Back office ops · Production

How Spendesk turned 90% adoption into embedded AI workflows with a champions program

The problem

After achieving 90% AI adoption, Spendesk found that depth of usage was uneven — most employees were using Dust as a generic chat replacement rather than in embedded workflows, and an initial limited rollout had created a perception that AI was only for certain roles.

First attempt

A summer hackathon that generated 11 new agents resulted in only 1 surviving after six months, with momentum evaporating almost immediately — demonstrating that burst-format events do not produce sustainable AI habits.

Workflow diagram · grounded in source
1
Company-wide platform deployment
trigger
“By May, the platform had expanded company-wide”
2
Champions program formalized
integration
“Spendesk formalized what had been informal: a structured Champions Program with clear roles, dedicated time, and real accountability”
3
Department agents built
ai_action
“Build: Create agents for department-specific use cases and workflows”
4
Governance touchpoints
feedback_loop
“Bi-weekly meetings (30 minutes): Champions share current use cases, ask questions, learn from each other”
5
Custom agents serve employees
output
“They were using purpose-built agents created by Spendeskers for Spendeskers, tools that solve specific problems in specific workflows”
6
360 Customer View unification
integration
“connects data that traditionally lives in separate systems: Salesforce, product databases, and other sources that don't naturally talk to each other”
Reported outcome

By December 2025, Spendesk achieved 93-94% monthly active users and 80%+ weekly retention, with 40%+ of messages going to purpose-built custom agents rather than generic LLM interactions.

Reported metrics
monthly active users (December 2025)93-94%
Weekly retention (key highlights)83%+
Weekly retention (results section)80%+
Messages sent to custom agents40%+
Show all 11 reported metrics
monthly active users (December 2025)93-94%
weekly retention (key highlights)83%+
weekly retention (results section)80%+
messages sent to custom agents40%+
AI Champions driving adoption13
initial AI adoption within six months90%
summer hackathon agents surviving after six monthsonly one of those 11 agents was still being used regularly
six-week hackathon agents still actively used8 agents created, 3 still actively used weeks later
AskProduct hallucinations reported since September2
monthly adoption by November80%
monthly active users by June65-70%
Reported stack
DustSalesforce
Source
https://dust.tt/customers/how-spendesk-turned-90-adoption-into-embedded-ai-workflows-with-a-champions
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

By December 2025, Spendesk achieved 93-94% monthly active users and 80%+ weekly retention, with 40%+ of messages going to purpose-built custom agents rather than generic LLM interactions.

What tools did this team use?

Dust, Salesforce.

What results were reported?

monthly active users (December 2025): 93-94%; Weekly retention (key highlights): 83%+; Weekly retention (results section): 80%+; Messages sent to custom agents: 40%+ (source-reported, not independently verified).

What failed first in this deployment?

A summer hackathon that generated 11 new agents resulted in only 1 surviving after six months, with momentum evaporating almost immediately — demonstrating that burst-format events do not produce sustainable AI habits.

How is this back office ops AI workflow structured?

Company-wide platform deployment → Champions program formalized → Department agents built → Governance touchpoints → Custom agents serve employees → 360 Customer View unification.