harnessing the speech analytics advantage
Post on 11-Feb-2017
268 Views
Preview:
TRANSCRIPT
» insights
harnessing the speech Analytics AdvantageHow split-second analysis of conversations can boost compliance and performance
Conversations with consumers—whether asking for payment or answering product questions—
are an expensive form of contact, fraught with compliance risk. When handled unsuccessfully, they
can negatively impact the bottom line, as well as increase risk of fines and reputational damage for
regulatory noncompliance.
Phonetics-based speech analytics provides a scientific, scalable way of evaluating and improving
conversations with consumers. The technology is helping companies lower compliance risk while
lifting agent productivity by double digits.
By combining speech analytics with other analytic techniques like predictive models, companies
can boost performance even further and build more valuable consumer relationships. They can
identify the precise conversational characteristics and agent behaviors that are predictive of specific
outcomes—such as, in sales, higher revenues; in support, shorter call handling time with higher
caller satisfaction; and in collections, higher payment rates and fewer compliance mistakes.
This paper looks at how companies can use these analytic insights to:
• Increase—and prove—regulatory compliance.
• Detect the conversational characteristics of best (and worst) performers.
• Score agents for likelihood of achieving targeted outcomes with
specific types of calls.
• Provide score-driven guidance to help all agents converse like the best.
• Generate supervisor dashboards and alerts for when to intervene.
• Lower human capital costs by accelerating agent training
and reducing churn.
Number 76
Companies using speech analytics often see double-digit performance gains—like a 30% increase in cash collected per agent hour
www.fico.com Make every decision countTM
www.fico.com page 2
Harnessing the Speech Analytics Advantage
» insights
Every contact center manager has experienced the good fortune of hiring the occasional
outstanding agents who do every aspect of the job exceptionally well. These agents rarely make
compliance mistakes, and they consistently achieve call objectives in a short amount of time
with high consumer satisfaction. “If only all my agents were just like them!”
But wishful thinking is one thing, and the daily challenge of running a business is another.
Complying with regulatory requirements is a big part of this challenge, as agencies—such
as the Financial Conduct Authority in the UK and the Consumer Financial Protection Bureau
and Federal Trade Commission in the US—expand oversight and step up investigations into
consumer complaints. As a result, many organizations are now focusing a good deal of attention
on just one aspect of agent performance: avoiding compliance mistakes.
Speech analytics is essential for achieving that objective. It monitors and performs detailed
checks on 100% of calls—consistently, reliably, through volume spikes, day in and day out.
Case in point: A collections company used the technology to free up 30 hours a month that
supervisors had previously spent listening to calls, redirecting this time into weekly agent
coaching sessions.
Supervisors also have the visibility to provide agents with immediate guidance or intervene to
correct mistakes while a call is still underway. As shown in Figure 1, analytics-driven dynamic
dashboards show agent compliance statistics and alert supervisors to conversations that require
their attention.
» Make Every Agent a Top Performer
Figure 1: A supervisor dashboard showing agent compliance statistics
April 2014
www.fico.com page 3
Harnessing the Speech Analytics Advantage
» insights » insights
This blanket call monitoring also enables
companies to fully assess compliance risk
exposure. As depicted in Figure 2, managers
no longer have to extrapolate compliance risk
based on a tiny random sample, high-level
aggregate statistics and anecdotal evidence.
They have statistically reliable data-driven
statistics with complete drill-down details.
Speech analytics indicates precisely how
many violations were made, which agents
made them and which consumers they were
speaking with. Complete audit data is available
to show regulators. The collections company
mentioned above has been able to reduce
the time it takes to respond to a compliance
inquiry from as much as two weeks to just
ten minutes.
In addition, speech analytics that capture
phonetic patterns (see sidebar on the next
page) provides companies with potential
visibility into a wider range of performance
dimensions. The prevalence of certain patterns,
forming phrases such as those shown in Figure
3, may expose the root causes of problems. For
example, a health plan contact center was able
to pinpoint a small flaw in its mailing process
to health care providers that was causing big
spikes in incoming call volume and average
handle time (AHT).
Phrase prevalence and order may also reveal
patterns in the conversations of outstanding
agents and help other agents emulate them.
A bank where customers were having trouble
with a new ID verification process used
this approach to understand how its most
successful personal bankers were positioning
the process to customers and when they
offered appropriate alternatives. Sharing these
insights in targeted training with other agents,
the bank increased customer verification rates
by 25%.
To fully realize the opportunity to understand and unleash drivers of higher performance,
however, companies need to apply predictive models and other analytic techniques to the
phonetic data.
Figure 2: Speech analytics replaces guesswork with evidence
Fractional monitoring 100% monitoring & analysis
+ aggregate statistics+ anecdotal insights= guesswork
+ detailed statistics(aggregate, team, agent)= evidence
Figure 3: Prevalent topics can reveal call context and point to the root causes of problems
www.fico.com page 4
Harnessing the Speech Analytics Advantage
» insights
With 100% of calls being processed by speech analytics, there is abundant historical data. Data
mining techniques can probe the historical data to find more complex and subtle characteristics (of
the agent, the consumer and previous agency interactions with that consumer) that can be used in
predictive models to determine the probability of specific outcomes. The algorithm might discover,
for example, conversational characteristics predictive of a low AHT in calls of a certain nature, with
consumers of a certain type.
Models incorporating multiple characteristics like these can analyze the conversations of individual
agents, scoring them for the likelihood of achieving targets for key performance indicators. Scoring
enables companies to identify where agents need help to improve, and to rank-order these needs
so that supervisors can prioritize coaching.
In this way, companies help agents adjust their behavior to the conversational patterns that have
been empirically proven to work. They shift the focus off of the lowest common denominator of
agent performance—no compliance mistakes. Speech analytics and predictive analytics provide
a data-driven way to work toward raising the performance of all agents to the level of the top
performers—no mistakes and best call results.
Phonetics-based speech analysis finds patterns of sounds in
speech. this powerful and flexible analytic method has several
advantages for contact center management. these include
real-time recognition of what agents and consumers are saying,
as well as speed in adapting to new regulations and customer
experience criteria.
A phoneme is the smallest unit of sound capable of conveying a
distinction in meaning in a language. For example, in English, the
word “best” is formed by the phonemes /b/e/s/t/; swap the /b/
sound for an /r/ sound, and the resulting word “rest” has a
different meaning. Changing the sequence of the phonemes in
“task” (/t/a/s/k/) produces “asked” (/a/s/k/t/).
Analytics that detect phonemes can recognize not only words,
but phrases and sequences of phrases in contact center speech.
By generating a phonetic index layered with a time-aligned index,
this technology can also determine when, during the duration of
a call, agents and consumers say specific things. For example, did
a collections agent make the so-called “mini-Miranda” disclosure
statement at the beginning of a call, and did she ask for full
payment before suggesting a settlement? Did a sales agent
make the most profitable offer first?
How does phonetics-based speech analytics work?
When a leading collection agency applied speech analytics to a
sample of calls, the company identified more than $200,000 in
potential violations of Fair Debt Collection Practices Act (FDCPA)
regulations. the company also discovered its agents were failing
to ask for payment in 60% of conversations, resulting in high
volumes of callbacks and other operational inefficiencies. When
agents did ask for payment, they frequently failed to aim initially
for full payment. instead they often immediately offered a 50%
settlement. the cumulative result was an estimated $800,000
left uncollected.
speech analytics has enabled the company to address both
problems. Within six months of implementing the technology,
the company has substantially reduced its compliance risk
while achieving:
f 30% increase in cash collected per agent hour
f 15% increase in RPC (right-party contact) promise-to-pays
f 30% reduction in call monitoring hours per manager
f 50% increase in agent coaching time
Speech analytics in collections
www.fico.com page 5
Harnessing the Speech Analytics Advantage
» insights
Speech analytics, especially when used with data mining and predictive models, enables
more successful conversations with consumers. In general, a successful conversation is one
that achieves all call objectives in the shortest amount of time, while being fully compliant and
leaving the consumer with an increased level of satisfaction. By identifying the conversational
characteristics most predictive of positive outcomes, these analytic techniques provide
companies with an empirical method of driving conversation success rates.
Every company will, of course, have
additional success criteria specific to its
organization or to a current initiative. An
advantage of these analytic techniques is
that they can be used to detect and measure
virtually any mix of criteria that make up a
contact center’s current definition of success.
Dashboards like the one shown in Figure 4
are easily configured to support changing
definitions and bring attention to what’s
most important now.
Phonetics-based speech analytics also easily
adapts to dynamic business environments
and new regulatory requirements. Unlike
dictionary-based systems that look for
words they’ve been trained to understand,
no retooling is required to accommodate
change. As new regulations come out,
or as new phraseology comes into use
with changing call strategies and market
conditions, the analytics automatically
begins detecting the phonetic patterns in
current conversations. And the technology
can also re-analyze historical data to look for
new patterns.
This flexibility is what makes the combination
of phonetics-based speech analysis and
predictive analytics so powerful for driving
conversation performance. An agent score
can be built for any call performance success
criteria (e.g., in collections, % of balance
collected, % of calls with promise to pay,
average length of call) where the outcomes
are known from historical data. A variety
of predictive modeling techniques, as well
as descriptive techniques such as cluster
analysis, could be used to identify the agent
conversational characteristics that are the
strongest predictors of these outcomes.
» Raising Conversation Success Rates
Figure 4: Dashboards are easily configured to show any mix of KPIs
Speech analytics in health care plan administration
in just four months, a major health care insurer reduced average handle time by
42%, for a projected yearly savings of $900,000. this result was achieved by
using speech analytics to not only improve agent training in first call resolution,
but also to correct a process gap.
Speech analytics in telecom customer serviceA leading outsourcing specialist used speech analytics to pinpoint the root cause
of rising average handle time: some agents struggling to explain a complex bill
process were losing control of calls. targeted coaching reduced these agents’
call times by an average 42 seconds, saving £200,000. speech analytics also
discovered process discrepancies that, when standardized, shaved an average
three minutes off of calls about faulty handsets replacement, saving £776,000.
www.fico.com page 6
Harnessing the Speech Analytics Advantage
» insights
As companies put more scorecards like these behind dashboard displays, they gain a wider, clearer
window into what’s driving or limiting their success. Agents and supervisors see scores on a
call-by-call basis, as well as trend lines. Training goes from an infrequent, expensive, after-the-fact
process to an activity that is integral to work and ongoing. In addition to pinpointing for supervisors
where an individual needs coaching, scores can drive game-like dashboards. Agents self-train by
competing against their own personal bests as well as group averages.
While analytics can help all agents raise their overall performance, it can also help match them
up with the types of calls and consumers they are most effective at handling. A machine learning
analytic technique, Latent Dirichlet Allocation (LDA), is one way contact centers could determine
where agents are likely to have the greatest success.
LDA is often used to analyze unstructured data, such as conversations, to identify similarities
between consumers. Mining large quantities of historical speech data, the algorithm discovers
archetypes that can help companies adjust their treatments, or
conversations, to these consumer categories. For example, in a collections
call center’s historical data, an LDA algorithm might discover archetypes
such as “Recent job loss but committed to repaying debt” and “Originally
intended to pay but considering a strategic default.”
As shown in Figure 5, the collections operation could then use speech
analytics and predictive models to score the performance of agents in
conversations with consumers that map to these archetypes. Business rules
could be written to prompt agents to direct certain types of calls to the
agents that score best with them. Based on the nature of the call, rules-
driven automation could prompt the agent to bring the right additional
resources into the call (“Excuse me, sir, I have a colleague who can help you
with this matter. May I put her on with you?”).
Such tactics may improve the outcome of the conversation. And, when
callers feel their need is being escalated quickly to the appropriate expert,
their satisfaction may increase.
FICO® Engagement Analyzer is a speech analytics
solution that enables collection organizations to gain a
multidimensional view of how their agents collect debt
from consumers. By indexing, searching and analyzing
content from all recorded conversations between agents
and debtors, Engagement Analyzer delivers accurate,
detailed performance statistics on every debt collection
agent. this application provides scalable, real-time analysis
of collections conversations, with results driving immediate
feedback to agents as well as to supervisor dashboards.
Driving collector performance and compliance with speech analytics
Figure 5: LDA learns about each agent’s strengths with every call
Her top scores are mapped to...Lindsay’s conversations with consumers
...consumer archetypes discovered by the machine learning algorithm, which mines large amounts of phonetic data from historical contact center conversations
Archetype A Archetype B Archetype C Archetype D Archetype E
Lindsay excels at handling these situations
SPEECH ANALYTICS /a/s/k/t/ /b/e/s/t/
+ PREDICTIVE ANALYTICSp(β1:K , θ1:D , z1:D , w1:D)
= ∏ p(βi) ∏ p(θd) i=1
K D
d=1
�∏ p(zd,n | θd)p(wd,n|β1:K zd,n)� N
n=1
www.fico.com page 7
Harnessing the Speech Analytics Advantage
» insights
Another developmental direction for improving conversation success may be to combine speech
analysis and LDA with strategy adjustment or optimization. In collections, agents often follow a rigid
“ladder” of steps that may eventually lead to a settlement offer. In many cases, however, this process
causes long AHT due to excessive negotiation and can result in less than optimal recovery amounts.
But speech analytics can prevent this waste by discovering topics in the conversation that trigger
rules guiding agents to skip over rungs of the ladder when a truncated course of action will be
more effective.
Topic discovery could be used to initiate real-time strategy optimization. For companies leveraging
optimization, the best settlement offer for each delinquent consumer has generally been
determined prior to the collections call. The process involves modeling all the factors in the decision
(see Figure 6), then using mathematical optimization to balance multiple objectives (e.g., amount
paid, loss, collection costs) and constraints (e.g., collection capacity, loss rate, interest revenue) to
maximize an overall goal (e.g., five-year profit). But the optimal settlement can be adjusted. Topics
discovered by speech analytics could trigger a real-time analytics engine to re-run the optimization
based on the how the conversation is progressing.
Figure 6: Mapping relationships between factors in a complex decision
Inputs Modifications Predictions Business Metrics Objective
NPV (Net Present Value)
p(Able to Pay)
Time to Liquidate
Monthly Payment
p(Walkaway)
p(Liquidation)
Future Value of Asset
Client PredictiveModels
Income,Household Size
Economic Forecast
Current Loan Amount,Interest Rate and Term
Current Valueof Asset
Region
Interest Rate, Term,Forbearance
RestructureShort Sell
PV of LiquidationCash Flows
PV of AlternativeCash Flows
Bureau Data
Mapping relationships between factors in complex decisions
This is a simplified view of a collections decision model that predicts the impact of likely customer reactions to a loan modification offer.
Harnessing the Speech Analytics Advantage
» insights
The Insights white paper series provides briefings on research findings, technology innovations and recommended best practices
from FICO. To subscribe, go to www.fico.com/insights.
FICO and “Make every decision count” are trademarks or registered trademarks of Fair Isaac Corporation in the United States and in other countries. Other product and company names herein may be trademarks of their respective owners. © 2014 Fair Isaac Corporation. All rights reserved.3088WP 04/14 PDF
Speech analytics is an essential technology for contact centers working to ensure regulatory
compliance and increase agent productivity. When used with predictive models and
other analytic techniques, speech analytics enables contact centers to identify the precise
characteristics of successful conversations—fully compliant and achieving call objectives in a
timely way—and help all agents increase their success rates. It’s a great demonstration of how
technology can be used to empower people—in this case, the fundamentally human activity
of talking to each other.
To learn more about the latest in unstructured data analytics, visit the FICO Labs Blog or read these
Insights white papers:
• Extracting Value from Unstructured Data (No. 71)
• Is It Fraud? Or New Behavior? (No. 69)
• When Is Big Data the Way to Customer Centricity? (No. 67)
» Conclusion: More Successful Conversations
For more information North America Latin America & Caribbean Europe, Middle East & Africa Asia Pacific www.fico.com +1 888 342 6336 +55 11 5189 8222 +44 (0) 207 940 8718 +65 6422 7700 info@fico.com LAC_info@fico.com emeainfo@fico.com infoasia@fico.com
top related