The Deflection Business Case That Didn't Deflect
Coastal Community Bank handles 2.1 million service contacts a year; 1.2 million are in scope for a GenAI customer assistant. The approved business case assumed 30% containment of in-scope contacts, a $6.40 average cost per contact, and $2.3 million in annual savings against $1.4 million implementation and $420,000 annual run cost. An eight-week pilot handled 180,000 sessions. Blended containment was 11%, but it ranged from 26% on password resets to 4% on disputed transactions. Customers who were contained were slightly less satisfied (CSAT 4.1 vs 4.4), and 18% of 'contained' sessions were followed by a call within 48 hours. The vendor says the pilot was 'not fully tuned.' The CFO wants the original savings number in next month's budget. The contact center VP wants to expand to all intents. You own the value case.
1. Recompute the annual benefit using pilot evidence and state the assumptions. Show the arithmetic and a range.
Read the worked answer
Start from the pilot's observed behavior, not the vendor's tuned aspiration. The original case: 1.2M in-scope contacts × 30% containment × $6.40 = $2.304M.
The pilot blended containment was 11%, but 18% of contained sessions were followed by a call within 48 hours, so effective containment is 11% × 82% = 9.0%, or 108,000 contacts. At $6.40 that is about $691k gross annual benefit. Against $1.4M implementation and $420k annual run, year one is roughly -$1.13M and steady-state net is about $271k, so payback is beyond four years unless performance improves.
A defensible range: low case $0.6M, base $0.7M, high case $1.1M only if containment reaches 20% across about 70% of in-scope volume (1.2M × 70% × 20% = 168,000 contacts, about $1.08M), which is a much broader scope than the pilot. The five highest-containment intents are about 45% of volume; at 20% containment that is 108,000 contacts, roughly the base case.
That is the key insight: the pilot does not support $2.3M, and routing the top intents alone does not close the gap; the original case required a containment rate and scope the evidence does not show. Also treat the CSAT drop (4.1 vs 4.4) as a real cost, not a footnote, and the 18% callback rate as leakage. Present the CFO a probability-weighted number and a range, not a point estimate.
2. Recommend a stage-gate decision and what must be true to expand. Address the CFO's request explicitly.
Read the worked answer
Recommend a 90-day stage gate, not full expansion and not cancellation. Expand only to the five highest-containment intents, keep the rest with human agents, and set explicit exit criteria: blended containment at or above 18%, 48-hour callback at or below 10%, CSAT within 0.1 of baseline, cost per contained contact below $2.50, and no rise in escalations or complaints.
During the gate, fix the tuning and the measurement: define 'contained' consistently, measure callbacks at 7 and 30 days as well as 48 hours, and instrument containment by intent and customer segment. The CFO request to book $2.3M should be declined; book $0 in the base budget and carry a $0.5M–$0.7M opportunity with finance-approved assumptions, or book only the pilot-proven $0.69M gross less run cost if the program continues.
The contact-center VP's 'all intents' proposal is the wrong sequencing: disputed transactions at 4% containment will destroy trust and create rework. Use the gate to prove the operating model-agent handoff, escalation, and QA-before adding volume. If the gate fails, stop; the $1.4M is largely sunk, but the $420k run cost is avoidable. If it passes, scale in two waves, with a second gate at 20% of traffic. Make the gate criteria public to finance and operations so the decision is not relitigated in each budget cycle.
3. Define benefits-realization metrics, baselines, and ownership.
Read the worked answer
Benefits realization needs a baseline, a formula, and an owner. Baseline: annual in-scope contacts by intent, average cost per contact, current CSAT, current callback rate, and agent handle time. Formula: value = (verified contained contacts × avoided cost per contact) − run cost − amortized build − rework cost. 'Verified' means no human contact within 30 days and no repeat contact for the same issue; count the 48-hour callback as a failure.
Track monthly: containment by intent, callback rate, CSAT, escalation rate, cost per contained contact, and adoption. Ownership: the contact-center VP owns the business outcome; finance validates the avoided-cost rate and signs off on realized benefits; the AI product owner owns model and assistant performance. Do not let the vendor report containment; use the telephony and CRM systems of record. Report a rolling three-month average to avoid celebrating a good week.
Include dis-benefits: CSAT decline, agent deskilling, and increased complexity for edge cases. Finally, define what would cause you to stop: if verified containment stays below 12% for two quarters, the program is not paying back and should be retired. A business case that cannot state its stop condition is not a business case; it is a budget request.