Anne K. Nies | VP AI, Machine Learning & Data Strategy

FLAGSHIP · COHERE HEALTH · 2021 to 2023

Forty models. Fifteen months. One operating pipeline.

Prior-authorization automation at Cohere Health. Unstructured clinical documents in, evidence-graded decisions out. The visualization below maps the actual production system the ML team shipped: NLP and document-AI models extracting clinical facts, policy-matching models comparing to health-plan rules, then a branching decision that auto-approves or annotates for the medical reviewer.

DATA & MODELS

BUSINESS OUTCOMES

HUMAN-IN-THE-LOOP

40+

MODELS DEPLOYED

All in production simultaneously by month 24, building on top of each other across the document-AI and policy-matching layers.

$4M+

BUSINESS VALUE / 15 MO

Measured in faster auto-approval, reduced manual review time, and fewer round-trips with the provider on missing information.

65%+

INCOMING FAXES AUTOMATED

Intelligent fax-processing pipeline on AWS with OCR and routing. Took the highest-volume manual workflow and made it the smallest.

75%

FASTER TIME TO DEPLOY

Standardized the ML deployment pipeline, established engineering principles, and grew the team from 6 to 20 along the way.

ABOUT

The problem I solve, and how I got here.

For 15+ years, the work has been the same problem in different clothing: take a business process that runs on human judgment under load, and automate the parts a model can do better. The path I took to get good at it is unusual. I started in operations.

At Vanguard Logistics Services, I ran North American Import Operations: 135 people, six offices, the kind of high-volume manual work that breaks under its own paperwork. To make my team faster, I taught myself to build software. The auto-rating system I shipped saved the company seven figures and impacted 3,000+ staff. That was my first AI project, before anyone was calling it AI.

The next decade was the formal version of the same instinct. Senior Data Strategist at Donlen (Hertz), where I identified $27.7M+ in client cost-savings across a portfolio of fleet operations. Senior Data Scientist at Allstate, leading Mobility Graph, the telematics graph-theory project, on a billion rows of location data. A patent came out of that work. Manager, Data Science at Rally Health (Optum), leading the team that built the readmission-risk model providers actually adopted to flag patients needing follow-up before things got serious.

Then Cohere Health for two years as Director of Machine Learning, where the operating doctrine on this site got proven out: 40+ models in production, $4M+ in business value, team scaled 6 to 20. Then Submittable as Director of Science, building the AI foundation, the Responsible AI principles, and the first suite of GenAI products on the platform. Most recently VP of Advanced Analytics & AI at Virta Health, owning the full data foundation, GenAI in product and operations, and the analytics + ML + data engineering org.

MS Plant Biology & Conservation (Northwestern). BS Mathematics (Northwestern). ML certificate (Stanford via Coursera). One patent. Two AI/ML industry recognitions. Open to VP and Chief AI Officer roles at companies ready to treat AI as an operating capability, not a research line item.

Anne K. Nies

MS · BS MATH · CARSON, WA

CASE STUDIES

Three orgs. Three production wins. One operating pattern.

VIRTA HEALTH · 2024 to 2026

VP, Advanced Analytics & AI

$3M

ORG BUDGET OWNED

THE FOUNDATIONS BUILD

Stood up the data foundation, AI/ML, and analytics org from the architecture up.

Ran a 15-person org across analytics, AI/ML, and data engineering at a metabolic-health company. Defined KPIs, designed the data architecture, led data governance. Shipped insights products to ~200 internal stakeholders and a comparable external customer base. Built predictive modeling for sales-funnel conversion that lifted forecast accuracy from 30% to 70%. Owned tooling decisions on Looker, dbt, and OpenAI. Collaborated with engineering on serving and scaling AI-driven services.

PERSON ORG

30→70%

FORECAST ACCURACY

~400

STAKEHOLDERS SERVED

SUBMITTABLE · 2023 to 2024

Director of Science

FIRST AI PRODUCT SUITE

THE FROM-SCRATCH AI FOUNDATION

Defined Responsible AI principles and shipped the first AI product line on the platform.

Set the strategic direction for AI software, ML, data science, and data engineering at a grant-management SaaS company. Partnered with Product, Marketing, and Engineering to translate business challenges into integrated data-driven solutions. Built the AI foundation: principles, standards, guidelines, leading practices. Shipped the platform's first GenAI suite, a chat-based application builder that helped customers compose multi-step grant applications. Released the first standardized dashboards and a new data-sharing platform on Azure and AWS.

RAI

PRINCIPLES SHIPPED

GenAI

PRODUCT v1 LIVE

CLOUDS (AZURE+AWS)

COHERE HEALTH · 2021 to 2023

Director of Machine Learning

6 → 20

ML TEAM SCALED

THE FLAGSHIP PRODUCTION SYSTEM

40+ models in production. $4M+ in value. 75% faster deploys.

Owned AI software, ML, and automation strategy at a healthcare prior-authorization platform. Grew the ML team from 6 to 20: hired 20 people across two years through deliberate staffing planning, job design, and interviewing. Led the team building NLP, deep learning, reinforcement learning, and LLM models on unstructured clinical documents. Deployed 40+ models with $4M+ in business value over 15 months. The fax-processing automation handled 65%+ of incoming faxes on cloud OCR. Standardized the ML pipeline and cut time-to-deployment 75%.

40+

MODELS LIVE

$4M+

VALUE / 15 MO

65%+

FAX AUTOMATION

THE AI-AS-OPERATIONS OPERATING MODEL

Four moves that get AI out of the notebook and into the P&L.

Built across operations leadership at Vanguard, analytics consulting at Donlen, data science at Allstate and Rally Health, and ML leadership at Cohere, Submittable, and Virta. Applied in order. No optional steps.

START WITH OPERATIONS

Map the business process before you map the data.

The single most common failure mode in AI orgs is starting from the data and looking for somewhere to apply it. The right starting point is the operational workflow that costs the most and breaks the most. At Vanguard, the breaking workflow was invoice creation. At Cohere, it was reading clinical documents to make a prior-auth decision. In both cases, the model only existed because the operational map already showed exactly where the model belonged.

WHAT TO BUILD FIRST

A workflow diagram of the human process. The pain points. The decision points. The volume and the variance. Then a hypothesis for which of those a model can do better.

WHAT TO AVOID

Inventory-led AI strategy. ("We have a lot of customer-support tickets, so we should do something with them.") The data is the second question, not the first.

SHIP TO PRODUCTION

A model that does not serve production traffic is a research expense.

Notebooks are not the deliverable. Dashboards are rarely the deliverable. The deliverable is a model serving production traffic with a measured business outcome attached to it. Everything before that, training, validation, paper-writing, is overhead in service of that moment. Build the deployment pipeline first. Standardize it early. Make shipping the boring, repeatable part of the job, so the team can spend its energy on the modeling.

LEADING INDICATOR

Time from first model commit to first prod inference, measured in days. At Cohere, this came down 75% once the pipeline was standardized.

TRAILING INDICATOR

Cumulative business value across all models in production. The number that a CFO can hold the AI function accountable to.

BUILD THE BENCH

Hire for engineering rigor first, modeling depth second.

A team of brilliant data scientists who cannot ship is a research lab, not an AI organization. Hire MLEs who can write production code, test it, deploy it, and monitor it. Pair them with data scientists who can pick the right modeling approach. Make the engineering practices, code review, CI/CD, observability, the floor of how the team works, not the ceiling. The team I grew at Cohere from 6 to 20 was selected and shaped against this rule.

ROLE MIX THAT WORKS

ML engineers, applied scientists, data engineers, and a small core of research-oriented data scientists. Production capability is in the majority.

ROLE MIX THAT FAILS

All scientists, no engineers. The team can model anything and ship nothing. Watch for it in interview panels and in the headcount plan.

GOVERN AS YOU SCALE

Responsible AI is built in, not bolted on.

The healthcare and regulated-industry experience makes the case clearly. By the time a model is serving real patients or real customers, the governance has to already exist: documented training data lineage, model cards, monitoring for drift and bias, an escalation path when something goes wrong. The work I did at Submittable was to put those principles in place as the foundation of the AI program, not as a compliance afterthought once an incident forced the conversation.

WHAT TO STAND UP EARLY

Responsible AI principles. Model cards as a default deliverable. Drift and bias monitoring on production models. A clear escalation owner.

WHEN TO STAND IT UP

Before the first production deployment. Retrofitting governance onto a scaled system is more expensive than building it in from day one.

CAREER

15+ years of data, ML, and AI leadership.

2024 to 2026

VP, Advanced Analytics & AI

VIRTA HEALTH · METABOLIC HEALTH

Owned the data foundation for business analytics and AI/ML work: KPI definition, data architecture, governance. Ran a 15-person org spanning analytics, AI/ML, and data engineering on a $3M budget. Built predictive modeling for sales-funnel conversion that lifted forecast accuracy 30% to 70%. Shipped insights products to ~400 internal and external stakeholders. Owned the GenAI roadmap in product and operations.

15 PERSON ORG

$3M BUDGET

+40% ACCURACY

2023 to 2024

Director of Science

SUBMITTABLE · GRANT-MANAGEMENT SAAS

Set strategic direction for AI software, ML, data science, and data engineering. Built the AI foundation: Responsible AI principles, standards, guidelines, leading practices. Released the first suite of AI products on the platform, a new data-sharing layer, and the first set of standardized dashboards on Azure and AWS. Championed data culture across Product, Marketing, and Engineering.

RAI FOUNDATION

FIRST GenAI SUITE

2021 to 2023

Director of Machine Learning

COHERE HEALTH · PRIOR-AUTHORIZATION PLATFORM

Defined and led AI software, ML, and automation strategy. Grew the ML team from 6 to 20 across two years. Led NLP, deep learning, reinforcement learning, and LLM development for Document AI on unstructured clinical documents. 40+ models deployed with $4M+ business value in 15 months. 65%+ of incoming faxes auto-handled. Standardized ML processes, cutting time-to-deployment 75%.

40+ MODELS LIVE

$4M+ VALUE

TEAM 6 → 20

2019 to 2021

Manager, Data Science

RALLY HEALTH (OPTUM/UHG) · CONSUMER HEALTHCARE

Led a 5-person data-science team supporting UHG's search product (find, price, and reward care). Built recommender systems, statistical models, and quantitative analyses for stakeholder management. Co-led the readmission-risk model that providers adopted to flag patients needing follow-up. Built a Python module that productionalized and shared the data-science code base. Established ML pipeline and deployment standards for the data engineering team.

5 PERSON TEAM

READMIT MODEL LIVE

2017 to 2019

Senior Data Scientist

ALLSTATE INSURANCE · TELEMATICS / MOBILITY

Hands-on project lead for Mobility Graph, a graph-theory project redefining how location data from mobile apps and vehicle onboard devices is structured and used. Leveraged 1B+ rows of telematics data using Spark, Python, pandas, scipy, and neo4j. Wrote a Python module and built a Hadoop database that opened the data to generalists across the enterprise. Initiated and led a cross-departmental data-science talent program. Patent filed on the underlying ML approach (US 11297466 B1).

1B+ ROWS

PATENT FILED

2014 to 2017

Senior Data Strategist

DONLEN CORPORATION (A HERTZ COMPANY) · FLEET

Provided analytic and strategy coaching to financial and procurement leaders optimizing corporate fleet strategies across multiple industries. Managed a 10-client portfolio. Identified $27.7M+ in client operational cost savings and avoidance through listening to client needs and applying data-science best practices. Built three predictive and ML models for optimizing fleet operations using Python, scikit-learn, and forecasting techniques.

$27.7M+ SAVINGS

10 CLIENTS

2006 to 2012

National Import Product & Operations Director

VANGUARD LOGISTICS SERVICES · GLOBAL FREIGHT FORWARDING

Ran North American Import Product and Operations: a 135-person division across six offices. Initiated, developed, and launched two major operations software enhancements that impacted 3,000+ staff using AI, operations research, and deep operational knowledge. Recovered $1M in outstanding revenue by spotting and fixing an operational-model gap. Improved operational efficiency 25% via strategic software, including automated invoice creation. Improved employee retention 60% by restructuring around aptitudes. This is where the AI-as-operations doctrine started.

135 PERSON DIV

+25% EFFICIENCY

+60% RETENTION

Models in production.
Value on the P&L.

The deliverable is not the model.
The deliverable is the outcome.

Forty models. Fifteen months. One operating pipeline.

The problem I solve, and how I got here.

Three orgs. Three production wins. One operating pattern.

Stood up the data foundation, AI/ML, and analytics org from the architecture up.

Defined Responsible AI principles and shipped the first AI product line on the platform.

40+ models in production. $4M+ in value. 75% faster deploys.

Four moves that get AI out of the notebook and into the P&L.

Map the business process before you map the data.

A model that does not serve production traffic is a research expense.

Hire for engineering rigor first, modeling depth second.

Responsible AI is built in, not bolted on.

15+ years of data, ML, and AI leadership.

Education, patents & recognition.

Ready to treat AI as an operating capability?

Models in production. Value on the P&L.

The deliverable is not the model.The deliverable is the outcome.

Forty models. Fifteen months. One operating pipeline.

The problem I solve, and how I got here.

Three orgs. Three production wins. One operating pattern.

Stood up the data foundation, AI/ML, and analytics org from the architecture up.

Defined Responsible AI principles and shipped the first AI product line on the platform.

40+ models in production. $4M+ in value. 75% faster deploys.

Four moves that get AI out of the notebook and into the P&L.

Map the business process before you map the data.

A model that does not serve production traffic is a research expense.

Hire for engineering rigor first, modeling depth second.

Responsible AI is built in, not bolted on.

15+ years of data, ML, and AI leadership.

Education, patents & recognition.

Ready to treat AI as an operating capability?

Models in production.
Value on the P&L.

The deliverable is not the model.
The deliverable is the outcome.