By function
15%
measured productivity gain in customer support in structured studies. Gains shrink when the work is judgment.
Stanford HAI, AI Index 20261Customers do not wait on intelligence. They wait on someone to reconstruct their history and take the next action.
Service teams lose time searching tickets, policies, orders, and inboxes before they can answer. The agent should own that reconstruction and the routine completion.
How I see it
Customer service agents
A service agent is valuable when it can assemble the customer picture, apply a policy, take an allowed action, and write the result back to the ticket and the system of record. A bot that deflects without completing work creates repeat contacts.
The first design choice is which cases the agent may close. Password, status, scheduling, and policy-clear requests are different from credits, complaints, and relationship exceptions.
Human control points belong on refunds, goodwill, legal risk, and anything that changes a financial or contractual position beyond a threshold.
Contact-center theater is the failure mode: a new channel, no baseline, and a deflection metric that hides repeat work.
Common mistakes
What teams usually get wrong.
Deflection as the goal
Moving a customer off the phone into an unsolved chat is not a win.
No policy source
If the agent cannot see the current policy, it will invent one.
Closing tickets the system cannot stand behind
A fluent wrong answer becomes a second contact and a trust problem.
A useful diagnostic
Five questions before you fund the work.
What share of contacts are status, policy, or simple completion?
That share is the first agent candidate.How long does an agent spend gathering context before acting?
That time is often larger than the decision itself.Is there a system of record for the customer case?
If the truth is in email, start there before autonomy.Which actions require a supervisor today?
Those are the control points.Can you measure repeat contact and time to resolution?
Without those, deflection will be used as a vanity number.
Economic model
Service surface areacontacts × handle time × loaded cost × automatable share = annual service opportunity
Count repeat contacts. An agent that creates them is negative value.
Three credible paths
How far should you go?
Do not force one solution. Choose the path the economics, the risk, and the organization can support.
Improve the work you already have
Keep the process mostly intact and use AI on the bottlenecks that create delay, rework, or follow-up.
The workflow is already sound and a few steps create most of the friction.
Gains are usually incremental. The operating economics do not change much.
Redesign the workflow around AI
Question every handoff, queue, and duplicate step, then rebuild the process around what AI can now do.
The process grew over years and coordination now costs more than the work itself.
Requires process change, clearer ownership, and a willingness to retire old steps.
Put an agent on a high-value outcome
Give an agent responsibility for one valuable result across systems, with humans at the control points that require judgment, authority, or risk acceptance.
The workflow is high-value, variable, multi-step, and worth engineering for production.
Needs stronger architecture, evaluation, controls, and monitoring. This is not a prompt project.
When this is the wrong next step
Do not fund an agent here.
- The real problem is product quality or staffing, and an agent would only hide it.
- Policies are contradictory and nobody will own a source of truth.
- Leadership wants a bot live before write-back and evaluation exist.
A useful next step
Bring one workflow. Get guided into production.
We guide the implementation, go deep on the technical path, and stay hands-on through operations — or tell you when a simpler answer is better.
Discuss an AI opportunity