Guided workflow walkthroughs to give millions of customers a self-serve path forward

When customers get stuck in a product workflow, generic self-help instructions often add confusion rather than clarity — pushing them to escalate to a live agent before the issue is resolved. Twenty percent of support contacts are workflow guidance questions: the exact category a Guidance Agent is built to deflect. Automating those walkthroughs is a $33M opportunity and a path to self-serve resolution for millions who today have no option but a live expert.

The north star was a "Product Support Done For You" experience inside VEP — with Intuit Assist as a guide, proactive when possible and reactive when needed. The agent scales along a spectrum from Guide (assist a customer through a step) through Execute (complete tasks on their behalf) to Orchestrate (reason across a multi-step job), applied to workflows ranging from connecting a bank to setting up payroll tax.

Guidance only works if customers trust it before they need it. We focused on conversational, step-by-step instructions that reduce cognitive load and build willingness to try AI help — with success measured by containment rate, customer satisfaction, and contact deflection.

First-time users often hesitate to try a new AI-guided flow — uncertainty about what will happen, whether they'll lose control, or whether it will actually help. A dedicated first-time-user introduction became essential: preview the experience before they commit, set expectations for how the agent will guide them, and give enough context to overcome reluctance and build trust. The hero above shows the in-context entry point; this walkthrough is the full onboarding flow.

The ideal-state prototype shows the core guided walkthrough in action: the agent meets the customer in context, walks them step-by-step through a real workflow, and keeps them oriented without leaving the product. Before, a customer reads generic steps and calls support; after, the agent guides them through step three of five with account-specific context. This is the near-term experience — guidance that feels patient, clear, and in control.

The future state pushes further: the agent acts on the customer's behalf, completing tasks inside the product instead of only describing them. When it hits a question outside its expertise or an issue it cannot resolve, it does not leave the customer stranded — it connects to a human expert with full context so the handoff feels seamless rather than like starting over.

I documented how the Guidance Agent should behave across workflows — interaction patterns, escalation rules, and the criteria for when the agent guides versus executes. This gives product and engineering a shared reference as the capability moves from provocation toward implementation. Open question we left on the table: at what point does guide become execute, and who decides?

We were able to test the experience with a limited version to real users — enough to validate the core model. Company reprioritization paused the full experience before it could scale. The broader industry has since moved the same direction: AI agents that act on a user's behalf, not just answer questions. We were early to that bet.