AI customer service
An AI customer service agent that knows when to hand off.
Answers your customers from your own help docs and order data, in chat or email, around the clock. When it isn’t sure, or a customer asks for a person, it hands off with the whole conversation attached.
Prefer to write? Send your requirements
In short
PutThrough builds AI customer service agents in about a week, at a fixed price agreed in writing after a free scoping call. The agent answers from your help center and policies with sources, looks up orders or accounts through tools you approve, works in your website chat or support inbox, and hands off to your team when it isn’t sure or when a customer asks for a person. It is tested against real past tickets before launch, logs every conversation, and runs under a monthly cost ceiling. Refunds, cancellations and anything else that moves money wait for a person.
01Who it’s for
For support teams answering the same questions every day.
-
Online stores
Where’s my order, can I return this, do you ship to me: answered from your order data and policies, at any hour.
-
Software companies
How-to, account and billing questions answered from your help center, with tricky ones routed to the right person.
-
Service businesses
Hours, pricing, availability and booking questions answered after hours, so nothing waits until morning.
02Example uses
What a one-week support agent usually does.
-
Order status and returns
Looks up the order, explains where it is and what your return policy allows, and hands off anything that needs a refund.
-
Answers from your help center
How-to, account and policy questions answered from your own docs, with the source linked in the reply.
-
After-hours and overflow cover
Answers when your team is offline or busy, and leaves a clean handoff for the morning.
-
Ticket triage and drafted replies
Labels and routes incoming tickets and drafts a reply for your team to check and send.
03What fits in a week
One support agent, with guardrails.
Fits in a week
- Answers from your help center and policies, with sources
- Order or account lookup through one or two tools
- Your website chat, or your support inbox or help desk where it has an API
- Handoff to a person, with the conversation attached
- A test set built from your real past tickets
- A monthly cost ceiling and a log of every conversation
Needs longer, or a call first
- Refunds, cancellations or payments without a person approving
- Phone calls, which are a separate voice agent build
- Replacing your help desk software
- Formal compliance programs such as HIPAA or SOC 2
04A sample week
Seven days, one support agent.
The same seven days as every build, with this product’s work in them. Your week is planned on the scoping call.
- Day 0
Scoping call and fixed quote
Which questions it answers, which it hands off, and which policies it must quote exactly.
- Day 1
Help center, tools and past tickets
Your help docs and policies loaded, order lookup connected, and a test set built from real past tickets with the right answers.
- Days 2–5
Build, in the open
The agent, its handoff rules and its tone, scored against the test set every evening, with a preview your team can try.
- Day 6
Harden
Upset customers, attempts to trick it and policy edge cases tested; the cost ceiling and alerts switched on.
- Day 7
Launch and handover
Live on a share of conversations first, then wider as the results hold. The handover covers adding answers and reading the logs.
05What you get
What you walk away with.
- The agent, live in your chat or support inbox
- The test set built from your past tickets, and the results it passed
- Handoff to your team with the conversation attached
- A monthly cost ceiling and a log of every conversation
- A guide to adding answers and changing its rules
Plus the standard handover every build ends with: source code, a written handover, a recorded walkthrough and every account in your name. The full Day 7 list
06Stack
Your help center in, your team in the loop.
The agent reads the help docs and policies you already have, so there is no second knowledge base to maintain. The model sits behind an interface you can swap as prices change.
- Models
- Claude or OpenAI, behind one interface
- Knowledge
- Your help center and policies, in Postgres with pgvector
- Channels
- Website chat, or your support inbox
- Quality
- A test set from real tickets, tracing and cost limits
07Pricing
A fixed price to build, and a ceiling on what it costs to run.
Your exact price is fixed in writing after the scoping call, before you commit to anything.
After launch, the care plan is from $499 per month.
Prefer to write? Send your requirements
08Questions
01 Will customers know they are talking to AI?
Yes. Replies are labeled as AI, and a person is always one step away. Being upfront is what keeps customers willing to use it.
02 Should we build one or buy the one in our help desk?
If your help desk’s built-in AI covers your questions, buy it, and we will say so on the call. Build when the agent needs your own systems, such as order or account lookup, exact policy answers or a handoff your tools don’t offer, or when per-resolution fees would add up.
03 What stops it making things up or promising refunds?
It answers only from your help center and policies, shows its sources, and is tested against real past tickets before launch. Refunds, cancellations and anything that moves money wait for a person to approve.
04 How do we know it is working?
Every conversation is logged with whether it was answered or handed off, so you can see how it is doing from the first day. It starts on a share of conversations and widens as the results hold.