Main site

Advertising AI agent · Client services

Client-Services & Account-Management AI Agent

Draft client updates, answer routine account questions from the figures of record, and track actions, approvals and open items — a named account owner sends every message, and scope changes go through change control.

Typical delivery
4–6 weeks
Deployment
Your stack
Client messages
Owner sends
After launch
Agent Care

What this agent does

Writes the update — the owner sends it

In

  1. 01

    Read the account — the signed scope, the rate card, the open actions and the last thing this client was told.

  2. 02

    Take campaign figures only from the reporting source of record, stamped with the window and the time pulled.

Reason

  1. 03

    Draft the update, the answer or the recap from the account record, in the account owner's own register.

  2. 04

    Mark what the signed scope covers and what is new work, and quote no rate — pricing is a commercial decision.

  3. 05

    Check every sentence that promises, agrees, approves or concedes against what the agreement supports.

Decide

  1. 06

    Hold anything that commits the agency, moves a date or quotes a figure that has not reconciled.

  2. 07

    Route the draft to the named account owner, and anything that changes scope to whoever signs it.

Out

  1. 08

    Log actions, owners, due dates and approvals against the account, with the message each came from.

  2. 09

    Retain the source of every figure, the draft, the owner's edit and the version the client received.

Product statement

The agent drafts and tracks. Nothing reaches a client until a named account owner sends it, and no scope, price or timeline changes without the contract's change control.

Example workflow

One client request, end to end

  1. 1Request or trigger receivedA client email, the weekly update slot, a meeting recap or an action falling due
  2. 2Account context gatheredThe scope in force, the rate card, prior approvals, open actions and the last thing this client was told
  3. 3Figures pulled and reconciledCampaign numbers taken from the reporting source of record and checked against the billed view, time-stamped
  4. 4Draft written and checkedThe reply, update or recap drafted, then every commitment, date, rate and figure checked against the record

No human action required

Stages 1 to 4 run without a person in the loop — the context, the figures and the draft are finished before the account owner opens anything.

5DecisionSplits on whether the draft commits the agency
In scope, figures reconciled

Reaches the account owner ready to send.

Commits, or moves scope or a date

Held with the sentence named, not sent.

Account owner and change approver

The account owner reads the draft against the account and sends it under their own name; anything that moves scope, price or a date goes to whoever signs a change order first.

Send · Rewrite · Raise a change

Approved — handed back

  1. 6Record updatedActions, owners, due dates and approvals written to the account, alongside the version the client actually received
  2. 7Outcome evaluatedFigure accuracy against source, commitments the reviewer flagged, approval-trail completeness and escalations

Owner edits

Every edit an account owner makes to a draft is counted.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Sending any message to a client.
    • Agreeing a scope change, a price or a rate.
    • Committing to a date, a deliverable or a make-good.
    • Recording an approval the client did not give.

  • Automation boundary

    Agent acts unaided

    • ✓Draft updates, replies and recaps from the account record and the scope in force.
    • ✓Pull campaign figures from the reporting source of record.
    • ✓Track actions, owners, due dates and outstanding approvals across accounts.
    • ✓Flag the sentence that commits the agency before anyone can send it.

    Write actions run only inside the approval boundaries agreed during implementation. Sending to a client and agreeing a scope change are not among them.

    • Quoting a performance figure that has not reconciled.
    • Answering from another client's account or spend.
    • Escalating, apologising or conceding fault on an account.
  • Changing the scope record or the approval thresholds.

Example output

One client update, annotated

Everything the agent drafts is attached to the account record and the figures it was built from.

Client-update output · single draft

Illustrative example
Request
Weekly client update
Account
Retainer, scope in force
Figures pulled
Today, 09:00
Draft
Held before send
Confidence
92%
Commitments
One found, flagged

As received

The request, the scope in force and the time the figures were pulled — nothing on this side is inferred.

Evidence used

  • Signed scope of work
  • Platform report, dated
  • Last approval on file

Why it is held

One line offers a creative round the scope does not cover. That is a change order.

What the score decides

How closely the account owner re-reads before sending — not whether the client will accept it.

Action

SendRewriteRaise a change

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every client request

From email, calls and briefs

03Account record

Answer from the account's record

Use the scope in force, the approval trail and the reporting source of record — not what the thread remembers.

01Approved path

Take the writing off the account team

Updates, recaps and routine answers are drafted before the account owner opens the account, so they edit and send rather than assemble.

02Human review

Catch the commitment in the draft

A sentence that agrees scope, a date or a rate is named before it is sent, instead of surfacing when the client asks the agency to honour it.

04

Build an evidence trail

Retain the source of every figure, the scope in force, the draft, the account owner's edit and the version the client received — on both paths.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Email & messaging

    • Outlook
    • Gmail
    • Slack
    • Teams

  • CRM & project management

    • Salesforce
    • HubSpot
    • Asana
    • Workfront
    • Monday

  • Campaign reporting

    • Google Ads
    • Meta
    • DV360
    • Looker Studio

Agent

Client services & account tracking

  • Drafts updates
  • Reconciles figures
  • Tracks approvals

  • Scope, time & documents

    • Harvest
    • Scoro
    • SharePoint
    • DAM
    • Meeting transcripts

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and your client

Each control wraps the one inside it. A draft clears every layer before an account owner reads it, and the send sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn drafting to the account team if evaluations or signals degrade.Roll back
L5TraceabilityRecord the sources read, the draft, the owner's edit and what was sent.Record
L4Client separationRetrieval is scoped to one account; no other client's data is reachable.Isolate
L3Approval gateThe owner sends; scope, price and date changes go to change control first.Gate
L2Commitment checkAny sentence that promises, agrees, approves or concedes is flagged.Flag
L1Source groundingEvery figure traces to the reporting source of record and its pull time.Ground
Model coreDraft proposed — the update, answer or recap, the figures in it and the commitments it makes
L1 – L2
Decide whether the draft may stand
L3
Decides who sends and who signs a change
L4 – L5
Keep accounts apart and the trail intact
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the message that was sent — not only the draft.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the update and the answer the client reads

  1. E1Final-output evaluationDoes every figure in the draft match the source of record?
  2. E2Step-level evaluationDid it read the scope in force and the approval trail before drafting?
  3. E3Tool evaluationDid it read the right account, and only that account?
  4. E4Commitment and tone reviewWhat did a reviewer flag as a promise, a concession or a misjudged escalation?
  5. E5Slice evaluationHow does performance change across account types, clients and request types?
  6. E6Business outcomeUnbilled out-of-scope work, corrections reissued and disputes the drafts caused.

Floor — the commitment the agency can actually meet

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    2 modes

    • CS-03

      Stale or unreconciled figure

      A number is quoted from a superseded pull.

    • CS-04

      Cross-account leakage

      Another client's spend reaches the wrong thread.

    Stage gathers

    Scope, approvals, history and the reporting figures

  2. 02 · Drafting

    2 modes

    • CS-01

      Unauthorised commitment

      The draft agrees work the scope does not cover.

    • CS-06

      Recap rewrites the meeting

      The note records a decision nobody made.

    Stage proposes

    The update, the answer or the recap, and its claims

  3. 03 · Review / send

    1 mode

    • CS-02

      Change order never raised

      Extra work is agreed and no change follows.

    Stage sends

    Only under the account owner who approved the draft

  4. 04 · Record

    1 mode

    • CS-05

      Approval never given

      A polite reply is logged as sign-off.

    Stage logs

    Actions, owners, due dates and approvals on file

  5. 05 · Change / Version

    1 mode

    • CS-07

      Superseded scope source

      Drafting runs against an amended scope of work.

    Stage tracks

    Model, prompt, template and scope-source changes

  • Sev-1 · the agency is bound by what was sent
  • Sev-2 · the client is told something unsupported
  • Sev-3 · the draft is wrong at source
See the complete Advertising failure-mode catalogue

Affected slices

The account in trouble is not the average account

Aggregate draft quality can look acceptable while a few account cohorts carry most of the flagged commitments, most of the figure corrections and nearly all of the scope disputes. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Unhappy or at-risk accounts5.1%3.0×Review
Scope and change questions3.7%2.2×Review
Same-category competing clients2.6%1.5×Watch
Routine in-scope status updates1.2%0.7×Normal

Bar: failure-rate lift vs. routine status-update baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The bill for a sentence arrives weeks after it is sent

A commitment only looks wrong when the client acts on it, weeks later. The cycle must reach the message that caused it, not last week's drafts.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    A dispute, an unbilled hour or a reissued figure traces back to a sent draft.

  2. 02

    Diagnose

    Isolated to the source read, the scope in force, the commitment check or the send.

  3. 03

    Improve

    The rule, source or template is changed, re-approved by the account lead and version-linked.

  4. 04

    Verify

    Re-run against held-out drafts from that account, including the one that caused the dispute.

  5. 05

    Learn

    The disputed sentence becomes a regression case and a line in the account playbook.

  6. Learn → Detect

    The return edge. Every cycle also re-reads the scope and rate card in force — last quarter's correct answer is this quarter's out-of-scope work.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, account and scope sourcing, drafting and pre-send checks, evaluation, review workflow, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Account, scope and rate-card sourcing.Week 1, Week 2
  3. 03Email, CRM and project-system access.Week 1, Week 2
  4. 04Reporting source-of-record mapping.Week 2
  5. 05Client-separation and scoping controls.Week 2
  6. 06Draft generation in the owner's register.Week 3
  7. 07Commitment and scope checks before send.Week 3, Week 4
  8. 08Action, approval and change-order tracking.Week 4
  9. 09Account-owner review and send workflow.Week 4, Week 5
  10. 10Evaluation suite, leakage probes and regression cases.Week 4, Week 5
  11. 11Meeting transcription and recap controls.Week 5, Week 6
  12. 12Observability, deployment and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne account teamProductionProduction CRM integrationAdvancedMulti-office / multi-client
Introduced at Pilot
Drafted updates and replies✓✓✓
Figures from the source of record✓✓✓
Account-owner send-off✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Commitment and scope flagging—✓✓
Action and approval tracking—✓✓
Client-separation controls—✓✓
CRM and reporting integration—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Meeting recaps and minutes——✓
Multi-office and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on accounts and systems in scope, reporting sources, client-separation requirements, message volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01The signed scope and rate card for each account in scope

    Account, scope and rate-card sourcing

    Week 1

  2. 02Who owns each account and who signs a change order

    Automation-boundary definition and the send-off workflow

    Week 1

  3. 03Access to email, CRM and project systems

    Email, CRM and project-system integration

    Week 2

  4. 04The reporting source you treat as authoritative

    Reporting source-of-record and reconciliation rules

    Week 2

  5. 05Your conflict rules for competing accounts

    Client-separation and retrieval-scoping controls

    Week 2

  6. 06Real threads where a commitment or a scope dispute happened

    Evaluation suite, leakage probes and regression cases

    Week 4

  7. 07Named account owners and change approvers

    Account-owner review, send workflow and pilot accounts

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Phases are drawn over the weeks they actually occupy. Week 5 carries both the evaluation work and the first drafts an account owner sends.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Accounts in scope, who owns each and who signs a change

W2Scope and rate-card sourcing, system access and client separation

W3Draft generation, commitment checks and the pre-send hold

W4Action and approval tracking, evaluation suite and leakage probes

W5Owner review workflow, first drafts sent and targeted corrections

W6Production validation, recap controls and Agent Care handover

Reading the band
Client separation is built in week 2, before a single account is drafted against in week 3. The bars show that dependency, not a smooth ramp.
At the end of W6
Drafts have gone out under named account owners, and every flagged commitment was cleared or removed first, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Advertising AI agent

Build a client-services agent around your account teams.

Show us one account — the signed scope, where its numbers come from, and who is allowed to send that client anything. We'll take that account's weekly update and one live scope question end to end, and show you what the agent would have held back.

Typical delivery
4–6 weeks
Deployment
Your agency stack
Client messages
Account owner sends
After launch
Agent Care available

Nestack Agents · Client services & account management

AGT-AM-05 · Agent Care available after launch

Build a client-services agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your setup

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.