Selected work

Problems we've actually solved.

A sample of real engagements across telehealth, e-commerce, HR-tech, software and agency support. Client names are withheld. Confidentiality comes before marketing.

Global HR-tech / employer of record

Two brands consolidated onto Zendesk, migrated off HubSpot

The problem. Support was running in HubSpot across two brands, with no single sign-on and no clean separation between what internal staff and external customers could see or submit.

What we did

  • Migrated support out of HubSpot into Zendesk, consolidating both brands in one instance
  • Implemented JWT single sign-on for silent login to the help center
  • Gated help-center forms so staff and end users each see only what they should
  • Rebuilt notification email templates on a dark-mode-safe skeleton
  • Stood up a change-management workflow with approval steps on Zendesk business rules

Outcome. One governed support platform across both brands, with staff and customer journeys properly separated and changes moving through a documented approval process.

MigrationSSOMulti-brand

Telehealth (GLP-1 / weight management)

AI first-response that never gives medical advice

The problem. A high volume of repetitive patient questions (shipping status, treatment check-ins) was consuming agent time, but the regulatory risk of automating replies on a clinical service is unforgiving.

What we did

  • Two-stage AI reply: one stage drafts an answer grounded strictly in approved knowledge-base content
  • A second stage reviews every draft for hallucination, sensitive data, and medical-advice creep before it can send
  • Clinically sensitive intents (side effects, symptoms) hard-blocked to a human regardless of confidence
  • Rolled out one intent at a time, each validated against real ticket history before going live
  • Every AI-handled ticket tagged so resolution and comeback rates stay measurable

Outcome. Repetitive contacts get answered automatically, while every clinically sensitive conversation still reaches a person. Coverage expands only when the data says an intent is safe.

AIRegulated supportAutomation

Multi-brand e-commerce (DE + UK)

Three brands, one helpdesk, always the right legal entity

The problem. Three consumer brands shared a single helpdesk, and replies risked going out under the wrong brand, and the wrong legal entity, across two languages.

What we did

  • Per-ticket signature switching driven by product and portal fields
  • English and German variants maintained in parallel
  • Consistent reply templates across all three brands

Outcome. Every reply automatically carries the correct brand, language and legal entity, with no agent having to remember which is which.

Multi-brandLocalizationTemplates

Technical software vendor

6,000+ documentation articles moved into Zendesk Guide

The problem. A large product documentation set lived in MadCap Flare and needed to become a maintainable Zendesk Guide knowledge base without losing structure or breaking cross-references.

What we did

  • Scoped against a measured crawl of the real source rather than an estimate
  • Automated the transform from Flare output into Guide-ready articles
  • Rebuilt the information architecture as categories and sections
  • Resolved every internal cross-link to its new destination

Outcome. Roughly 6,200 articles across 579 sections published one-to-one, with internal links intact and a structure the team can maintain.

MigrationKnowledge baseGuide

Health & nutrition coaching

A 280-article knowledge base built from a year of tickets

The problem. The company had no real knowledge base. Every answer existed only inside past support tickets and scattered website pages.

What we did

  • Clustered twelve months of resolved tickets to find the real recurring questions
  • Filtered out noise before any content was generated
  • Synthesized structured articles using a locally-run language model, so customer data never left their own infrastructure
  • Published into the help center to serve both self-service and the AI assistant

Outcome. 280 structured articles live, powering customer self-service and grounding the AI bot, built entirely on-premise for privacy.

Knowledge baseAIOn-premise

Content & SEO agency

Turning a helpdesk into a deal pipeline

The problem. Sales, fulfillment and payment were tracked informally inside a helpdesk never designed for it, on top of an inherited automation mesh nobody fully understood.

What we did

  • Designed a unified multi-stage pipeline covering lead → sale → fulfillment → payment
  • Added the custom fields and stage tracking to make conversion measurable
  • Audited the inherited automation mesh end to end
  • Root-caused duplicate replies, blank auto-responses, and a dead end where a whole class of tickets got no first response

Outcome. One coherent pipeline the team actually works in, and the silent reply failures eliminated.

FreshdeskProcess designAudit

Why no client names or logos? Because we don't publish a client's name without their written permission, and several of these operate in regulated or sensitive spaces. If you'd like references, we're happy to arrange them directly once we're talking.

Your turn

Got a problem that looks like one of these?

Start with a free Zendesk Health Check, a prioritized read on what to fix first, no obligation.

Request your Health Check →