Most helpdesks are not underpowered. They are unrevisited.

The pitch you usually get is that you are only using a fraction of your helpdesk. It is a good pitch because it is unfalsifiable and it makes you feel behind.

Here is the version we actually see when we open instances. The configuration was right the day it was built. Then the business changed and the configuration did not.

What that looks like in practice

Every one of these came out of a real instance.

• A trigger routing to a group that was dissolved in a reorganisation. Tickets still land there. Nobody watches it.
• Forty macros, six of which get used. Two of the six contradict each other and agents pick by habit.
• An IVR menu offering a returns department that closed two years ago.
• A required field added for one report that nobody runs any more, still slowing every ticket.
• An AI agent answering questions with content from a Help Centre article last edited in 2023.

Every one of those decisions made sense the day it was made. Then the team changed and nobody had time to revisit a rule that was working.

Why the three platforms fail differently

Zendesk fails by accumulation. It is the most configurable of the three, which means it holds the most history. Triggers, automations, views and Explore reports built by admins who have since left, all still firing.

Intercom fails at the AI layer. Fin does what its Guidance tells it, and most Guidance we open is either empty or a template somebody pasted in during onboarding. The mechanism is fine. Nobody wrote the instructions.

Gorgias fails on assumptions about what is included. Order Management is Shopify only. So is AI Agent. If you are on BigCommerce or Magento, the feature your demo showed you is a build, not a checkbox.

What we change first

Not everything. In order of what moves the number:

• Routing, because a ticket in the wrong queue is already late.
• The knowledge the AI reads, because an agent grounded in stale content is worse than no agent.
• Escalation, because containment without a clean handoff just moves the complaint.
• Macros, cut to the ones people use and made to agree with each other.
• Reporting last, once the fields underneath it mean something.

The numbers we will stand behind

We will not tell you a percentage before we have seen your tickets. What we can tell you is what our own builds have produced. Across the Gorgias instances we have configured, containment lands between 26 and 56%. The Zendesk help centre we built for The San Francisco Marathon reached 56% chat deflection in production and 67% the following year. Those are our measurements from our own work, not vendor figures.

Sixteen years of experience and 500 or more implementations behind that. What your number turns out to be depends on your knowledge quality, your escalation design, and how far you are willing to let the AI go.

The cheapest way to find out

A $500 audit reviews one instance in one to two weeks and ends in a ranked written fix list with effort estimates. On the walkthrough call we tell you which items your own team can run without us. If we do not think you should hire us, we say so there.

It is credited against any project booked within 30 days.