational costs by 60%, and 49% of employees were actively using Claude within 2.5 weeks.

Reported metrics
MCP Connector organizations in first 30 days1,400+
MCP Connector operations supported36+
Distinct capabilities discovered without documentation27
MCP Connector usage beyond querying48%
Show all 20 reported metrics
MCP Connector organizations in first 30 days1,400+
MCP Connector operations supported36+
Distinct capabilities discovered without documentation27
MCP Connector usage beyond querying48%
Code documentation time reductiontwo weeks to four hours
AWS cost management tool build time30 minutes
Projected annual savings from AWS cost toolhundreds of thousands of dollars annually
Employees completing training on Day 1272
Employee base actively using platform within 2.5 weeks49%
Helpdesk volume reduction30%
Code shipped by Claude Code users vs peers3x more code
Pull requests merged by Claude Code users vs peers31%
Highest-performing engineer output increase5 to 7x their previous output
Cycle time reduction across org28%
Operational cost reduction from prompt caching60%
Response latency improvement from prompt caching20%
Cache hit rate overall70%
Cache hit rate on follow-up queries90%
Nonprofit applications evaluated in under an hour3,200+
Early adopter shipping rate increase5x their previous rate
Reported stack
ClaudeClaude CodeClaude EnterpriseClaude SonnetSmartsheet MCP ConnectorAmazon BedrockSnowflake
Source
https://www.anthropic.com/customers/smartsheet
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Smartsheet MCP Connector reached 1,400+ organizations within 30 days of launch, with 48% of usage going beyond queries into active work creation.

What tools did this team use?

Claude, Claude Code, Claude Enterprise, Claude Sonnet, Smartsheet MCP Connector, Amazon Bedrock, Snowflake.

What results were reported?

MCP Connector organizations in first 30 days: 1,400+; MCP Connector operations supported: 36+; Distinct capabilities discovered without documentation: 27; MCP Connector usage beyond querying: 48% (source-reported, not independently verified).

What failed first in this deployment?

Existing AI coding tools were limited to line-level code completion and could not work across the data boundaries Smartsheet's engineering team needed.

How is this it support AI workflow structured?

User submits request via Claude → Claude interprets natural language → MCP Connector routes to Smartsheet → Work actions executed in Smartsheet → 9-agent SDLC template runs → Token usage linked to engineering output.

Related it support cases