Skip to content
Velum
← All solutions

Customer support

Tickets resolved, not deflected.

Most support queues are the same thirty questions asked a thousand ways. This pipeline reads each ticket, identifies the customer and their order or account, finds the answer in your knowledge base and systems, and either resolves it or hands a fully prepared case to the right person. Money-moving actions wait for approval, and every answer cites its source.

60–70%
Tickets resolved by agents
< 5 min
First response
0
Refunds without approval

How it runs

Step through the pipeline.

Click a step or let it play. The ticket record on the right fills in as the agents work.

Step 01 of 08

Take in every channel

Email, chat, WhatsApp and web forms land in one queue. The agent detects the language, the customer and the intent.

Zendesk / Intercom / FreshdeskWhatsApp
Ticket record1 / 8 fields
Ticket
Email · Spanish · ‘My order arrived damaged’
Customer
—
Priority
—
Answer
—
Action
—
Reply
—
Escalation
—
Knowledge
—
  1. 01 · Take in every channel

    Email, chat, WhatsApp and web forms land in one queue. The agent detects the language, the customer and the intent.

    Tools: Zendesk / Intercom / Freshdesk, WhatsApp. Output: Ticket.

  2. 02 · Identify the customer

    The customer, their account, orders, plan and past tickets are pulled so the answer is about them, not generic.

    Tools: Shopify / CRM, Billing. Output: Customer.

  3. 03 · Classify and prioritise

    Intent, urgency and sentiment are scored against your SLAs. Anything legal, safety-related or angry is flagged for a person immediately.

    Tools: Velum agent. Output: Priority.

  4. 04 · Find the grounded answer

    The agent retrieves the relevant policy, help article or runbook and checks it against the customer's situation, citing the source it used.

    Tools: Knowledge base, Notion / Confluence. Output: Answer.

  5. 05 · Take the action

    Standard actions such as a replacement, a tracking update or an address change are carried out in the right system. Refunds and credits above a threshold are proposed for approval.

    Tools: Shopify / ERP, Stripe. Output: Action. A person approves before this step completes.

  6. 06 · Reply in their language

    The reply is written in the customer's language and your brand voice, with tracking links and next steps, not a template.

    Tools: Help desk. Output: Reply.

  7. 07 · Escalate with context

    Cases the agent should not handle go to the right team with the customer, history, attempted answer and a suggested resolution already written.

    Tools: Help desk, Slack. Output: Escalation.

  8. 08 · Close the knowledge gaps

    Questions the knowledge base could not answer are clustered weekly and turned into draft articles for the support lead to approve.

    Tools: Knowledge base. Output: Knowledge.

Impact

What changes when it runs.

Illustrative targets from comparable engagements. Your baseline is measured in the audit and the targets are agreed before we build.

MeasureBeforeWith agentsChange
Routine tickets resolved without a person0%60 to 70%Two thirds
Median first response6 hoursUnder 5 minutes−98%
Cost per resolved ticket$8 to $15$1 to $3 for AI-resolved−80%
Escalations with full contextRarelyEvery one100%

Hours saved and ROI

What it is worth to your team.

Start from our defaults for this pipeline, then move the sliders to your numbers. The maths is simple and shown.

Hours reclaimed18 hper support agent, every week

Based on routine tickets taking five to eight minutes each and two thirds being resolved end to end, leaving the team with complex cases and reviews.

Worked example

Eight support agents each spending 30 hours a week on routine tickets at a $32 loaded hourly cost, with 60 percent resolved end to end: about 144 hours a week back, roughly $240,000 a year, while response times fall from hours to minutes.

Gross capacity, not cash savings. Excludes implementation, software and supervision costs, which are quoted before we start.

Your numbers

8
30 h
$32
60%

An assumption to explore, not a promised result.

What that is worth

144 h

reclaimed every week

$239,616

value per year

624 h

every month

3.6 FTE

of capacity returned

Hours on repetitive work, per week

Today240 h
With agents96 h
Validate these numbers

Gross capacity estimate, not cash savings. Excludes implementation, software, and supervision costs. We validate assumptions in the audit.

Questions about this pipeline

01

Will customers know they are talking to an agent?

Yes, we recommend being open about it. Customers care about fast, correct answers, and a person is always one reply away.

02

What stops it making up an answer?

Answers must cite a source from your knowledge base or live systems. If none is found, the ticket is escalated instead of guessed.

03

Which help desks do you work with?

Zendesk, Intercom, Freshdesk, Gorgias, HubSpot Service and most others with an API, plus WhatsApp Business and email.

Want this running for your team?

We connect your help desk and knowledge base, run the agent in draft mode on real tickets, then release the categories it gets right.