Tony Coconate

For Dale and the executive team

The first
ninety days.

How I would lead EMRge, the AI OS, and the engineering organization — and what I will put on the table at each mark.

  • People first
  • Metric driven
  • Scale what works
  • Stay in the work

Day 30 — mapping the loop

Listen, and make the system legible. I arrive ready to draw what exists, what it costs, and where the time goes.

  • The engineering senior director, all six product teams, and client delivery. One client, shadowed, in the first month.
  • EMRge on a single page. Databricks, platform services, what clients actually touch, the infrastructure, and where SOC 2 is a habit.
  • The AI OS on a single page. Claude, Glean, ClickUp, Slack, Outlook. What is connected, what writes back, and what is still a demonstration.
  • A cost baseline someone owns. Cloud, Databricks, and model inference.
  • The hiring bar, drafted before a req opens into fog. A weekly half hour with Dale: decisions and risks.
Hold me to

A one-page map of both systems, a cost baseline, and the sequenced work for the next sixty days.

Put a foundation under it, and close one loop. Standards that are enforced. One workflow whose metric was written down before the code.

  • Golden paths for build, release, infrastructure as code, and observability. Repeatable, auditable, and quiet in the best way.
  • Governance for the AI OS, written with IT and Security. Access, credentials, audit logs, least privilege. In place before the workflows spread.
  • LLMOps, small and real. An evaluation, a cost per workflow, a trace. We scale what we can see.
  • A build-versus-buy pass on models and vendors. Speed, cost, and the right to leave.
  • One pilot with a startup’s blast radius. A meeting that becomes an action written back to ClickUp or Slack, or a measurement insight written back into planning. Name the blockers early. Name who is already succeeding, and protect that.
Hold me to

One workflow in production, scored against a metric we agreed up front. A governance note IT and Security have signed.

Scale what worked. Name the year. If the pilot moved the number, productize it. If it did not, stop it where the organization can see. Either way, the next year is on paper with Product.

  • Take the golden path to a second team, or end the pilot and keep the lesson. Do it in the open.
  • Write the AI OS down. Orchestration, write-back in both directions, knowledge. An architecture that outlives any one person.
  • Name the EMRge bets, and the work we will leave alone. A sharp core, then the thin layer a client actually uses.
  • A hiring plan tied to the year, and career paths a manager can say out loud.
  • One conversation outside the building, with a client or a prospect, so the technology has a voice in how the firm wins and keeps work.
Hold me to

A one-year technology roadmap, co-owned with Product. Cost, risk, and the numbers that mean it worked.

Shaped by time with Troy, Carrie, Alex Looker, Dianne, Beth, Ericka, Bonnie, and others across the firm. Erin and Kyle already let me sit down and build. I plan to keep showing up that way.

Measurement

Count what grows the business.

Traditional metrics optimize for what’s easy to count. EMR optimizes for what actually grows the business — incremental revenue, healthier customer files, and profitability.

The industry is moving from single-source attribution to triangulation. Ovative is already there, with a unified framework and the platform to operationalize it.

EMR Brand Revenue Customer

  1. Enterprise revenue

    Online and off. Store and wholesale included. If that dollar is missing, the picture is wrong.

  2. Incrementality

    The causal part. Sales and members who would not have arrived without the marketing.

  3. Future customer value

    New and reactivated. The long health of the file, beyond this week’s order.

  4. Profitability

    Margin. The number finance can stand behind.

ROAS and last click

Sees the online order, hands the credit to the final touch, and misses the store, the brand, and everything privacy changes knocked out. Cookie loss. iOS ATT.

Platform-reported metrics

Google, Meta, and the rest each claim the same conversion. Added together, the reports routinely outrun the growth of the business.

Basic multi-touch

A step past last click, and still a correlation. Blind offline. Signal loss has taken an estimated 30–40% of conversions that used to be trackable.

A note for the measurement practice. I will not out-model the people who do this every day. I do want the arguments to be inspectable. Bayesian versus frequency. Saturation and diminishing returns. Fallout and creative fatigue. When a recommendation moves because one of those moves, a client should be able to see why.

The two systems

One platform that measures. One layer that acts.

EMRge

The asset. A data foundation, measurement, and the surfaces our teams and our clients use to plan and grow. Databricks underneath. I read the public product as six capabilities already in market: predictive planning, holistic reporting, modern MMM+, automated operations, EMR activation, and precision testing. The work is to make these scale, stay enterprise-grade, and stay pointed at incremental revenue, a healthier customer file, and profit.

AI OS

The loop. An orchestration layer across Claude, Glean, ClickUp, Slack, Outlook, and the next tool that earns a place. A meeting becomes intelligence. Intelligence becomes an action. The action writes back into the system where the work already lives. Knowledge is still findable the week after. IT and Security are in the first design, because this reaches across the company. Every workflow gets an owner, an evaluation, and a cost.

Buy where the market is already excellent. Build where the workflow is Ovative’s: the measurement logic, the write-back into how this firm actually works, the golden path our own teams will live on. Models are a portfolio. Keep the right to leave.

How I run it

A high bar, and a short path to the customer.

  1. 01

    Accelerate the people here

    Ovative’s advantage is the talent already in the building. Technology should multiply that. Clear careers, a bar we can describe at every level, and managers who know what great looks like.

  2. 02

    Metric before build

    Write down what success is. Test. If it moves the number, scale it quickly. If it does not, stop it where everyone can see, and keep the lesson.

  3. 03

    Engineers near customers

    Keep the people who build close enough to hear what is actually needed. Give them room to step outside a lane when the client problem asks for it. That proximity is how a firm this size stays able to pivot.

  4. 04

    Build the 5%

    Get the core right, once. Then build the thin layer each customer actually uses, and make that layer theirs. In the age of AI, that is what finished looks like. Specific, on top of something solid.

  5. 05

    Golden paths

    Care in delivery comes from a standard way to start and a standard shape of output. A path people trust, and permission to leave it when they should.

  6. 06

    Hands on

    I can sit with a stakeholder who does not write code, and I can still read the diff. Name the blocker early. Name the success early, and protect it. AI carries an idea further. Judgment stays with the people.

The work, briefly

Twenty years, still in the code.

The pattern is the one I would bring here. Get close to the user. Build the smallest thing that teaches you something. Scale it when the metric moves.

2015 — now

Google

Tech lead, manager, and staff engineer. Marketing platforms.

  • SACA. Machine learning over search-term performance, to see what was resonating. It led to YACA, the same idea pointed at YouTube creatives. Themes, not just terms.
  • Video creative performance. A patent-pending model for how a video will land.
  • SDF, from Bulkdozer. A sheet, into DV360.
  • Data lake designer, for Publicis. Ads, maps, and social, through Fivetran, into one channel view.
  • Anomaly detection. ARIMA, seasonal decomposition, an alert when the pattern breaks.
  • In the open. The Agent Development Kit, and MCP servers for Google Ads.

2012 — 2015

DevMynd

Director. We grew the consultancy from 8 people to 32.

  • Local dealers. Automotive bidding, built for the way a storefront actually buys media.
  • Farmland. A chemical-buying tool drawn on geo maps. The user was never an engineer. Adoption was the product.

2011 — 2012

Lightbank and oBaz

CTO-in-residence. The idea that held was oBaz.

  • Group buying, gamified. Deals like Sperry and Ray-Ban. From the idea through to an acquisition into Groupon Goods.

2006 — 2011

Tribune

Steve Gable got us the MacBooks. I started the Tribune Interactive group.

  • newspaper.com. A CMS for the sites.
  • The archive. Watermark a photo, then license it on to iStock.

Past ninety

The year is the point.

My job as CTO is to make that measurement system more scalable, more AI-augmented, and more tightly closed-loop into planning and activation — while keeping the human expertise that turns numbers into decisions.

  1. Get the core right Data lake, insights, planning, and protection. The foundation has to be boring before the ambition gets loud.
  2. Close the loop Measurement that feeds planning and activation, with a person still making the call. One new surface I would pilot, and not promise: read organic video the way SACA taught us to read search. Themes that resonate, not only media we paid to run.
  3. Run AI like a product Governed, evaluated, costed, and small enough to change. The AI OS earns a second team only after the first one can show its work.
  4. Grow without the machinery Doubling the business in three years works if engineers can still hear a customer. Hiring follows the year. The bar stays high. The path between the build and the client stays short.