Decision systems for customer-led growth

Most companies have the data.
Few have decided
what to do with it.

Most companies already have models and dashboards. The value appears only when the business changes what it does next — repeatedly, transparently, and with clear ownership.

Former Customer & Marketing Data Science Leader · Macy's
Columbia University · MS Operations Research
Focus: Growth, retention, LTV, and experimentation
$570M+ Incremental revenue impact
25+ Decisions operationalized
12 Years across four industries
Discuss a decision →
Decision system
Signals
browsing, purchase, engagement
Customer State
curiosity, evaluation, purchase readiness
Decision
what the system should do
Action
intervention, personalization
Outcome
revenue, repeat rate, CAC payback
Learning Loop
outcomes refine the decision logic over time
Most companies build the left side of this system.
Audiences AI focuses on designing the decision layer in the middle.
The Problem

Your team knows
something is
off.

Symptom
Data and models exist. Decisions haven't changed.
Scores sit in dashboards. LTV estimates circulate in decks. The meetings where decisions actually happen don't reference either.
Pattern
The model keeps improving. The outcome keeps not moving.
Each iteration makes the model more accurate. The business process it was built to change has not been touched.
Diagnosis
No one in the organization is responsible for the decision layer.
Strategy exists. Models exist. The layer that connects model output to what the business does next sits between functions and belongs to none of them.
If any of these sound familiar, that's where we start.
You may recognize this pattern
These signals often appear when companies have data but lack a decision system.
Decisions still happen in meetings.
Teams have models and dashboards. The final call still happens in discussion, by whoever argues most convincingly that week.
The same growth decisions repeat with no accumulated logic.
Which customers to retain. How much to spend on acquisition. When to intervene. Each cycle starts from scratch.
Experiments run. Learning does not compound.
Tests produce results. Results get reviewed. The organization never builds a decision framework from them.
When these patterns appear, the issue is rarely data.
It is the absence of a decision system.
The Work

Decision architecture
in practice.

Most companies already have data and models — what they lack is clarity on which decisions the system should own. Below are examples of the types of decisions Audiences AI helps companies operationalize.

Frameworks and outcomes are real. Client names are kept anonymous by design.

Omnichannel Retail · Retention & LTV
LTV Models That Changed Personalization Decisions
Goal
Improve retention and LTV without over-intervening on customers who would have returned regardless.
Decision
Which customers belong in which segment — and what does each receive?
Outcome
$55M+ incremental · 3% churn improvement
Omnichannel Retail · Personalization
Recommendations Architecture Revamp
Goal
Improve recommendations relevance and boost conversion rate.
Decision
Which product should this customer see next?
Outcome
$40M+ incremental · +1.6% conversion
Omnichannel Retail · AI Governance
AI Council — From Committee to Decision Engine
Goal
Govern AI decisions at scale without adding overhead.
Decision
Which AI outputs are approved for full automation?
Outcome
Prioritization scoring system · build-vs-buy decisions reduced by 6 weeks
See all case studies → See the framework →
How work starts

Every engagement starts
with one decision

Rather than launching a large program, we start with a focused decision sprint.

In roughly 30 days, the goal is to identify where value leaks, define the logic, and test whether the decision produces measurable lift.

If the decision proves valuable, the system expands from there.

01
Identify where value leaks today — a recurring decision made by humans despite available data.
02
Define the signals and decision logic — what the system should know and what it should decide.
03
Design how the system should act — actions, ownership, thresholds, and reason codes.
04
Test whether the decision produces measurable lift — with a learning loop that improves over time.
If the decision proves valuable, the system expands from there.
About

Rohan Talati

Founder, Audiences AI

Audiences AI designs decision systems that turn customer data into automated business actions. Getting an organization to trust a model enough to stop overriding it — that was the work.

Twelve years inside the problem across omnichannel retail, financial services, and e-commerce. The frameworks are calibrated to how organizations actually behave — not how they behave in case studies.

Audiences AI operates between the C-suite setting direction and the data teams building systems — the only position from which you can see the full gap and close it.

Former Sr. Director, Data Science — Macy's
Columbia University — MS Operations Research
HBR Advisory Council Member
Full methodology and background →
By the numbers
$570M+
Revenue impact
25+
Decisions operationalized
12
Years across four industries
Works at two levels simultaneously: decision authority and governance at the leadership level; signal architecture, model integration, and learning loops at the execution level.
Work Together

The conversation
starts with one
question.

Most engagements begin with one question: which decision in your business most frequently requires a meeting despite having data available?

That conversation is 30 minutes. It is free. What it surfaces usually is not.

Book 30 minutes →
Free diagnostic conversation · No commitment required
Email contact@audiencesai.com LinkedIn linkedin.com/in/rohanstalati
Based Ahmedabad, India · Working globally

Every engagement starts with one decision.
If the decision proves valuable, the system expands from there.