OneBenefits: marketing AI in eight weeks.
The same architecture pattern we deploy for clients, shipped end-to-end on ourselves. The proof point we cite when buyers ask if any of this is real.
Every buyer eventually asks the question. "Have you actually shipped this end-to-end somewhere?" The honest answer most agencies give is "almost." Customer Zero is the answer that earned a different reaction.
The before state
OneBenefits had a marketing operation that looked, on paper, like every B2B SaaS marketing team in 2024. A CDP, a marketing automation platform, a CRM, a content library, an event stream, a small data team. The components worked. The handoffs between them did not.
Campaigns took six to eight weeks from brief to launch. Most of that time was orchestration. The marketer wrote the brief. The data team translated it into a segment. The creative team generated the assets. The MarOps team scheduled the cadence. The data team checked the segment again because something had shifted upstream. The creative team revised. By the time the campaign hit the inbox, the prospect's intent had moved on.
The signals were there. The visit-level data was clean. The segment library was robust. What was missing was the connective tissue that turns a signal into a same-week campaign without four handoffs and three revisions.
The work, in eight weeks
We built three components on top of the existing stack. None of them required ripping anything out. All of them extended what was already running.
Persona Segmentation as a Customer Intelligence Graph
The existing segment library was rebuilt as a real-time graph on a governed cloud data lake. Visit signals, CRM activity, intent feeds, and benefits-product context all join in one entity layer. Every persona segment becomes a query against the graph instead of a batch-rebuilt list. Schema drift is monitored. Consent state is enforced at query time, not at activation time. The data team gets a versioned API for the rest of the marketing org to consume.
Campaign Manager as a Brief-to-Assets Agent
The campaign brief becomes a structured input to a governed agent. The agent reads the brief, queries the persona graph for the relevant segment, generates draft creative grounded in the existing content classifier, and surfaces a reviewable bundle to the marketer inside their existing workspace. The marketer approves, edits, or rejects. The agent never publishes without sign-off. Time from brief to first reviewable bundle dropped from days to under an hour.
Unified Scheduler as Next-Best Action coordination
The scheduler decides which channel each segment hits and when, against the live state of every other channel. Email cadence rules talk to the chat front door. The web personalization layer reads from the same graph as the sales handoff. A prospect who engages on the website does not get the same activation email tomorrow that they would have gotten yesterday. Same audience, different orchestration.
The architecture, simplified
Customer Zero · Reference Architecture
Cross-cutting: MLflow, OpenTelemetry, OneTrust consent, RBAC, RLS, audit trail. Governance was wired in from day one.
What we called it, where it lives
| What we called it | Where it lives in the stack |
|---|---|
| Persona Segmentation | Customer Intelligence Graph on a governed cloud data lake. |
| Campaign Manager | Brief-to-Assets Agent inside the existing marketing workspace. |
| Unified Scheduler | Next-Best Action coordination across web, email, chat, and sales. |
Where the lift came from
The +40% engagement lift number gets a lot of attention. The harder question is where it actually came from, given that the audience and the content were the same.
About a third of the lift came from timing: the same email landing in the same inbox on the day the recipient was researching, instead of three days later. About a third came from sequencing: the channel chosen for each touch reflecting what the recipient had just done in another channel. The remaining third came from content selection: the agent picking from a wider content library faster than a human marketer could scan it for the best fit.
None of those lifts required new content investment. None required new audience acquisition. The graph and the agent layer extracted value from inputs the marketing team already produced.
The −75% time-to-launch number is the more durable one. A six-to-eight-week campaign cycle dropped to roughly one-and-a-half. The marketing team did not need approval for new headcount. The savings funded the work the team actually wanted to do.
The agent layer changed how our team works in a way the platform vendors had been promising for two years and never delivered. The difference was that Innovative Group built it on top of what we already had, instead of selling us another stack.
What this proves out for clients
Customer Zero is the work we cite when a CMO or a CTO asks whether any of this can ship inside a stack they already run. The same pattern applies. Different industry, different MAP, different CDP. The four-layer architecture is the same. The four tensions are the same. The eight-week zero-to-live target is the same.
The pattern is not a slide deck. It runs every day, on our own marketing org, against our own audience.
Right offer, right place, right time.
Our full Next-Best Action solution page covers the four-layer architecture in depth, the four tensions slowing most enterprise programs, and the three doors into the work.
See the full Next-Best Action page →