How Afiniti AI Works
How our AI learns, pairs, and optimizes performance responsibly
Afiniti as a fully managed service
Historical data is used to build interaction algorithms, which predict the outcome of all possible caller/agent pairings based on previous similar interactions. Data is refreshed nightly, so the algorithms continue to learn as caller and agent behavior evolves.
Precise attribution through unbiased benchmarking
Afiniti is the only company that uses built-in benchmarking to measure the exact incremental value our technology delivers. By turning our AI “ON” and “OFF” throughout the day, we are able to prove the real impact on your key business metrics.
Factors That Affect Performance
Data
Model performance is directly correlated with the amount of data available for each caller and agent
- Telephony
- Outcomes
- Contextual
- Operational
- Third-party vendor sourced data
Variability
Variation in possible matches is the basis for optimization
- Caller tenure
- Caller history
- Caller profile
- Caller intent
- Agent experience
- Agent work shift
- Offers available
- Product mix
- Competitor
- information
Scale and Choice
As free agents or calls in queue accumulate, our opportunity to pair increases
AI Management Framework
Input data
Input data, vetted and approved by the client, goes through rigorous selection process to ensure merit and deter bias
Data used
- Telephony
- Outcomes
- Contextual
- Operational
- Third-party vendor sourced data (if client approves)
Data not used
- Agent surveys
- Protected class data
- Web scraped data
Modeling & testing
AI modeling and testing, subject to robust AI governance process and controls, ensures operational stability and successful outcomes
Data not used
Controls ensure run time algorithms don’t affect operational performance outside of approved SLAs. Should the Afiniti service be impacted, the routing will revert to the client’s legacy routing.
Interaction models
Controls ensure run time algorithms don’t affect operational performance outside of approved SLAs. Should the Afiniti service be impacted, the routing will revert to the client’s legacy routing.
Validation
Multipronged model validation and testing of comparative advantage and pairing strategies to confirm operational performance.
Outputs & analysis
Client specific dashboards ensure transparent outputs across performance and operational metrics
Sample dashboards
- Telephony data
- Outcomes data
- Operational data
- Operational data
Periodic bias testing
Periodic testing conducted to ensure models and data used don’t introduce bias (in conjunction with our clients, since Afiniti does not access/process customer protected data).
Client compliance analysis
Afiniti provides which caller was paired with which agent and outcomes data allowing the client to conduct independent analysis to ensure risk and compliance adherence.
Input Data
Afiniti builds its pairing algorithms and models using multiple data sources that enable identification of patterns of success. Each data element is vetted and approved by the client and is fully documented. Additionally, Afiniti conducts account level reviews to ensure all data are collected and consumed ethically.
Data used
Telephony data
Outbound and inbound call details sourced from the dialer/ACD and/or CTI engine I.e. dialer data, IVR attached data, DNIS, ANI, call start & end times, agent ID information.
Operational data
Environment information, used to understand agent behavior and events affecting call center (as necessary) i.e. agent compensation, skilling and training.
Third-party data
Commercially procured data on consumers to aid in matching and identification.
Outcomes data
Outcomes of interactions between callers and agents I.e. new customers, customers retained, cross-sell offers accepted, promises to pay kept.
Contextual data
Data about callers and their recent interactions I.e. CRM data, recent journey analytics and any internal customer analysis / segmentation.
Data not used
Agent surveys
Protected class data
Web scrapped data
How Afiniti uses data
Data transfer
Two-phased process:
- Secure read-only access to enable discovery of relevant data for Afiniti’s AI solution
- Transfer only the minimal data for consumption, respecting data ownership and privacy
Data processing and matching
Afiniti performs data processing (i.e. matching of calls and outcomes) on the Afiniti appliance
Data design
The client provides business logic to extract certain events and cases from raw records; Afiniti provides data and queries to support reconciliation of measurement.
Data security
Data remains behind the client’s firewalls, data sources are mutually agreed and fields can be encrypted/hashed as necessary.
Previous
Up Next