case-study 001 --active-engagement
An embedded agent fleet for a BI and performance marketing agency.
A connected system that brings scattered company knowledge, recurring work and cross-department execution together.
company profile
A 100+ person agency spanning the full campaign lifecycle.
scale
100+ people
Senior operators working globally across the agency.teams connected
6 core functions
Strategy, performance, creative, data, project management and client operations.business model
Services + SaaS
The agency also owns an analytics product with bespoke customer dashboards.engagement
6 weeks in
A 3–4 week operational audit followed by heavy implementation.the constraint
Expertise was distributed across people, tools and client work.
The agency operates across creative, performance, project management, data and client operations. Important context lives in meetings, messages, task systems, documents, design files, dashboards and the heads of individual specialists.
A generic assistant could answer isolated questions. It could not safely maintain the operating context, understand which source was authoritative, execute recurring work and preserve the boundaries between teams, markets and clients.
The requirement was not “install a chatbot.” It was to build an operational system the agency could trust and expand.
It also did not require a wholesale migration. The fleet plugs into the collaboration, project, creative, meeting and data tools already used by the business.
the audit · weeks 1–4
We spent the first month understanding the operation before setting the build order.
The forward-deployed audit followed real work across departments: where requests began, which tools and people they touched, what had to be rebuilt at each handoff, and where expert judgement genuinely mattered.
Company knowledge was scattered
Decisions and customer context lived across meetings, Slack, project tools, design files, dashboards and individual specialists.
Work crossed too many inboxes
Campaign, creative, data and project teams repeatedly translated the same request as it moved from one department to the next.
Specialists became queues
Campaign managers waited for analysts to answer repeatable questions because the knowledge required to answer safely was not available at the point of work.
Approval did not mean delivery
Approved creative and dashboard requests still triggered manual preparation, naming, checking, upload and coordination across several people.
What the audit changed
The problem was not a shortage of AI tools. It was the amount of knowledge and work repeatedly transferred between departments.
the architecture
One path from company knowledge to approved delivery.
01 / existing tools
Work stays where it is
Slack, project tools, meetings, design files and data systems remain in place. No wholesale migration is required.02 / shared knowledge
The operation becomes readable
A dedicated worker keeps approved company and customer knowledge current for the agent.03 / team agent
Slack becomes the front door
People ask questions, develop work and approve delivery with the embedded agent in the tool they already use.04 / department rules
Experts define what good means
Data, creative and campaign specialists maintain the definitions, methods and quality standards the agent follows.05 / connected execution
The agent carries the work
It moves one request across departments, prepares outputs and completes repeatable steps without losing the original context.06 / review + delivery
People keep final control
Important work is checked and approved by a specialist before it reaches a campaign, executive or customer.The system connects to the tools the agency already uses. The architecture separates knowledge, conversation, execution and approval so each part can expand without giving one agent unrestricted access to everything.
implementation · 6 weeks so far
Four weeks to understand the operation. Two weeks into heavy execution.
Week 1
Embed and listen
Joined the operation, mapped the major teams and tools, and established access to the sources needed to understand how work really moved.Weeks 2–3
Follow the handoffs
Traced recurring work across creative, performance, data, project management and client operations, identifying where context was lost or repeatedly rebuilt.Week 4
Turn the audit into a build plan
Agreed the agent architecture, knowledge boundaries, first high-value workflows and the order in which they should be implemented.Weeks 5–6
Build the live foundation
Put the embedded agent, shared company knowledge, recurring updates, reporting and first connected workflows into active use.Now
Execute against the audit
Expanding from assistance into cross-department delivery while instrumenting waiting time, manual touches, rework, throughput and cost.working now
The foundation is live and carrying real work.
Builds shared company knowledge
Meetings, presentations, messages, project work, design files and customer feedback are brought together so the agent understands what each team is working from.
Works where the team works
The agent works in Slack, where it can prepare briefs, find information, support project work and share finished materials without forcing people into another application.
Runs repeatable work
Scheduled workflows monitor creative work across two markets, show what is moving or stuck, and keep people responsible for important decisions.
Produces reports that show their sources
Weekly learnings and executive reports include the source material, confidence and missing information so incomplete evidence is never quietly presented as fact.
OPERATIONS Creative status monitoring across two markets with clear updates and no repeated noise.
KNOWLEDGE Historical meetings and live company information are kept current for the agent.
REPORTING Source-linked learning and executive reports that clearly show confidence and missing information.
RELIABILITY Daily checks and a clear success or failure report for every company-knowledge update.
cross-department execution
One agent chain replaces a pile of manual handoffs.
The foundation is live. The connected execution steps below are labelled by their current state so planned work is not passed off as completed.
01 / contextlive foundation
Bring the operation into shared context
The agent reads approved meetings, conversations, project systems, creative feedback and analytical context so departments are no longer working from separate fragments of the truth.02 / answersexpanding
Give campaign teams direct answers
Campaign managers ask the embedded agent instead of joining an analyst queue for every repeated data question. Analysts maintain the definitions and rules the agent uses to answer correctly.03 / feedbackactive development
Turn client feedback into draft creative changes
The agent interprets shared feedback and prepares adjustments before human QA and finishing, reducing the amount of manual translation between account, campaign and creative teams.04 / launchactive development
Move approved assets into campaign execution
Approval can trigger preparation, QA, naming, export and paid-media upload as one controlled chain, with humans reviewing the work at the consequential checkpoint rather than moving every file by hand.05 / learningroadmap
Carry performance learning into the next campaign
Creative-performance evidence becomes the starting point for future work, helping the agent produce a stronger first pass while specialists tune the strategy, judgement and quality bar.06 / reportingactive development
Combine company knowledge and data for leaders
Analysts define what the numbers mean. The agent combines that with meetings, conversations and work history to produce detailed reports and useful answers for performance leads.07 / deliverydelivered
Turn a Slack thread into a customer dashboard
The team develops the idea with the agent in Slack. The agent performs the analysis and builds the dashboard, a specialist reviews it, and one final approval delivers it to the customer.customer dashboard delivery
From a Slack conversation to a customer-ready dashboard.
The agency has its own analytics product. Its customers receive bespoke dashboards built around their business, campaigns, goals and data.
Delivering a new dashboard used to be a large piece of work spread across several teams. Analysts interpreted the request, specialists assembled the context, developers built it, and multiple rounds of review moved between departments.
Now the dashboard is developed collaboratively with the embedded agent in a Slack thread. The agent performs the analysis and ships the code. A specialist checks the work, and one final approval delivers the finished dashboard to the customer’s account.
A customer asks a question
A customer or campaign lead explains what they want to understand and the decision the dashboard should help them make.
The team develops the idea in Slack
Data, campaign and customer experts shape the analysis together in one thread with the embedded agent.
The agent gathers what it needs
It draws from the customer history, meetings, conversations, existing dashboards and the data rules maintained by the analytics teams.
Write the dashboard
The agent produces analysis and dashboard code against the maintained templates, components and delivery contract.
A specialist checks the work
A human reviews the analysis, presentation and customer-facing result, then requests any final adjustments in the same conversation.
One approval delivers it
With a final approval prompt, the dashboard appears inside the agency’s analytics product for the customer to use.
delivered outcome
Work that once took days or weeks can move from customer request to finished dashboard through one Slack conversation.
Data specialists maintain the definitions, analysis methods and dashboard standards. The agent applies that knowledge to the work, while the specialists focus on improving the approach and checking the result instead of manually carrying every dashboard through the process.
results so far
Six weeks in, the operation is already moving differently.
cross-department communication
Fewer departments acting as each other’s inbox
Campaign, creative, performance and operations teams can work through one embedded agent with shared company knowledge instead of repeatedly forwarding requests between specialists.customer delivery
Days or weeks of dashboard work through one Slack conversation
The agent carries the request from collaborative analysis through code, specialist QA, final approval and delivery into the customer’s analytics account.specialist leverage
Experts tune the machine instead of carrying it
Analysts define the data. Creative experts set the quality bar. Campaign managers make important decisions. The agent performs more of the transfer and preparation work between them.Measurement now being added
Analyst-response waiting time Approval-to-launch time Manual touches per campaign Rework Throughput Cost per campaignwhat this demonstrates
Forward-deployed engineering compounds.
The first value comes from solving recurring work. The larger value comes from instrumenting that work and creating a durable picture of how the operation behaves.
As the system sees more decisions, exceptions, handovers and outcomes, it becomes possible to improve the automation, redesign the workflow and identify changes to the business itself.
Discuss an operational constraint