Tax Collections Optimizer - New York State Office For Technology

ovalscissorsInternet et le développement Web

30 juil. 2012 (il y a 6 années et 2 mois)

651 vue(s)

Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
1

of
7






















Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
2

of
7



B.
Executive Summary


With a deficit of
more than

$9 billion,
New York
State
is facing historic budget shortfalls
that have forced legislators to make deep cuts in programs and dire adjustments to
hiring.


However,
the
New York State

Department of Taxation and Finance
(DTF)
ha
s

added cutting
-
edge smart technology to
an
already impressive tax collections/

enforcement arsenal

to
obtain delinquent taxes and
help the State close the deficit gap
.

The Tax Collections Optimizer

is a recent e
xample of such innovation.


Since 1999, DTF has partnered with IBM to
use operational analytics to identify
suspicious tax returns, detect suspect filing patterns or questionable returns and
clamp down on civil and criminal tax fraud
. Through this collab
oration, the DTF
developed analytical applications to identify questionable refund claims, which
,

since
2004,
has resulted in retrieving over $1 billion in owed taxes
.


Through

continued

collaboration with IBM,
DTF added the
Tax Collections
Optimization Sy
stem

to its
pioneering

approaches around tax compliance and
collections
in 2009
.
This

new tool

will

bring in an additional $100
m
illion over a
three
-
year period
,

and this amount is
expected to rise substantially
overtime
after
as
more data is collected.


The Tax Collections
Optimizer
uses

a
unique combination of data analytics and
other models to create action plans for each case
. The
action
plan optimizes the
order of activities agents will take to maximize the total amount of debts collected while
takin
g into consideration the caseload, personnel resources, and the anticipated
effectiveness of the suggested actions.


The solution optimizes the collection actions of agents by taking into account the
complex dependencies between resources, business needs a
nd legal constraints.
Using a variety of taxpayer data, such as the amount owed and past payment history,
the system develops a plan for collecting from the entire population of delinquent
taxpayers. The system creates a work flow that automatically look
s for the focus areas
that would
maximize
the best use of an agent’s time. For example, the system
determines whether the next collection attempt should be a letter, a phone call or a
personal visit while weighing those actions against the potential resul
t of that action.
Th
e

use of advanced analytics improve
s

the collections results and help
s

enhance the
productivity of agents.


T
he Tax Collections Optimiz
ation System

has received significant press attention,
including a feature
story
on
a prominent New Y
ork City news channel.


This application addresses NASCIO’s State CIO Top Priorities for Budget and Cost

Control; Consolidation of Resources; and Priority Technologies, Applications and Tools:
Business Intelligence and
Business Analytics Applications.

Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
3

of
7


C
.
Description


With government budgets feeling the impact of
the current economic climate, tax collections
are quickly emerging as a renewed focus area.
It is estimated that the average state has
billions of dollars in delinquent taxes. As
governments
worldwide continue to focus on
increasing revenue, they are looking for new
and innovative approaches to improving
collections with fewer resources.



The problem of optimally managing the
collection process by taxation authorities is one of prime importan
ce, not only for the
revenue it brings but also as a mean
s

to administer a fair taxing system.

The analogous
problem of debt collection management in the private sector, such as banks and credit
companies, is also increasingly gaining attention.
Since 20
04, Department of Taxation
and Finance has saved
more than
a $1 billion identifying questionable refund claims
through the
application of data analytics

and last year targeted improvement in their
collections processes with the use of leading edge data mod
eling and optimization
techniques.


Problem

The current processes used to recover unpaid funds by tax agencies are complex,
outdated, costly, and generally ineffective.

While Department of Taxation and Finance
strategically focuses on the use of technol
ogy in pursuit of activities to move taxpayers
“up” the Compliance Continuum to greater voluntary compliance not all taxpayers are
ready, willing, or able to pay what they owe. Effective and coordinated audit, collection,
and criminal investigation activi
ties to deter tax avoidance and abuse, under reporting,
and failure to file are fundamental to the support of a compliance system.


The tax/debt collections process is complex in nature and its optimal management
need
s

to take into account a variety of con
siderations. At a high level, the collections
process management problem is that of determining the answers to the following
questions:


1.

Which of the debtors should be approached
?

2.

Which of the possible collection actions are to be taken
?

3.

Who should take t
hose actions
?

4.

When should they be
taken?

Collecting Delinquent Taxes

T
o Ensure Fairness

Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
4

of
7


T
he answer to each of these questions will
depend on a number of factors. The answer to
the first question will depend on the information
the collection agency may have on the debtors,
such as the demographics and

the amount of
debt owed. The
answer to the second will also
depend on the nature and status of the debtor,
such as how collectible they appear to be, since
actions of varying severity may be appropriate
for debtors in different categories. The third wi
ll
additionally depend on what resources are
available within the collection organization.
With regard to the fourth, there are several
complication
s

that impact the optimal timing to
take actions on debtors. In the tax collection
case, there are certain

legal requirements that
govern the sequential course of collection actions.

There are also business
considerations that affect the appropriate timing of a collections action in general.



Solution

Due to the high complexity involved in the collection pro
cess, and because of legal and
business constraints, it is common practice to follow rigid manually constructed rules to
guide the collection activities. DFT’s Tax Collections Optimizer developed a collection

process automation system
primarily based on d
ata modeling and optimization and
accepts input from a rules engine as constraints on the optimization process.


The Tax Collections Optimization System is an event driven
,

end to end collection case
management system
using
the power of mathematical modeli
ng combined with the
implementation of business rules that are both legal and department policy

to determine
actions
taken on tax collection cases. The system takes actions recommended by an
optimization

model and delivers them to DFT’s

Websphere Process
Server (WPS)
software to carry out the actions. WPS
carries

out the actions by integrating with the
legacy
Case and Resource Tracking System

(
CARTS
)

collections system u
sing

business components and smart services. Work

lists are used to deliver case work

to
collection representatives in the call center. Graphical User Interface screens were built
to provide a more user friendly depiction of the traditional CARTS green screens.



The
goal of the system is to allocate the

best

use of
Collections and Civil

Enforcement
Division

(
CCED
)

resources to collect tax debt owed. The optimization model u
ses

a
snapshot of the current state of the taxpayer obtained through our first iteration of a
Collection Taxpayer Profile (an account of taxpayer interactions with th
e department
over time) along with business rules that determine if an action can be performed based
on both legal and department policy. When an event occurs or a period of time has
passed the collection case is re
-
evaluated to determine if a more approp
riate action
should be taken then what is occurring at the present time. The model takes into
“IBM Helps New York

Go After Tax Deadbeats”

money.cnn.com 4/12/20



Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
5

of
7


account resources available at the time of
the recommendation along with the cost
associated to taking the action.
For
example,
the results may suggest that
con
tacting certain delinquent sales tax
vendors by phone or in person is more cost
-
effective
than
sending a notice that
emphasizes

potential criminal prosecution.



D.
Significance


DTF

use
d

a novel approach based on the
framework of constrained Markov Decis
ion Processes (MDP), and developed concrete
algorithms and methods for performing data modeling and resource optimization in a
unified fashion, while addressing all of the considerations
to the collections questions
.
The constrained MDP formulation, allow
s DTF to take into account the various
constraints that govern the actions under consideration. Th
is

framework provides a
comprehensive solution to the problem of collections optimization, which tightly couples
data modeling and constrained optimization i
n a unified manner.


A tax collection optimization engine was implemented and deployed as part of a larger
collection management solution at the Department of Taxation and Finance. The
collection process starts on a
taxpayer when an upstream process (e.g.

auditing) makes
the determination that a certain taxpayer has an outstanding debt and an assessment is
created on that taxpayer.

D
ata consisting of the taxpayers’ background information,
their complete history of transactions (payment
s) and collection ac
tions taken is
collected and
sent
,
by default
,
to the Call Center (CC). The case will belong to CC for a
certain period of time, and various contact and collection actions can be taken (e.g.
mailing, phone call, warrant, levy, etc.) either manually by CC
staff or sometimes
automatically. When
a
determination is made that the collection process should be
elevated for a given case, the case is moved to a more specialized organization
. These
organizations
are
those that handle certain specialized types of c
ases or the district
offices that handle elevated cases belonging to specific regions.

“NY Turns $10 Million Investment

into $1 Billion in Savings”

gcn.com 5/17/10


Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
6

of
7


Chart 1: The
Collections Systems
has a complex,

but effective

IT

a
rchitecture
.

Chart 1 is
a schematic of the overall collections systems architecture:





Given this overall flow of the collections process, the task of the recommendation
engine,
based on
D
TF
’s
modeling and optimization engine, is divided into the modeling
phase and the scoring (action allocation) phase. In the scoring phase, it is given as
input: (1) a set of collection actions under consideration; (2) a set of modeling feature
ve
ctors for taxpayers of interest; (3) a set of constraint feature vectors for the same
taxpayers, specifying for each taxpayer which of the actions are permissible (obtained
by applying business and legal rules to their profile data); (4) a description of t
he
available resources in terms of man hours in each of the multiple organizations; (5) a list
of additional constraints in terms of u
p
per and lower bounds on the type of actions that
are performable; it is to output an optimize
d

action allocation, mapping

each of the
taxpayers in (2) to a recommended action.


In addition to the modeling features, the engine makes use of another group of features
called action constraints, which are binary features
speci
f
ying
, for each of the actions
considered, whether or
not the action is a
llowed on the case at that time
, according to
the business and legal rules.


I
n deploying
the

optimization

engine,
D
TF

made a few extensions and enhancements to
the modeling engine.

One extension that is worth mentioning has to do with
the way the
segmentation is done in the modeling engine.

The regression tree modeling engine
use
d

performs tree
-
based automatic segmentation of the feature space into a number of
Category: Improving State Operations


NYS DTF
-
-
Tax Collections Optimization

Page
7

of
7


segments, which are uniform with respect to the regression of the objective
function and
large enough to admit sufficient statistical significance.

While automatic segmentation
is critical for our purposes, there is a
motivation to guide the segmentation
process with some coarse segmentation
that is derived from the domain
knowle
dge.

For example, the stages
within the collections process, such as
whether the case is in the call center or in
a district office, or whether the case has
been warranted, have a major impact on
the way they should be handled.


Benefit of the Project

Th
e
Department of Taxation and
Finance’s strategic goal is to achieve
greater voluntary compliance with tax filings and regulations and to improve the ways
taxpayer data, returns, and payments are processed.
DTF continually seeks to
enhance

and improve the
collection and enforcement activities through:




Improve audit selection and speeding up the audit cycle;



Providing automated services to i
ncrease the rate of collections;



Improve methodologies to identify and pursue non
-
filers and fraudulent filers; and



E
stablish a high dollar, high risk enforcement program
coordinating audits, collections and criminal
enforcement activities.


Tax Collections Optimization
process
is
a novel approach to
the debt collections optimization problem

and
bolsters

resources alread
y at
Tax’s

disposal
to
prosecute

and d
eter
noncompliance, including:




O
ur
Case Identification and

Selection System (CISS),
which
screens the flow of return data for suspicious returns.



Office of Tax Enforcement

Data Resources Unit, which
analyzes all the

data
at
our disposal
to detect suspect filing patterns, questionable returns, and to home
in on the most productive enforcement opportunities.



Special investigations units that have joined prosecutors statewide to clamp
down on civil and criminal tax fraud
.


The system went live in December of 2009
and the

state expects savings of about 100
million dollars in the next three years.



Tax Evaders Beware:


NY State Has An App For That”

5/
12/10 fcw.com