Fullstory for Customer Agents: session context to resolve, not interrogate

Support teams don't fail because they aren't smart. They fail because every conversation starts blind. Fullstory for Customer Agents streams real-time behavioral context to your AI and human agents, so the first connection with the customer starts with already knowing what happened.

If you're in support, you've lived this moment; your deflection rate hit a wall about six months after you turned on your AI support agent.

The first stretch looked great. Password resets, billing questions, where-do-I-find-this. The agent handled them, volume dropped, and the business case wrote itself. Then the curve flattened. Not because the agent got worse, but because the tickets left over are the most complex to investigate and the right answer depends on what the customer was actually doing.

Those conversations still open the way every conversation opens: "Hi, what can I help you with?"

By then the customer has tried three times. They hit an error. They clicked the same button until it stopped feeling like their mistake and started feeling like the product's fault. And now something is asking them to recall every attempted action with precision.

Then it escalates, and your human agent opens the same ticket and starts the same conversation over.

That context gap is not an intelligence problem. Your agent reasons well, your knowledge base is comprehensive, and your support team is good at their jobs. But nobody in that conversation can see what the customer tried to do.

Our latest solution: Fullstory for Customer Agents closes that gap. It delivers real-time, AI-generated session summaries directly into your support stack, so the agent handling a conversation, human or AI, already knows what the customer did before they typed a word.

Why deflection plateaus and escalations drag

An AI support agent is a reasoning engine attached to a knowledge base. Both of those have gotten very good, very fast. Situational awareness has been the historic gap.

Without situational awareness, agents compensate the only way they can, by asking. Every question it asks to reconstruct the situation is a moment the customer can abandon the conversation, escalate, or decide the whole thing is not worth it. The questions are also the part customers describe as robotic, because a human colleague who could see the screen would never ask them.

Your human agents pay the same tax in a different currency. When a ticket reaches a person, the first stretch of their handling time goes to reconstruction: reading back through the thread, checking the account, asking the customer to walk through it one more time. None of that is solving. It is catching up. And the customer, who has now explained the same problem twice, is already unhappy before the real work starts.

David DiGiammarino, who leads partnerships at Fullstory, recently made the longer version of this argument: an AI agent is a brain, and a brain with no eyes has to open with a question. The same is true of a person handed a ticket with no context.

What is Fullstory for Customer Agents?

Fullstory for Customer Agents is a packaged offering that streams behavioral context from a live or recent session into whoever is handling the conversation. It sits on Anywhere: Activation and works with any support platform that can call an MCP server or an API.

That context arrives in one of two shapes, and which one you use depends on who is consuming it.

A session summary is a narrative recap of what the person did, generated by Fullstory and shaped by a prompt profile you define. It reads the way a colleague would catch you up. That is what a human agent wants when a ticket lands in their queue, and it is what Ninety.io feeds to Fin.

Session context is that same ground truth delivered as structured events rather than prose, transformed for a language model to consume directly. Fullstory does not write a summary; your own model reasons over the raw signal. That is what you want when you have built your own agent and you want it drawing its own conclusions with full situational awareness. No signal loss and event logs built for LLMs to scan through and reason with.

Most teams end up using both, and the agent decides which to reach for. A fast recap for a simple ticket. Full context when the case is tangled enough to reason through.

Either shape carries the same underlying record:

• What the person did in the session, in order, including navigation and the actions that matter to your product

• Errors they hit, including the ones your instrumentation never captured

• Frustration signals such as repeated clicks on an unresponsive element

• User, organization, location, and device context

• Any custom events you already send to Fullstory

How it works

Three steps, and only the third one takes any setup.

1. Capture

Fullcapture captures every interaction automatically from one line of code: every click, navigation, error, frustration signal, and custom event, across web and mobile. Nothing gets tagged in advance, which matters here more than anywhere else. Support conversations are about the thing that went wrong, and the thing that went wrong is almost never the thing someone thought to instrument last quarter.

2. Summarize, or hand over the raw signal

When a conversation starts, Fullstory either generates a summary of that user's session on demand (generate_session_summary) or returns the structured session context for your own model to work from (generate_context). What goes into a summary is yours to define. Prompt Manager lets you write prompt profiles that pull out exactly what your support team needs to triage, and you can return the result against a custom schema if your platform expects structured fields.

3. Deliver

The context reaches your agent at the start of the conversation over one of two paths. Fullstory for Customer Agents connects to any MCP-capable client and authenticates with OAuth. The API connector works with support platforms that expect a standard data connector. Either way, the same context follows the conversation, so when an AI agent hands off to a person, the person starts where the AI left off instead of starting over. This integration works seamlessly with any of the leading platforms on the market today: no rip and replace, just a better path to successful outcomes.

What changes for your human agents

The AI half of this gets the attention, but the escalated ticket is where handling time actually lives, and that is a human being's problem.

When a conversation reaches a person, the session context is already in the ticket. No hunting, no separate tab, no asking the customer to start over. The agent opens the queue and the first thing they read is what the customer did, what broke, and where they gave up.

That removes the reconstruction phase from the front of every hard ticket. It is the single largest recoverable chunk of average handling time in most support orgs, and it is pure overhead: the customer already lived it, the product already logged it, and the agent was rebuilding it by hand anyway.

It also changes what gets escalated. An agent who can see the actual failure writes a bug report an engineer can act on, instead of a ticket that says "customer reports checkout not working, unable to reproduce."

Why this isn't just another AI session summary

AI session summaries are becoming table stakes. Most experience analytics tools ship some version of one now, and if you are evaluating this category you should assume everyone will claim it. Two things separate a summary an agent can act on from a summary that sounds plausible.

The first is what the summary is built from. A summary assembled out of tagged events can only describe the moments someone predicted in advance. Everything between those events is inference, and inference is where support cases live: the spinner that never resolved, the validation error that flashed for half a second, the button that did nothing. Fullcapture captures the session completely, so the summary describes what happened rather than narrating around the gaps. For a support agent that difference is the difference between a correct first response and a confident wrong one.

The second is who decides what matters. A canned summary optimizes for a generic reader. Your support team is not a generic reader. Prompt Manager lets you shape the summary against your own taxonomy and domain expertise, so the agent gets the three things your team always needs to know and none of the things it never uses.

And when a summary is the wrong instrument, you are not stuck with it. An agent that needs to reason for itself can pull the full structured context instead.

What Ninety.io saw

Ninety.io runs a SaaS platform for team alignment and execution, and their support team was living the plateau. Their AI workflows lacked situational context, so customers repeated steps they had already taken and agents rebuilt user journeys before they could start resolving anything.

They used Anywhere: Activation to deliver Fullstory session summaries into Fin at the start of each conversation. Fin could then reference what the person had actually done inside the product. When a conversation escalated, the human agent inherited the same context.

Tickets that reached Fin with Fullstory context resolved 79% of the time, a 6% increase over those that did not.

"We all experience AI fatigue; AI interactions can feel less friendly. They lack the conversational warmth of a human touch. This is something we didn't realize we were missing. Fullstory adds a layer that makes Fin a more personable and friendly experience for our clients."
— Taylor Paletta, Director of Support and Digital Success, Ninety.io

Read Ninety.io's full story →

Where support teams start

Four places, in the order most teams get value from them.

1. Cut handling time on the tickets that need reconstruction. Most of an agent's time on a hard ticket goes to figuring out what happened, not to fixing it. Context at the start of the conversation removes that step for both AI and human agents.

2. Push deflection past the plateau. The tickets your AI agent currently escalates are mostly the ones where it could not establish what was going on. Give it that, and a portion of them resolve without a person.

3. Lift resolution rate, not just response rate. A meaningful share of tickets stall because nobody can confirm what the customer actually experienced, so they sit, bounce between queues, or close without a real fix. When the session behind the ticket is available from the first reply, fewer tickets end in a workaround and more end in an answer.

4. Resolve the first time correctly. Repeat contacts and escalations drive satisfaction scores more than response speed does. An agent that opens with an accurate read of the situation gets more first attempts right.

Availability

Fullstory for Customer Agents is generally available now. Enterprise and Advanced plans include a complimentary Activation quota to get started, and you can add quota as your usage grows.

Interested? Request a demo to get started or explore what else support teams can do with Fullstory.