Your best CS lead has seen this before.
Now your whole team has too.
The people who run customer success make hundreds of judgment calls a week, and the same problems keep coming back. Cortex remembers every call they made and why, so their team and your AI agents can ask them anything, any time, without asking them.
Illustrative example. Names and accounts are fictional.
The problem
The answer exists. It is in one person’s head.
Every CS team has one person everyone pings. They know which customer gets a credit and which gets a call, which escalation is real, and what was promised two renewals ago. None of that is in the playbook, so the whole team queues behind them.
of knowledge workers say the only way to get the information they need is to ask someone or book a meeting.
Atlassian, State of Teams 2025
of the knowledge a role needs lives only with the person in it. If they leave, nobody can do that part of the job.
Panopto, Workplace Knowledge Report
of service issues are fully resolved through self-service. The rest end up with a person.
Gartner, 2024
of people using AI at work bring their own tools. Your team does not share one assistant, so it cannot share one memory.
Microsoft and LinkedIn, Work Trend Index
How it works
Decision by Cortex. Action by your tools.
Cortex does not replace your helpdesk, CRM or agents. It sits underneath them as the one place where your lead's judgment lives, and answers whoever asks.
01
Capture the call, not just the ticket
The lead talks to Cortex the way they would brief a deputy: a short interview, then their calls, escalations, notes and the odd voice memo after a hard conversation. Cortex records what they decided, and the reasoning they would never type into a CRM field.
02
Remember problem, customer, decision
Every judgment is stored with its context. Problem A, for this kind of customer, at this stage of the contract, got fix B, because of C, and option D was rejected. Over a few weeks that becomes a working model of how this person decides.
03
Answer anyone, in the tool they already use
A rep, a manager or an AI agent asks "what would she do here?" Cortex answers in the lead’s voice, cites the past decisions it is leaning on, and says plainly when a case is new enough that the lead should look.
What Cortex keeps
Not a transcript. A decision, with its reasons.
Search tools find the conversation. Cortex keeps what was decided, for whom, why, and what was ruled out. That is what lets it answer the next case, not just recall the last one.
A day on the team
Four questions. One person used to answer all of them.
09:12 · New rep, week two
“Customer wants to move billing to a different legal entity mid-contract. Can we?”
Today
Waits for the lead to come out of standup. Customer waits too.
With Cortex
Gets the lead’s rule and the two times it was bent, with the exact paperwork. Replies in four minutes.
11:40 · Support, tier two
“Third outage complaint this quarter from a top-20 account. Escalate or handle?”
Today
Escalates by default. The lead’s afternoon goes to triage.
With Cortex
Cortex flags it as the lead’s pattern for a save play, drafts the apology the way she writes them, and loops her in only for the call.
15:05 · Claude, running the renewal queue
“30 renewals due next month. Which ones are at risk, and what is the play?”
Today
A generic health score that treats every account the same.
With Cortex
Claude asks the lead’s Cortex account by account. The decision comes from her history; the follow-up tasks get created by Claude.
18:30 · The lead is on leave for two weeks
“Everything above, every day, while she is offline.”
Today
Decisions stall or get made inconsistently. She returns to a backlog.
With Cortex
The team keeps deciding the way she would. She returns to a log of what her Cortex answered, and corrects anything she would have called differently.
Customer success
Eight calls your lead makes every week.
Start with whichever one costs the most interruptions. The rest follow, because it is the same judgment underneath.
#01
Escalation triage
Which issues the lead personally takes, which go to engineering, and which are handled with a template and a follow-up call.
#02
Refunds, credits and goodwill
When a credit is right, when a refund is, how much, and what to say so the customer does not read it as an admission.
#03
Renewal risk and save plays
The signals the lead actually trusts, and the play she runs for each account shape. Not a health score, a judgment.
#04
Pricing and contract exceptions
What has been granted before, to whom, and what precedent it set. Sales stops asking the same question every quarter.
#05
Incident communication
Who hears first, how much detail goes out, and the tone the lead uses with a furious customer versus a worried one.
#06
QBR and exec reviews
What the lead would raise with this account, what she would not mention yet, and the story she would tell about the quarter.
#07
New rep ramp
A new hire asks the lead’s Cortex instead of the lead. She stays the expert without being the help desk.
#08
Feature requests and roadmap asks
Which requests to escalate to product, which to decline, and how she says no without losing the account.
Beyond CS
Operations runs on the same kind of judgment.
Anywhere one senior person signs off on the exceptions, the pattern is identical: the policy is written down, the judgment is not. A lot of operations work can be solved this way.
#01
Vendor and supplier risk review
How the risk lead weighs a missing SOC 2 report against a strong track record, which gaps block onboarding, and which get a remediation deadline.
#02
Payments and reconciliation exceptions
The mismatches that need a human, the ones that clear themselves, and who to call at the bank for the rest.
#03
Compliance and policy questions
How the compliance owner reads the policy in the grey areas the document does not cover, with the past rulings attached.
#04
Procurement and approvals
What the finance lead approves without a second look, what needs three quotes, and what always comes back with questions.
#05
Disputes and chargebacks
Which disputes are worth fighting, what evidence wins, and when to refund and move on.
#06
Logistics and fulfilment escalations
Reroute, refund or reship: the operations head’s call for each failure, by customer and by margin.
Worked example · Supplier risk
A supplier review that takes a team weeks is mostly the same twenty questions, answered the way one risk head would answer them.
An agent pulls the public filings and the supplier’s questionnaire. It asks the risk head’s Cortex how she weighs each gap: a missing SOC 2 report, a thin financial history, a sub-processor in a new jurisdiction. It scores the supplier the way she would and sends her only the cases that do not look like anything she has decided before. The routine work gets done; she reviews the hard part.
One source of truth
Switch assistants. Keep the judgment.
Half the team is on Claude, someone prefers ChatGPT, engineering lives in Codex and Cursor. If the lead's judgment lives inside one assistant's memory, it is gone the day you switch. Cortex connects to all of them over MCP and an API, so every tool reads the same answers.
Your lead’s Cortex
Decisions · reasons · voice
Trust
Customer conversations are sensitive. We treat them that way.
Private by default
A Cortex answers only the people and agents its owner allows. Nothing is public unless someone chooses to make it so.
Your data is not training data
Conversations and decisions are used to answer your team, not to train models. Provider retention is switched off on every request.
Every answer shows its sources
Cortex cites the past decisions it leaned on, so a reviewer can check the reasoning instead of trusting a black box.
The owner stays in control
The lead can review what her Cortex said, correct it, and mark topics it should hand back to her.
Rollout
Live in a month. Measured from week one.
What to measure
Interruptions to the lead
Slack pings and "quick questions" per week, before and after.
Time to a decision
From a rep asking to a customer hearing an answer.
Decision consistency
The same case, answered the same way, whoever picks it up.
Ramp time
Weeks until a new rep handles escalations without help.
Questions
What teams ask first.
Is this a chatbot for our customers?
No. It is for your team and your agents. Customers keep talking to people; those people just stop waiting on one person for every judgment call.
How is this different from a knowledge base?
A knowledge base tells you the rule. Cortex tells you how the person who owns the rule applies it, including the exceptions nobody wrote down.
What if Cortex gets a call wrong?
It says how confident it is, shows the cases it is drawing on, and routes genuinely new situations back to the lead. Her corrections become part of what it knows.
We use Claude today and might switch. Does that matter?
No. Cortex connects over MCP and an API, so any assistant your team adopts reads the same judgment. The memory belongs to you, not to the tool.
Do we need to clone the whole team?
No. Start with the one or two people everyone waits on. Everyone else asks them.
Stop being the help desk for your own team.
Start with one lead and one queue. In a month, the team decides the way they would, and they get their week back.
makecortex.com/use-cases/customer-success