- Item
- Ceramic vase, sand
- Total
- €64.00
- Carrier
- DHL Express
Photo shows damage on arrival
Spark puts AI agents on chat, email and voice that refund, rebook, reset and resolve, inside the rules you write and with every step on the record.

Playbooks in action
Spark follows your procedures step by step, calls the systems that hold the answer, and closes the loop with the customer. Pick a playbook to watch it run.
Photo shows damage on arrival
So sorry, Maya. I've refunded €64.00 to your Visa ending 4821, and there's no need to send it back.
Customer away until Saturday
Hi James, your parcel will now arrive on Saturday between 9am and 12pm. Reply CHANGE to pick another day.
Duplicate charge detected
Hola Lucía, hemos devuelto el cargo duplicado de 29,00 €. Lo verás en tu cuenta en 3 a 5 días.
Identity check required
You're verified, Aiko. A fresh reset link is on its way to your inbox and works for the next 30 minutes.
Asked about annual pricing
Welcome to Pro, Rohan! You're on the annual plan at €190 a year, with €11.40 credited for this month.
Above the agent's authority
Tom needs a claims specialist: kitchen water damage, about €6.2k. Photos and policy attached.
Capabilities
Answering is the easy part. Spark is built for the rest: acting in your systems, staying inside policy and knowing when to bring in a person.
Refunds, rebookings, plan changes and lookups, through the same APIs your team uses.
shopify.find_order#48213 · delivered200stripe.refund€64.00 · succeeded200zendesk.add_notecarrier claim200hubspot.updatelifecycle: customer200Each reply is checked before it's sent.
Detected in the first message and answered in kind.
Help centre, macros, past tickets and PDFs, kept in sync.
The same brain on chat, email, voice, WhatsApp and SMS.
When a person should take over, they get the whole story.

Summary for Ana
Water damage claim, €6.2k. Photos and policy attached. Customer prefers a call after 5pm.
The platform
One workspace for the people who write the procedures, the people who answer for the results and the agent that does the work.
1Build
Describe the procedure the way you'd train a new hire. Mention a tool and Spark calls it; set a limit and Spark won't cross it. Simulate hundreds of conversations before anything goes live.

2Launch
Turn on chat, email and voice from one configuration. Your team sees every conversation as it happens and can take over with a click.

3Improve
Every conversation is scored for accuracy, policy and tone. Spark spots the questions it can't answer yet and drafts the fix for you to approve.




Results
Median across Spark customers live for six months or more.
Integrations
Native connectors for helpdesks, commerce, billing and CRM, plus a typed API for anything in-house. Most teams connect their first three systems in an afternoon.
Connectors include Zendesk, Shopify, Stripe, Slack, Salesforce, Intercom, HubSpot, WhatsApp, Twilio, Notion, Jira, Gmail, Okta, Snowflake, Google Drive, Confluence.
Browse all 120+ integrationsCustomer stories

“We expected Spark to deflect the easy tickets. It ended up closing refunds and delivery changes end to end, and our team finally has time for the conversations that need a person.”
84%resolved end to end
0policy breaches in a year
“Compliance signed off because they could read the playbooks themselves. Spark writes the reply; the limits decide the money.”
Jonas LindqvistHead of Digital Service9×peak-day capacity
“Storm days used to mean a four-hour hold queue. Now the agent rebooks thousands of passengers before our morning shift logs on.”
Sofia AlmeidaDirector of Customer Care4 daysfrom contract to live
“We connected Shopify, Stripe and Zendesk on a Tuesday and had Spark answering live chat by Friday. The simulations meant no surprises.”
Marcus BellHead of Support EngineeringSecurity
Spark runs in your region, keeps your data out of model training and records every action with the rule that allowed it.
Card numbers, IDs and health data are masked before any model sees them.
Host in the EU, US or UK, or deploy privately in your own cloud.
SAML, SCIM and permissions down to a single playbook.
Every action, rule change and release, exportable to your SIEM.
Conversations never train shared models. It's in the contract.
Multi-region failover across cloud and model providers.
FAQ
Can't find what you need? Our team replies within a working day.
Contact usOr email hello@example.com
A chatbot answers questions. Spark finishes the job: it follows your playbooks, calls your systems to refund, rebook or update an account, and only hands off when a person is genuinely needed.
Most teams are live on one channel within a week. Connecting your helpdesk and knowledge takes an afternoon; the rest of the week goes on writing playbooks and simulating them against past tickets.
It hands the conversation to the right queue in your helpdesk with a summary, the customer's details and everything it already tried, so nobody asks the same question twice.
Spark routes each step to the model that suits it best across several leading providers, and fails over automatically. Enterprise plans can bring their own model or keep processing in their own cloud.
No. Your conversations stay in your workspace and are never used to train shared models. That commitment is written into every contract.
Per resolved conversation, not per seat. Hand-offs to your team are never billed. See the pricing page for plans and a full comparison.
From the blog

Language models are good at conversation and bad at policy. Here is how we split the two, and what it did to our error rate.

Reviewing fifty tickets a week felt rigorous. Scoring all of them showed us how much the sample was hiding.

People notice a pause on the phone at around a second. This is the latency budget we work to, stage by stage.
Bring a week of real tickets. We'll show you how many Spark resolves, and exactly how it got there.