Use casesServices

Agent assist

New call-center agents take months to be productive and put customers on hold to search the knowledge base.

Kinds of AI
SLMsAgentic AI
Who decides
Call-center agent
FIG. 1 · Side panelIllustrative
Side panel · live callCustomer asks to skip an installment

SLM · the procedure

  • 4.2.1Eligible if ≤2 skips per yearprocedure paragraph

Agent · this customer

  • met1 skip this year

SLM · say

You can skip this month's installment; it moves to the end of the term.

  • SLM
  • Agent

DecidesCall-center agent

1. The desk

Today

  1. FindCall-center agents search the knowledge base while the customer waits; procedures change, and new agents learn them by escalating.
  2. SummarizeAfter-call summaries and disposition codes are typed by hand and vary by call-center agent.
  3. FileEvery field of the CRM entry is filled in by hand after the call.
  4. DecideThe call-center agent decides what to tell the customer, often after putting them on hold.

With the system

  1. FindThe speech model transcribes the live call, and the SLM suggests the next step with the exact procedure paragraph, found by retrieval, within one to two seconds.SLM
  2. SummarizeAt the end of the call, the same SLM writes a three-line summary, a disposition code, and a follow-up task.SLM
  3. FileThe agentic assistant fills the CRM fields as drafts; it never finalizes a record.Agentic AI
  4. DecideThe call-center agent decides everything said to the customer, corrects the drafts, and finalizes the CRM entry, as before.Person decides

What lands on the desk

While the customer is still talking, the call-center agent sees the procedure paragraph that applies, whether this customer qualifies, and a sentence to say. After the call, the summary and code are waiting as drafts for the call-center agent to correct and finalize.

FIG. 2 · Side panelIllustrative
Side panel · live callCustomer asks to skip an installment

SLM · the procedure

  • 4.2.1Eligible if ≤2 skips per yearprocedure paragraph

Agent · this customer

  • met1 skip this year

SLM · say

You can skip this month's installment; it moves to the end of the term.

SLM · after the call, draft

  • SummaryCustomer requested a skip; eligible; confirmed by the call-center agent
  • DispositionINST-SKIP-APPROVED
  • Follow-upSend confirmation
  • SLM
  • Agent

DecidesCall-center agent

2. The system

For your technical team — skip it if you like.

How it works

In

  • Live call audio

  1. Speech model · transcribes

    Streaming speech recognition

    The contact-center speech model, run on the live call

    Reads

    • Everything under In

    Out

    • A live transcript

  2. SLM · understands

    Procedure assistant

    Fine-tuned on the institution's procedures, read by retrieval

    Reads

    • The transcript
    • The procedures, by retrieval
    • The customer record

    Out

    • A next-step suggestion
    • The exact procedure paragraph
    • Within one to two seconds

  3. SLM · understands

    Call summarizer

    The same SLM: a summary task

    Reads

    • The call, at its end

    Out

    • A three-line summary
    • A disposition code
    • A follow-up task

  4. Agent · prepares

    CRM-drafting assistant

    Fixed workflow, light

    Reads

    • The summary, the code, and the task

    Out

    • CRM fields filled as drafts

  5. Decision

    Call-center agent

    Reads

    • The suggestions
    • The CRM drafts

    Out

    • Decides everything said to the customer
    • Finalizes the CRM entry
How the agent's work is arrangedFixed workflow

Fixed workflow

There is one step after each call (write the drafts to the CRM), so a light fixed workflow is enough.

How the models are trained and tested

Your expert writes the test first. A step the test does not need is skipped, and we say why.

Show the training steps

Small language model (SLM)

  1. Test set

    200–400 call moments graded by senior call-center agents (right procedure, right paragraph, summary accuracy).

  2. Dataset

    Procedures, and past summaries and dispositions.

  3. Continued pre-training (CPT)Skipped

    Skipped: retrieval supplies the procedures' wording.

  4. Supervised fine-tuning (SFT)

    On the suggestion format and the summary and disposition codes.

  5. Reinforcement learning (RL)Skipped

    Skipped: fine-tuning holds the format.

  6. Retrieval

    Retrieval supplies the procedures, so a procedure change needs no retraining.

Where the agentic system stops

The call-center agent decides everything said to the customer and finalizes the record; the supervisor owns procedure changes.

Show its tools, limits, and record
Tools and access
Write draft fields in the CRM only.
May not
Speak to the customer, finalize a record, or take an account action.
Who decides and the written limit
Call-center agentThe call-center agent decides everything said to the customer and finalizes the record; the supervisor owns procedure changes. The assistant drafts only.
Stops when
No procedure paragraph is found: the suggestion is withheld.
Record
The suggestions shown, the paragraphs cited, and the call-center agent's edits.

How agentic systems stay inside limits

3. Value and conditions

Where the time goes, and what your institution will measure.

Where the value comes from

  1. Handling time.
  2. Time to proficiency for new call-center agents.
  3. Consistent summaries.

What your institution measures

  • Average handling time
  • After-call work time
  • Time to proficiency
  • First-contact resolution
  • Summary edit rate
  • Customer satisfaction
Published elsewhere
In a study of 5,179 customer-support agents, access to a generative-AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice and low-skilled workers.

Brynjolfsson, Li, and Raymond, "Generative AI at Work", NBER Working Paper 31161

Measured in the Decision Blueprint

Baseline handling and after-call time set the starting point; your institution sets the target.

Conditions

The rules, the data, and the known error, with its guard.

  • The KVKK points of the contact-center speech case apply here too.
  • A suggestion without a cited paragraph is not shown.
  • The latency budget limits model size; hardware is sized in the Decision Blueprint.

Start here

We take a sample of your calls and procedures, measure today's handling and after-call time, and test the suggestions on call moments your senior call-center agents grade before anything is built.

Request a Decision BlueprintWhat the Decision Blueprint covers
  • SLMsAgentic AI

    Contact-center speech: quality and compliance on every call

    TodayEvery call is recorded, quality hears two in a hundred, and off-the-shelf speech-to-text mangles your product names.

    With the systemEvery call checked for quality and compliance; the coaching point in context.

    Team lead decides

  • SLMsAgentic AI

    Outsourced complaints desk

    TodayOne team answers written complaints for several clients, each with its own rules, tone, and deadlines.

    With the systemEach reply drafted in the client's own rules and tone.

    Team lead decides

Equalize the gap.

Bring one decision. Leave knowing the first step.

A short description of the decision is enough; no confidential records.

Start with a Decision BlueprintTalk to the engineer

https://eqlx.ai/use-cases/agent-assist · Printed copy; the website is the current version.