Delancy
SOCIAL ENGAGEMENT AGENT

Answer the comment and the message before the buyer moves on.

Messages and questions arrive on your accounts at all hours and get picked up when somebody remembers. This agent reads them as they land, answers what is in scope from your own material, and puts anything needing judgement in front of a person.

What the engagement agent does

Read it. Answer it. Escalate the rest.

Social messages get treated as marketing and behave like enquiries. Somebody asks a real question under a post, or sends a direct message asking whether you handle a particular kind of work, and it sits until the person who checks that account next opens it. No firm would run a phone line that way, and the standard on social is quietly far worse than the standard on the phone.

The agent connects to the accounts you nominate and reads what arrives across all of them. Questions within the scope you agreed get answered from your own material. Anything outside it, and anything reading as a complaint or a dispute, goes to a named person with the thread attached and nothing said in the meantime. That limit is fixed, because an automated reply to somebody who is unhappy, in public, is how a small problem becomes a screenshot.

It also works to the rules the platforms impose rather than around them. There are windows inside which a business may reply to a message, and a thread left until Monday can simply be closed. Connecting through the official interfaces means the agent behaves the way the platform expects, which matters more in 2026 than it did, given how firmly the platforms moved against tools reaching them by other means.

This is scoped to social. Website conversations are the live chat agent, and phone and email enquiries are the AI receptionist. Where the posts generating the questions are the point, the employee advocacy agent covers what your people publish.

Capabilities

What it handles.

Comments and messages in one place

Everything arriving across the accounts you run, read in one stream rather than found by whoever happens to open each app. The platforms have been consolidating comments and direct messages into unified threads, and the agent works to that.

Answers drawn from your own material

Responses come from what your firm has actually said about its services, its process and its terms. Where the answer is not in the material the agent does not construct one, because a confident wrong answer in public is worse than a slow correct one.

Complaints routed to a person

Anything reading as a complaint, a dispute or a matter needing judgement goes to a named person with the thread attached rather than receiving an automated reply. This is a deliberate limit and it is where most automated social response goes wrong.

Covers the hours nobody is watching

Expectations do not adjust for evenings, weekends or the fact that your marketing coordinator works three days. A message arriving on Friday evening that gets a first response on Monday has usually already been answered by somebody else.

Works inside the platform rules

The platforms impose windows on when a business may reply to a message, and missing that window closes the conversation. The agent is built to those rules rather than around them, which is one reason it responds rather than queueing for the morning.

Runs on your own connection

Your own application registration and your own credentials, one firm to one deployment. Your accounts are not pooled inside a shared platform, and the threads it reads stay in your environment.

Who it is for

Built for firms answering in public, or failing to.

Firms whose enquiries arrive as a message

For a growing share of buyers the first contact is a direct message rather than a form or a phone call. If that surface is checked when somebody remembers, then the enquiry is being handled at a standard nobody would accept on the phone.

Client facing teams with a public account

Questions arrive publicly under your posts, where everybody reading can see whether you answered. An unanswered question sitting under a post for four days is read by more people than the post was written for.

Firms running several accounts

Two or three platforms, each with its own notifications, each checked by a different person or by nobody. Coverage is uneven by accident rather than by decision, and the gaps are invisible until something is missed.

Firms who have left a complaint visible

The cost of a slow response is not the delay, it is that it happened where prospective clients can see it. Speed matters most on exactly the messages people least want to answer quickly.

What is included

What you get.

A direct connection to the accounts you nominate, through the official interfaces, on your own credentials.

Comments and direct messages read in one stream across those accounts.

Answers drawn from your own material, with the scope of what it will answer agreed during the build.

Complaints, disputes and anything needing judgement escalated to a named person with the full thread.

Coverage through the hours and days your team is not watching, inside the reply windows the platforms set.

A log of everything answered and everything escalated, so the standard is visible rather than assumed.

How it gets built

How it gets built.

01

Agree the accounts and what each platform allows

Which accounts are in scope, and what the official interface on each one actually permits. Coverage differs considerably between platforms and it is better to know that at the start than to discover it after the build.

02

Draw the line around what it answers

The questions it handles, the material it answers from, and the categories that always go to a person. The boundary starts conservative on purpose, because widening it later is easy and pulling it back after a bad reply is not.

03

Test against messages you already received

We run it over real threads from the last few months and check what it would have said and what it would have escalated. The important test is not how much it answers, it is whether it recognised the ones it should have left alone.

04

Connect it and set the escalation route

Connection is made on your own application registration and credentials, one deployment to one firm. You decide who receives escalations, how fast they need to see them and where they land, so an escalated thread is not simply a different queue nobody checks.

Why Delancy

Against a notifications tab somebody checks on Monday.

Scoped to social, not to everything

This handles the comments and messages arriving on your social accounts. Website chat is a different surface with different expectations and it has its own agent. Two narrow agents that each do one thing beat one that half covers both.

Escalation is the design

The valuable part is not how much it answers, it is how reliably it recognises what it should not answer. Complaints and judgement calls reach a person quickly and with context, which is the opposite of an automated reply making things worse in public.

Connected the way the platforms intend

Through the official interfaces rather than the unofficial routes that drew enforcement across 2026. What it can do on each platform is what that platform actually permits, and we are specific about that before anything is built.

No ongoing dependency

The agent runs on your own credentials and your own application registration, with the price agreed before build. Your accounts and your conversations are not held inside a shared platform you have to keep paying to reach.

Frequently Asked Questions

Common Questions.

Scope. The live chat agent handles the conversation on your website, where somebody has already arrived and is asking about your services. This handles comments and direct messages on your social accounts, which are public or semi-public, arrive at all hours and carry different expectations. Firms often run both, and keeping them separate is what keeps each one good at its own surface.

No. Complaints, disputes and anything reading as a matter needing judgement are escalated to a named person with the thread attached. That is a fixed limit rather than a setting. An automated reply to somebody who is unhappy, in public, is how a small problem becomes a screenshot.

What each platform allows differs considerably, and we are specific about it during scoping rather than promising uniform coverage. Some provide full support for reading and replying to comments and messages, others provide much less. The honest position is that coverage is agreed platform by platform against what the official interfaces actually permit.

Only against rules you set, and everything it does is logged. Most firms restrict it to clearly abusive content and leave everything else visible, on the basis that hiding criticism is usually worse than answering it. The default is to escalate rather than to remove.

It says so and routes the question to a person rather than producing an answer. The scope of what it will address is agreed during the build, and the boundary is deliberately conservative at the start and widened once you have seen how it behaves on real traffic.

Pilots start from £2,500. The price is agreed before any build starts. Where the agent relies on third-party services to run, those operational costs sit with you and are made clear before build begins.

FREE ASSESSMENT

Is an AI agent the right shape for this?

Ten questions on the task, the volume and the output you need. You get a diagnosis of whether an agent fits, not a score.

Take the assessment

Find the oldest unanswered message on your accounts.

Book a discovery call. We will look at what arrives on your social accounts, how long it currently waits and what should never be answered automatically.

Book a discovery call →