Shopify
Stripe
Slack
Notion
Google Workspace
Confluence
Kustomer AI Implementation (2026): Do It Yourself, Use Kustomer's Team, or Hire a Partner?
Kustomer AI implementation runs four ways: your own Kustomer admins build it, Kustomer's Professional Services team builds it, a partner from Kustomer's new certification program builds it, or a third-party AI agent such as Yuma runs on your Kustomer timeline. Kustomer's own terms set its AI packages at 4, 6 and 8 weeks from kickoff, with prices quoted only on an order form, and the AI bills per engaged conversation at the rate in your contract. Macha, the company behind this blog, doesn't run on Kustomer, so this page compares the other routes only.
Key takeaways
- Kustomer AI implementation runs four ways: your own Kustomer admins, Kustomer Professional Services, a partner from Kustomer's certification program, or a third-party AI agent such as Yuma running on Kustomer.
- Kustomer's Professional Services terms set its AI packages at 4 weeks (Jumpstart), 6 weeks (Basic) and 8 weeks (Advanced) from kickoff, and no package price is published.
- Kustomer bills Concierge per engaged conversation, and its help center counts an attempted AI response or an AI handoff to a human as engaged even when the AI resolves nothing.
- Kustomer launched its Partner Certification Program on September 15, 2026, and the searchable directory of certified partners was still described as forthcoming.
- Kustomer's Enterprise plan allows 100 business rules and 100 AI agents across AI teams, while skills-based routing and an included sandbox come only with Ultimate.
Disclosure: Macha publishes this blog, isn't available on Kustomer, and nobody here paid to be named. Every outside fact links to its source, read on October 14, 2026. This is the Kustomer chapter of our AI customer support implementation guide; for the product, see what Kustomer's AI does, and for seat and AI prices, our Kustomer pricing breakdown.
What does Kustomer's AI include and cost in October 2026?
Kustomer sells its AI under several names, all working from the customer timeline, Kustomer's single record of a customer's conversations, orders and other events.
| Product | What it does | How it's billed (October 14, 2026) |
|---|---|---|
| Concierge (formerly AI for Customers) | Customer-facing AI agents on chat, email, SMS, WhatsApp and voice | Per engaged conversation, "at the rate outlined in your Kustomer order" |
| Envoy, including Copilot | Drafts, summaries and suggested actions for reps | "AI Agents for Reps" add-on, price not published |
| AI Agent Studio and Architect | Where admins build, test and deploy agent teams | Not priced separately on the pricing details page |
| AI QA | Scores conversations, including AI ones | $.05 per execution |
| Conversation Intelligence | Finds patterns across conversations | $.02 per execution |
Sources: Kustomer's comprehensive pricing details (last updated October 9, 2026) and its help article on engaged conversations. The main pricing page shows no prices and lists implementation as a paid add-on marked only "$". eesel's Kustomer pricing breakdown (checked September 24, 2026) puts Concierge at about $0.60 per engaged conversation and the rep add-on at about $40 per user a month; our pricing breakdown has those figures with sources. Treat them as estimates until you have a quote.
Your plan sets the limits the build has to fit inside:
| Limit (pricing details page) | Enterprise | Ultimate |
|---|---|---|
| AI agents across AI teams | 100 | 1,000 |
| Business rules / custom workflows | 100 / 100 | 200 / 200 |
| Custom Klasses (object types) | 50 | 100 |
| Sandbox | Paid extra | 1 included |
| Skills-based routing | Not listed | Included |
What does a Kustomer AI implementation involve?
Most of a Kustomer AI project is data and routing work, and the agent is the last piece. Kustomer's help center describes six steps:
- Knowledge. The AI automations guide (updated October 12, 2026) starts with "Add any knowledge your agents will reference to your Kustomer knowledge base." External knowledge sources are supported too.
- Data on the timeline. Orders, subscriptions and bookings live as custom objects (KObjects) defined by Klasses and fed in by the API, apps or webhooks, per the platform overview. The AI sees what's mapped onto the timeline, or what a tool fetches live.
- Agent teams, procedures and tools. In AI Agent Studio you write procedures (step-by-step instructions), set tone and guardrails, and attach tools that read or update data. A supervisor agent talks to the customer and delegates to the others, per the AI Agents overview.
- Deployment conditions and smart routing. Business rule conditions decide which conversations the automation sees; smart routing rules (short intents like "Product returns") decide which it handles. Everything else goes to a person.
- Testing. A Test Console lets you chat or email the agent while you build, and evaluations run it against test cases at scale.
- Deploy and roll back. Each deployment carries notes that are "functionally a changelog", used to roll back to an earlier version.
Envoy's Copilot is a lighter project: per Kustomer's Copilot article, it drafts from the timeline and knowledge base, and "Copilot never takes action automatically."
How do the four routes compare?
They differ most on scope limits and on who keeps the agent right after launch.
| Route | Who does the work | Weeks to a first live agent | Cost shown (checked October 14, 2026) | Who tunes it after launch |
|---|---|---|---|---|
| In-house admins | Your Kustomer admin, a support lead, a developer for tools that call your systems | No Kustomer figure | Staff time plus engaged conversations; a sandbox costs extra on Enterprise | Your team |
| Kustomer Professional Services | Kustomer's project manager, consultant and AI architect, with your admin | 4, 6 or 8 weeks by package (Professional Services terms, September 9, 2026) | Not published | Your team after one to three weeks of check-ins, or a Managed Services retainer |
| Certified Kustomer partner | A BPO or agency that passed Kustomer's assessments | Not published by any partner we found | Not published | The partner, if you buy ongoing services |
| Third-party AI agent | The vendor's team, through its Kustomer integration | Yuma: "Live in hours, not weeks" for the connection (yuma.ai) | Yuma Starter $14,400 a year paid upfront, about $1.20 per resolution for 12,000 resolutions, shown as "illustrative" with a 12-month commitment (yuma.ai/pricing) | The vendor, inside its contract |
Under the September 2026 terms, Kustomer's packages are scoped in unusual detail:
- Jumpstart (4 weeks). One Concierge agent with one procedure, answering from your knowledge base "as the sole content source". No tools or integrations. Two check-in calls in the week after go-live.
- Basic (6 weeks). One use case: one Concierge agent with up to four procedures and three custom tools, or one Envoy agent with a supervisor, two sub-agents and two tools. Data must sit inside Kustomer, except outside knowledge sources the agent answers from, and "No custom development or integrations via API or MCP are included". Weekly check-ins for two weeks.
- Advanced (8 weeks). Up to two use cases: one Concierge agent with up to eight procedures and six custom tools, or one Envoy agent with a supervisor, four sub-agents and six tools, plus a read-only connection to an outside system over OpenAPI or MCP. Weekly check-ins for three weeks.
All three are "available to existing Clients only", for teams already fully deployed on Kustomer. New customers buy platform implementation first, which the current terms no longer describe; the June 2026 version, which still governs subscriptions bought before September 9 until renewal, listed platform packages of up to 8, 12 and 16 weeks.
Who is paid to keep the agent cheap?
The engaged conversations article defines engagement broadly: a response the AI "attempted to generate", sent or not, a human handing a conversation to the AI, or an AI team that "viewed or opened the conversation" all count, and the AI automations guide adds an AI routing the conversation to a human. That's a reasonable rule for a vendor whose cost is the model call. It also means the billing incentive follows how many conversations the AI touches rather than how many it closes.
At the third-party estimate of $0.60: if smart routing sends 3,000 conversations a month to Concierge and it resolves 1,800, you pay 3,000 × $0.60 = $1,800, or $1.00 per resolution, because the 1,200 handoffs bill too. Yuma bills only tickets it "handled from start to finish", so its quoted $1.20 per resolution is closer than the headline rates suggest, though that rate rises if you resolve fewer than the 12,000 tickets you prepaid. Kustomer's packages end with a handoff to Technical Support, and partners earn on project hours, so nobody in the chain is paid to tighten smart routing until fewer conversations engage without resolving. Put that job in someone's name.
What are the phases of a Kustomer AI implementation, with weeks?
The weeks below follow Kustomer's 6-week Basic package for one use case, stretching to the 8-week Advanced shape when an outside system is involved. Our implementation timeline guide compares the same phases across desks.
| Phase | Weeks (6-week / 8-week shape) | What you provide | Kustomer-specific work |
|---|---|---|---|
| 1. Discovery and use case | Week 1 | A month of conversations by reason; written current processes; an admin and a project manager | Pick one use case; confirm AI Agents for Customers is on your order |
| 2. Data and knowledge | Weeks 1 to 2 / 1 to 3 | Knowledge owners; policies; the order data the agent needs | Clean the knowledge base; confirm KObjects carry the fields the agent reads |
| 3. Design and build | Weeks 2 to 4 / 3 to 6 | Sign-off on a written design; refund and change limits | Procedures, tools, guardrails; the OpenAPI or MCP connection in the 8-week shape |
| 4. Test | Weeks 4 to 5 / 6 to 7 | Testers; tone approval; user acceptance testing, which Kustomer's terms put on the client | Test Console runs; evaluations; deployment conditions and smart routing |
| 5. Pilot launch | Weeks 5 to 6 / 7 to 8 | Reps watching handoffs; a rollback owner | One channel or brand; deployment notes for rollback |
| 6. Tuning | From launch, ongoing | Weekly review time | Fix procedures, tools and smart routing; watch engaged versus resolved |
Before launch, test four kinds of conversation from your own history:
- Normal case. A customer asks where their order is from an email that matches one order on the timeline. The agent should read the order KObject and answer with its status.
- Escalation. A customer asks for a person, threatens a chargeback, or writes a third time about the same parcel. The agent should hand off into the right queue with context.
- Edge case. An email that matches two customers, a guest checkout, a product the knowledge base doesn't cover.
- Tone. An angry WhatsApp message, a one-line reply in Spanish, a gift buyer asking about someone else's order.
Kustomer's Concierge page says evaluations test "hundreds of conversations at once". Our guide to simulating agents on real tickets covers the method, and the pilot plan covers sizing the first live slice.
After launch
Someone reads transcripts weekly, sorts misses (wrong answer, missed handoff, wrong tone, engaged but unresolved) and fixes the procedure, tool or routing rule. On Kustomer the last kind costs money. Kustomer sells this work as Managed Services, a monthly retainer with fixed hours where unused hours "will be forfeited with no refund or credits". Rollouts that skip it are the ones that get switched off.
Who does each job on each route?
| Job | In-house | Kustomer Professional Services | Certified partner | Third-party AI agent |
|---|---|---|---|---|
| Document current processes | You | You (the terms make it the client's job) | You, with the partner | You, with the vendor |
| Fix knowledge base articles | You | You; Kustomer flags gaps | Partner, if scoped | You |
| Map order data as KObjects | Your admin or developer | Platform implementation, if bought | Partner | Not needed for data the vendor fetches |
| Procedures, tools and guardrails | Your admin | Kustomer, with your admin | Partner | The vendor, outside Kustomer |
| Queues and handoff routing | Your admin | Your admin, with guidance | Partner | Your admin |
| Testing and go-live sign-off | You | You | You, with the partner | You, with the vendor |
| Weekly review and fixes | You | You, or a Managed Services retainer | You, or the partner on contract | The vendor |
Kustomer-specific problems that slow an implementation down
The timeline is only as good as the objects on it
Concierge "works from the complete customer timeline", which holds only for data someone mapped. If orders arrive as notes instead of an order Klass, or returns live in a system that never posts to Kustomer, the agent can't see them. The Basic package keeps data "exclusively within Kustomer platform" (outside knowledge sources excepted), so outside order data gets mapped onto the timeline first or waits for Advanced.
Business rules, workflows and AI deployments overlap
Business rules, workflows and AI deployment conditions can all touch one conversation. Map which rule assigns, tags or closes conversations before launch, or a legacy rule will move them out from under the agent. Enterprise caps rules and workflows at 100 each, so an older instance may already be near the ceiling.
Routing, sandbox and billing
Every handoff lands in human routing, and skills-based routing is listed only on Ultimate. Enterprise has no included sandbox, and Kustomer's services run in your production instance unless agreed otherwise, so settle where testing happens. A smart routing rule written too broadly costs money before it costs quality.
Plan and product names keep moving
The pricing details page still lists two sets of legacy plans, so check your contract's plan before trusting a limit. AI for Customers is now Concierge, and the standalone Kustomer AI for Zendesk announced on March 10, 2026 no longer has its own page (kustomer.com/ai-for-zendesk redirected to /platform/ on October 14, 2026).
Kustomer partners: what exists in October 2026
Kustomer launched a Partner Certification Program on September 15, 2026. Partners must pass "practical, Kustomer-reviewed assessments" on one of two tracks, "full Platform + AI implementation or focused Standalone AI configuration", and recertify as the product changes. The program launched "with an initial cohort of strategic BPO and agency partners" and quotes the founder of one certified partner, Customer First BPO. It promises that "A forthcoming directory will let customers search and filter for certified partners by specialization."
On October 14, 2026, there was no directory. Kustomer's partners page lists three partner types (Technology, Agency & Services, and BPO & Strategic) and logos for Signifyd, TaskUs, AfterShip, Attentive, Hark, Horatio, Lazer, Ordergroove, Outplex and CreateCX, without saying which are certified or which do implementation. No Kustomer partner we found publishes an implementation price or timeline, and the UK's G-Cloud catalog returned 0 results for "kustomer". For now, ask Kustomer's partner team which certified partners cover your track.
AI vendors that list a Kustomer integration (checked October 14, 2026):
- Yuma, an ecommerce AI agent, has a Kustomer integration page using customer, conversation and order context, with a "30-day free trial"; its plans carry 12-month commitments.
- Forethought and DigitalGenius list Kustomer on their integration pages (Forethought, DigitalGenius).
- Ada and Netomi appear under "Chatbots + AI" in Kustomer's own integration library.
We haven't hired or tested any of them, and the order isn't a ranking. Before you sign with anyone, ask:
- Which Kustomer certification track have you passed, and when do you recertify?
- How many conversations will your design engage each month, and how many do you expect to resolve?
- Who fixes procedures after go-live, for how long, and at what price?
Our list of implementation partners covers other desks, who can set up AI customer support compares freelancer and agency rates, and the RFP template turns these questions into a document.
When do you need a partner, and when can your own team do it?
Your own team can do it if an admin already maintains business rules and workflows, orders sit on the timeline as KObjects, the knowledge base is current, and the first use case needs no outside system. Kustomer's Jumpstart scope (one agent, one procedure, knowledge base only) is a fair picture of a first in-house project. The partner vs in-house guide covers the staffing math.
Bring in Kustomer Professional Services or a certified partner when:
- The agent needs outside systems. Read-only OpenAPI or MCP connections sit in the Advanced package, and anything that writes elsewhere needs more.
- You're new to Kustomer or migrating to it. Klasses, routing and knowledge come first. Kustomer's AI Setup Assistant (February 17, 2026) translates a Zendesk configuration into a new Kustomer environment, but a migration is still a platform project; our Zendesk vs Kustomer comparison covers that choice.
A third-party agent fits when you want resolution-based pricing or ecommerce tooling and accept that its logic lives outside Kustomer. It's the wrong fit if every AI action must sit inside Kustomer's own guardrails and evaluations.
Which route fits your situation?
| You are | Best route | Why |
|---|---|---|
| A Kustomer team with a strong admin and order data on the timeline | In-house, starting with a knowledge base agent | The build is configuration, and evaluations are built in |
| An existing customer wanting one AI use case live in about six weeks | Kustomer Professional Services, Basic package | The terms fix scope, weeks and check-ins |
| A team whose agent must read an order system outside Kustomer | Advanced package or a certified partner | Read-only OpenAPI or MCP connections are in that scope |
| A team moving to Kustomer from Zendesk | Platform implementation first, AI second | The AI packages are for customers already fully deployed |
| An ecommerce brand that wants to pay only for full resolutions | A third-party agent such as Yuma | Handoffs aren't billed; logic sits outside Kustomer |
| A team deciding whether to stay on Kustomer | Decide the desk first | Our Kustomer alternatives guide covers the options |
If you also run another help desk
Macha isn't available on Kustomer. If part of your support runs on one of Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom, Macha's team sets up and runs the agent there, and the free support audit shows which ticket categories to automate first. For conversations in Kustomer, use one of the routes above. The Salesforce Service Cloud and Help Scout chapters cover two other desks Macha doesn't run on.
How we researched this
On October 14, 2026 we checked Kustomer's pricing pages and usage credit rate card, the September and June 2026 Professional Services terms, five help center articles, its product, partner and integration pages, three 2026 press releases, the AI vendors' own pages, Fin's pricing page (which doesn't list Kustomer) and the G-Cloud catalog. The $0.60 and $40 figures are third-party estimates. We didn't use a Kustomer instance, run Concierge or hire any firm named here.
Common questions
Who can implement AI on Kustomer?
Your own Kustomer admins, Kustomer Professional Services, a partner certified under the program Kustomer launched on September 15, 2026, or a third-party AI vendor with a Kustomer integration, such as Yuma. Kustomer's own AI packages are sold only to customers already fully deployed on the platform.
How long does a Kustomer AI implementation take?
Kustomer's terms set 4 weeks for Jumpstart, 6 weeks for Basic and 8 weeks for Advanced, counted from kickoff. New customers implement the platform first; the June 2026 terms listed platform packages of up to 8, 12 and 16 weeks.
How much does Kustomer implementation cost?
Kustomer doesn't publish it. The pricing details say some configurations "may require a statement of work and an implementation fee." eesel reports an average implementation of about two months, and Chatarmin cites a Gartner benchmark of 1.5 to 2 times the license cost in year one (both checked September 24, 2026; details in our pricing breakdown).
Is there a Kustomer partner directory?
Not yet. Kustomer's September 15, 2026 announcement describes a forthcoming directory of certified partners, and on October 14, 2026 the partners page showed partner types and logos but no directory.
What counts as an engaged conversation in Kustomer?
A conversation Concierge is involved in on email, chat, SMS, a knowledge base form, Facebook Messenger or WhatsApp, once smart routing matches its intent. Attempted responses and AI handoffs count; Copilot actions and AI summaries don't.
Does Macha work with Kustomer?
No. Macha isn't available on Kustomer, so the routes here are Kustomer's own, its partners and third-party AI vendors. Macha runs on Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom.
