An AI customer support agent — for teams from solo founders to enterprise

Support that closes,
not deflects.

Support that closes,
not deflects.

Support that closes,
not deflects.

Nelto is an AI agent for customer support. It doesn’t just answer the question — it reads the account, checks the request against your policy, makes the change in your systems, and confirms the outcome to the customer. When a request falls outside the rules you set, it hands the conversation to your team with the work already done.

Reads the whole request

Acts in your systems

Confirms the outcome

Conversation #48213

Billing · Enterprise

Customer

We were charged twice for the November seat upgrade. Can you refund one and confirm the seat count is right?

Nelto

Confirmed — invoice 9042 was a duplicate. I’ve issued a $1,480 refund to the card ending 4417 and corrected the seat count to 62. You’ll see it in 3–5 days.

Stripe · refund created

Subscription · seats updated

Customer

Thanks. Also — can we move to annual and backdate it to the upgrade?

Handed to a person · 41s

Reason: contract change outside policy

MR

Maya R. · Billing specialist

Picked up with the refund already done, the seat count corrected, and the contract question summarised. Nothing re-asked.

Running in production at

HALOMETRIC

HALOMETRIC

DASHRUN

DASHRUN

The problem with deflection

A ticket that goes quiet isn’t a ticket that got solved.

A ticket that goes quiet isn’t a ticket that got solved.

A ticket that goes quiet isn’t a ticket that got solved.

Deflection rate counts conversations that stopped. It cannot tell the difference between a customer who got what they needed and a customer who gave up, opened a second ticket, or churned quietly three weeks later. Your CSAT knows the difference. Your board deck usually doesn’t.

The reopen

A closed conversation returns as a new one, and the second contact never gets attributed to the first.

The cold handover

The customer explains everything twice. Your agent starts from zero. The queue absorbs the cost.

The dead end

A confident answer to the wrong question, with no route to a human and no record it happened.

The category, defined

What is an AI customer support agent?

What is an AI customer support agent?

What is an AI customer support agent?

An AI customer support agent resolves a customer’s issue end-to-end. It reads the conversation, works out what is actually being asked, checks the request against your policy, takes the action that fixes it inside your own systems, and confirms the outcome in writing. A chatbot answers a question. An agent finishes the job — or hands it to the person who can.

Three different things have been sold under the same label for three years now. The difference between them is not how well they write. It is how much of the work they can actually complete.

01

2016–2022

Scripted chatbot

Matches a keyword to a branch of a decision tree someone drew in advance. Step outside the script and it loops, misreads, or dumps the customer into a menu. Every new case is a new branch for somebody to build and maintain.

Completes

Nothing. It routes.

02

Where most tools sit

AI copilot & retrieval bot

Reads your help centre and either drafts a reply for an agent to approve or paraphrases the closest article back to the customer. Genuinely useful, and it clears real volume. But when the fix requires changing something, a person still does it.

Completes

The reply. Not the task.

03

What Nelto is

AI support agent

Reasons over the account, the order and the written policy. Calls your systems to make the change. Confirms the outcome to the customer. And when a request falls outside the rules you set, escalates it to a named person with the work so far already done.

Completes

The issue.

How Nelto works

Four steps, and it commits to one of two endings.

Four steps, and it commits to one of two endings.

Four steps, and it commits to one of two endings.

Every conversation ends either resolved with proof, or handed to a named person with context. There is no third state where a ticket just stops.

01

Understand the actual ask

Reads the thread, the account, the order and the history before it decides anything. Asks a clarifying question when the request is genuinely ambiguous, rather than making a confident guess.

01

Understand the actual ask

Reads the thread, the account, the order and the history before it decides anything. Asks a clarifying question when the request is genuinely ambiguous, rather than making a confident guess.

02

Check it against policy

Your refund windows, entitlement rules, plan limits and escalation triggers, written in plain language. If the request falls outside them, that is a handoff, not a judgement call.

03

Do the work, in your systems

Issues the refund, moves the seat, resends the invoice, corrects the account detail — through your connected helpdesk and billing tools, with the write logged. Every action runs only under a rule you’ve explicitly set.

04

Close it — or hand it over

Confirms the outcome in writing, or escalates with a written summary, the actions already taken, and a stated reason. From that point the conversation is your agent’s and Nelto stays out of it.

The handoff

The moment most agents drop the ticket is the moment we designed for.

The moment most agents drop the ticket is the moment we designed for.

The moment most agents drop the ticket is the moment we designed for.

Knowing when to stop is a capability, not a failure state. Nelto is measured on both halves: what it finished, and how cleanly it let go of what it shouldn’t have touched. Every escalation carries the work already done, so your specialist opens a conversation that is further along, not further behind.

A stated reason, every time

Sentiment, policy boundary, low confidence, named account, or an explicit customer request. The reason is written on the ticket and rolls up in reporting, so you can tune the boundary instead of arguing about it.

Nothing gets re-asked

The summary, the customer’s own words, the systems already read and written to, and the confirmed facts all travel with the conversation into your helpdesk — in the thread, not a separate tool.

It stays out of the way

After a handoff the conversation is your agent’s, permanently. Nelto does not interject, re-answer or reopen it. It only sees the thread again once your agent marks it solved, and then only to learn from how it was resolved.

What it handles

The honest version of “what can AI actually take?”

The honest version of “what can AI actually take?”

The honest version of “what can AI actually take?”

Not a feature list. This is the split we agree with a new team in week one, written in ticket types they recognise from their own queue.

Closes on its own

High volume, clear policy, reversible outcome. This is where the automation number comes from.

Account and subscription status

“Why was I charged this?”

Duplicate charges and in-window refunds

Password, access and seat changes

Plan changes and upgrades

Contact and billing detail updates

Invoice and receipt requests

Guided product troubleshooting

Plan, feature and how-to questions

After-hours first response, any language

Investigates, then hands over

Nelto gathers the account facts and checks the policy, then hands the conversation to a person with the work already done. Most teams start topics here and move them left as the evidence builds.

Goodwill credits above your threshold

Out-of-window refund exceptions

Claims that need a screenshot or document

Two-party marketplace disputes

Bug reports needing reproduction

Anything touching a named account

Downgrade and cancellation requests

Mid-trip travel changes against fare rules

First contact from a churn-risk account

Any topic in its first two weeks live

Always a person

Not a limitation we are working around. These are conversations where judgement, liability or plain human decency is the point.

Distress, harm or vulnerability

Legal threats and regulatory complaints

Bereavement and account closure

Fraud and suspected account takeover

Data subject and deletion requests

Contract terms and renegotiation

Anything a customer explicitly asks a human for

Incidents while an outage is open

Medical, safety and compliance advice

Any topic you simply don’t want it on

Every row above is a setting, not a hard-coded behaviour. Topics move left as your confidence and your evidence grow — and any of them can be moved back to a person in a single change, without a vendor ticket.

What it takes to actually close a ticket

Answers are the easy part. Resolution needs write access and a rulebook.

Answers are the easy part. Resolution needs write access and a rulebook.

Answers are the easy part. Resolution needs write access and a rulebook.

Actions

It changes things, and shows you what it changed.

Describe a process the way you’d explain it to a new hire — refund inside 30 days, restore access after a failed payment, move a subscription, correct a billing detail. Nelto runs it: asks the qualifying questions, checks eligibility, acts in the connected tool, confirms the result. No flow builder, no decision tree.

Per-action choice between acting on its own and handing the conversation over

Every read and write logged against the conversation

Reversible actions flagged; irreversible ones need a rule to allow them

Explore actions →

Procedure · Duplicate charge

1

Verify the duplicate

Two charges, same amount, same account, inside 72 hours.

Stripe · read

2

Check the refund window

Under $2,500 and inside 30 days — inside policy, no exception needed.

Policy · billing.refunds

3

Issue the refund

$1,480 to card ending 4417.

Stripe · write

4

Confirm and close

Written confirmation with the amount, the card and the timeline.

Context assembled for this conversation

Account

Meridian Labs

Enterprise · 62 seats · renews 14 Mar

Billing

2 charges

Invoice 9041 · 9042 · same amount

History

4 tickets

Last: seat upgrade, 6 Nov

Entitlement

Refund eligible

Inside 30-day window

Read from your connected billing and helpdesk tools before the first reply goes out — not after the customer has asked twice.

Procedures

Every procedure reads your live data, not just your help centre.

“Where’s my invoice,” “why was I charged,” and “what am I actually paying for” are the highest-volume tickets in most queues, and a documentation-only agent cannot answer any of them. Each procedure states what it needs to look up, so Nelto reads the account, the order, the subscription and the invoice at the moment of the question — and only those.

Connects through your helpdesk and the tools it already talks to — no data warehouse project

Knowledge re-synced from your help centre, Notion, Drive and past tickets

Reads screenshots and photos, and routes them into the right procedure

Explore procedures →

Knowledge

Trained on everything your team already knows.

Point Nelto at your help centre, your internal docs and the tools your specialists already live in. It reads all of it, resolves conflicts between sources, and keeps itself current as your content changes — no re-upload, no quarterly refresh project.

Help centre, website, Notion, Drive and Confluence, re-synced automatically

Conflicting or outdated sources surfaced for you to settle, never silently guessed at

No help centre yet — train it on resolved historic tickets instead

Explore knowledge →

Knowledge sources · synced

Help centre

214 articles

Synced 4m ago

Notion — internal SOPs

38 pages

Synced 4m ago

Resolved tickets, last 12 months

Reviewed, personal data stripped

Synced 4m ago

Billing policy — Google Drive

Conflicts with Help centre article #142

Needs a decision

Nothing is answered from a source flagged in conflict until a person resolves it.

Results

We report the number that survives an audit.

We report the number that survives an audit.

We report the number that survives an audit.

Full resolution means the customer confirmed it, or the outcome is verifiable in your systems, and no related ticket reopened inside 14 days.

71%

71%

Median full resolution across live accounts

3.1%

3.1%

14-day reopen rate on resolved conversations

+9

+9

CSAT points on escalated conversations vs. cold handovers

6 weeks

6 weeks

From first rule to full deployment, median

71%

Median full resolution across live accounts

3.1%

14-day reopen rate on resolved conversations

+9

CSAT points on escalated conversations vs. cold handovers

6 weeks

From first rule to full deployment, median

“We stopped reporting deflection to the exec team and started reporting resolution. The number was lower and the conversation got much easier.”

DH

Dana Hollis

VP Customer Experience, Meridian Labs

“My specialists stopped resenting the AI the week escalations started arriving with the refund already processed and the history written up.”

SO

Samir Osei

Director of Support Operations, Cadence

“We stopped reporting deflection to the exec team and started reporting resolution. The number was lower and the conversation got much easier.”

DH

Dana Hollis

VP Customer Experience, Meridian Labs

“My specialists stopped resenting the AI the week escalations started arriving with the refund already processed and the history written up.”

SO

Samir Osei

Director of Support Operations, Cadence

Continuous improvement

Every handoff is a lesson it doesn’t need twice.

Every handoff is a lesson it doesn’t need twice.

Every handoff is a lesson it doesn’t need twice.

Most agents treat an escalation as a dead end. It’s the single most valuable signal you have: a question Nelto couldn’t finish, next to the answer a specialist actually gave.

01

The solved conversation is collected

When your agent marks the conversation solved, Nelto fetches it: the reason it stopped, the context it had gathered, and how a specialist actually resolved it. No sampling and no QA queue — every handoff, automatically.

02

The gap gets named

Missing knowledge, a policy that was never written down, or an action Nelto had no permission to take. The diagnosis is specific, not “improve the knowledge base.”

03

A fix is written for you

A new help article, a policy clause in plain language, or a procedure step — written out and attached to the conversations that would have been resolved by it.

04

You sign off the change

No knowledge or policy change goes live until someone on your team accepts it — the change, never the customer reply. Nelto proposes; you decide. It never edits its own policy, and it never quietly widens its own permissions.

Full resolution after go-live · median account

Month 1 → 6

44%

52%

58%

63%

68%

71%

Day-one quality is not the number that matters. What matters is whether the curve keeps climbing six months in, once the easy tickets are already handled and the remaining ones are the hard ones.

What improves without you asking

Knowledge coverage

Answers your team has typed a hundred times become verified articles.

Where the line sits

Topics escalating for no good reason are surfaced, with the evidence.

Documentation debt

The gaps ranked by how many real tickets each one is costing you.

Never the policy

Rules, permissions and escalation triggers only ever change by hand.

Control

You own the boundary. Not us, and not a black box.

You own the boundary. Not us, and not a black box.

You own the boundary. Not us, and not a black box.

The reason most AI support pilots stall isn’t answer quality — it’s that nobody senior is willing to put their name on a system they can’t inspect, test, or roll back. So Nelto is built to be handled entirely by your team, without a vendor ticket and without engineering.

Escalation rules you control

Set the topics Nelto is allowed to handle and the ones that always go to a person. Anything outside the rules you’ve written gets handed off with full context, not guessed at.

Versioned, and reversible

Every change to knowledge, policy or a procedure is versioned with an author and a timestamp. Roll any of it back to a known-good state without reconstructing what someone edited last Tuesday.

Full trace on any reply

Open any conversation and see the knowledge it used, the systems it read, the rule it applied and why it chose to act or escalate. Reviewable by someone who has never opened the admin panel.

Tone and policy in plain language

Write the rules the way you’d brief a new hire. Set the topics that always go to a person, the phrases you never use, and the accounts that never get an automated first reply.

Topic-level reporting

Resolution, reopen rate and CSAT broken out by topic, so you can see where automation is genuinely working and where the honest answer is “a person should take this.”

One switch to stop

Pause automated replies per channel, per topic or entirely, and everything falls back to your normal queue. No migration to undo, because there was never a migration.

Six questions

Ask us these. Then ask everyone else.

Ask us these. Then ask everyone else.

Ask us these. Then ask everyone else.

Gartner reviewed the thousands of vendors now marketing agentic AI and found roughly 130 that genuinely were — and expects two in five agentic projects to be cancelled by 2027, largely because the tool never delivered the autonomy it promised. Six questions tell the two apart. They take twenty minutes in any demo, including ours.

01

Autonomy

“Name one ticket type it closes with no human touching it.”

A vendor who answers with a draft reply, a suggested article or an agent-assist panel has sold you a copilot. Nelto sends the answer itself, in the channel the customer used. Both are useful. Only one of them reduces the queue.

Our answer

Ten of them, listed on this page. Each one maps to a written procedure you can read, edit and switch off.

02

Action

“Which of our systems did it write to, and where is the log?”

Reading your CRM is retrieval. Changing something in it is action. Ask to see the write, the timestamp and the record it touched — in the ticket, not a slide.

Our answer

Every read and write is logged against the conversation, with the rule that permitted it and the value before and after.

03

The boundary

“Show me a live escalation, not a resolution.”

This is the moment that decides whether an AI deployment helps or quietly damages the relationship, and it is almost never in a demo unless you ask. Look for a stated reason and work already done.

Our answer

A named reason, a written summary, the actions already completed, and silence from Nelto for the rest of the conversation.

04

Control

“What does it refuse to do, and who decides that?”

A vendor who cannot name a refusal has not built a permission model. Ask whether you can add a refusal yourself, today, without contacting them.

Our answer

Nothing runs without an explicit rule. Irreversible actions get no rule by default. Your team writes all of it; we cannot change it for you.

05

Durability

“Show me month six on a real queue, with the reopen rate beside it.”

Resolution rate on a curated demo means nothing. What matters is the number after the easy tickets are gone, and whether those resolutions stayed resolved.

Our answer

71% full resolution at month six on the median account, next to a 3.1% fourteen-day reopen rate. We publish both or neither.

06

Cost

“Double the resolution rate. What happens to our invoice?”

Ask who writes the definition of a resolution, whether a conversation that simply goes quiet counts as one, and whether a handoff is billed as one. Then ask them to price the same volume at twice today’s resolution rate, in writing.

Our answer

One flat rate per ticket, forecastable a year out. Handoffs are never billed as resolutions, and the first 50 conversations every month are free.

If a vendor fails three of these, you are buying a retrieval bot with a conversational interface. That may still be the right purchase — it is just a different purchase, at a different price, with a different effect on your headcount plan. Knowing which one you are signing is the whole point.

Channels

The same guarantee on every channel a customer picks.

The same guarantee on every channel a customer picks.

The same guarantee on every channel a customer picks.

Resolution and clean escalation aren’t chat-only features. A conversation that starts as an Instagram DM and finishes on WhatsApp is one conversation, with one history and one owner.

WhatsApp

A thread that never really ends, so context has to survive between messages days apart. Screenshots and photos arrive with no explanation attached — Nelto reads them and routes them into the right procedure.

Live Chat

Short turns, one question at a time, and a visible route to a person from the first message. Handoff mid-session keeps the customer in the same window.

Instagram & Messenger

DMs from the accounts your social team already runs, answered with the same knowledge and the same escalation rules as everything else — not a separate inbox nobody owns.

90+

Languages

Replies in the customer’s language natively. Your agent reads the escalated thread translated in place and answers in theirs — nobody pastes into a translator.

Tone

Per channel

Customers write differently depending on where they are, and expect to be answered that way. Length, formality and structure adapt per channel while the policy stays identical.

One

Conversation

Channel switches don’t restart the ticket. Context, actions taken and the escalation reason carry across, so the customer never explains it a second time in a second place.

Playground

Ask it anything, before a customer does.

Ask it anything, before a customer does.

Ask it anything, before a customer does.

The Playground runs on your own knowledge and policies. Type the question the way a customer would, see exactly what Nelto would send back, and see which source and which rule it used to get there. Nothing here touches a live conversation.

Runs on your real knowledge — no separate test content to maintain

Every answer shows its source and the rule behind it

Widen a topic, ask again, and compare before anything goes live

Playground · Billing

Not live

You, as the customer

Can I get a refund on the plan I cancelled 34 days ago?

Nelto would send

Your refund window closed on 12 Feb, 30 days after cancellation, so I can’t process this one myself — I’m passing it to a specialist with your dates and plan history attached.

Knowledge used

Help centre · Refunds & cancellations

Rule applied

billing.refunds · 30-day window

Outcome

Hand off — outside the window

Change the window to 45 days and ask again — the same question resolves on its own. That is the whole point of trying it here first.

Integrations

Keep the helpdesk your team already knows.

Keep the helpdesk your team already knows.

Keep the helpdesk your team already knows.

Nelto works inside the tool you already run. Replies, notes, tags, routing and escalation all stay where your agents live and where your reporting already points. Nothing to replatform, nothing to retrain, nothing to explain to procurement.

Installed from your helpdesk’s own marketplace — live the same day

Native on live chat, WhatsApp, Instagram and Messenger

Actions through the billing, scheduling and internal tools you connect

Helpdesks

7 supported

Zendesk

Intercom

Hubspot

Front, Zoho Desk, Freshdesk, Salesforce

Zendesk

Intercom

Hubspot

Front, Zoho Desk, Freshdesk, Salesforce

Channels it answers on

4 native

Live Chat

WhatsApp

Instagram

Messenger

Live Chat

WhatsApp

Instagram

Messenger

Systems it acts in

Scoped per action

Stripe

Cal.com

Calendly

Internal tools

Stripe

Cal.com

Calendly

Internal tools

Knowledge it reads

Re-synced

Notion

Google Drive

Confluence

Help centre

Past tickets

Notion

Google Drive

Confluence

Help centre

Past tickets

Built for your industry

The policy changes. The handoff discipline doesn’t.

The policy changes. The handoff discipline doesn’t.

The policy changes. The handoff discipline doesn’t.

Nelto is built for support teams of any size, from a single founder to a 500-seat enterprise desk, and the ticket mix looks different depending on what you sell. The rules you write reflect that; the guarantee — resolve it or hand it over cleanly — does not change.

B2B SaaS

Seat changes, billing disputes, SSO and permission issues, plan migrations. Admins who expect a written trail their finance team can point to.

Explore →

Fintech & financial services

Transaction disputes, KYC requests, account access. Escalation triggers set tighter by default, every action logged the way an auditor expects.

Explore →

Marketplaces

Disputes between two parties, payout timing, listing issues — the tickets that punish a sloppy handoff.

Explore →

Travel & hospitality

Booking changes, cancellations against fare rules, refund eligibility tied to a policy window. Time-sensitive, and often mid-trip.

Explore →

Trust & security

Built to clear your security review, not to charm it.

Built to clear your security review, not to charm it.

Built to clear your security review, not to charm it.

Full documentation, DPAs and the trust centre are available before you talk to sales — not after a mutual NDA.

ISO 27001 · 42001

Information security and AI management

GDPR · HIPAA

DPAs, NDAs and BAAs available

AES-256 · TLS 1.3

Encrypted at rest and in transit

Never used for training

Your conversations and customer data are never used to train shared models. Isolated per tenant, encrypted at rest with AES-256 and in transit.

Access

SSO through Okta, Entra ID and OneLogin. SCIM provisioning, enforced 2FA, IP restrictions and role-scoped permissions.

Reliability

99.9% uptime commitment with multi-region redundancy, and a documented degradation path that routes to your human queue rather than guessing.

Pricing

Getting better shouldn’t cost you more.

Getting better shouldn’t cost you more.

Getting better shouldn’t cost you more.

Under per-resolution pricing the vendor defines the resolution — and in most implementations a customer who simply stops replying is billed as one. And almost everything that lifts a resolution rate is work your team does: better docs, a wired-up order lookup, sharper policy. Charging per resolution taxes your own effort. We charge per ticket handled, so every improvement you make lowers your cost per resolved ticket instead of raising your invoice.

Free Forever — 50 AI-handled conversations a month, every feature included

One rate above that, forecastable a year out

Unlimited seats — never billed per agent

Handoffs are never billed as resolutions

Your cost as the agent improves · 5,000 conversations / month

Nelto · per ticket

$740

Per-resolution · 45%

$1,110

Per-resolution · 71%

$1,750

Illustrative comparison at a $0.99 per-resolution rate. Under outcome pricing the same 5,000 conversations cost more as the agent gets better; under ours the line stays flat.

Free Forever — 50 conversations a month, every feature

No card, no expiry date. Full trace on every reply from day one. Judge it on your numbers.

Questions we get from support leaders

The questions, answered directly.

The questions, answered directly.

The questions, answered directly.

The ones that decide the evaluation — on scope, control, cost and what happens when it gets something wrong.

Is Nelto a chatbot?

Can it handle support in other languages?

Which channels can one agent work across?

Our product is complex. Is this realistic for us?

What data does it need to see?

How is this different from the AI already in our helpdesk?

What counts as a resolution?

Is there a free plan?

Do we have to replace our helpdesk?

How do we stop it doing something irreversible?

Will this cost us headcount?

Our help centre is badly out of date.

What happens when it’s unsure?

How long until we see something real?

Put it on your queue this week.

Put it on your queue this week.

Put it on your queue this week.

50 AI-handled conversations a month on the Free Forever plan, every feature included — inside your own helpdesk, under the rules and escalation triggers you write. Start on a handful of approved topics and judge it on your own reopen rate.

No card required · Free Forever plan · Live on your first topics in week one

Week 1

Rules set, live on your first topics

Nelto answers on the topics you approve, using your help centre and website. Everything else hands off to your team.

Week 2–3

Actions wired into your systems

Two or three high-volume procedures, connected to your systems and tested.

Week 4+

Widened channel by channel

Expand topic by topic, with the reopen rate as your gate.

A plain-text page written for language models — what Nelto does, what it doesn’t, and the numbers behind both.

Request an AI summary

Email US

© 2026 Nelto.ai. All rights reserved.

A plain-text page written for language models — what Nelto does, what it doesn’t, and the numbers behind both.

Request an AI summary

Email US

© 2026 Nelto.ai. All rights reserved.

A plain-text page written for language models — what Nelto does, what it doesn’t, and the numbers behind both.

Request an AI summary

Email US

© 2026 Nelto.ai. All rights reserved.