Tacnode™

Duplicate credits from concurrent AI agents

Three AI support agents each give one customer a goodwill credit, because each checks a synced copy of credits that hasn’t received the others yet.

Photo: Vitaly Gariev / Unsplash
→→At0:20
The story

Marcus’s order arrives damaged, so he opens a chat.

  • His $180 coffee grinder came cracked. He sends a photo to the support chat.
  • Each support channel is answered by its own AI agent: chat, email and voice.
  • The policy allows one goodwill credit per incident, up to $100.
The support console
What’s happening underneath
MarcusOrder #48213 · damagedChat agentAI agenttalking to MarcusEmail agentAI agentidleVoice agentAI agentidleissue_creditcheck_creditsBillingBilling system$0 issuedCredits copyPostgres$0 issuedSyncFivetran synccopies on a schedule
1 / 10

Illustrative scenario. Timings show the shape of the problem, not a measurement of any one company.

Your stack today, and with the Tacnode Context Lake™

Today

Billing
Billing system
Sync
Fivetran sync
Chat’s $80 not copied yet
Credits copy
Postgres
AI agents
Chat, email, voice

Most support stacks give each AI agent read access to a copy of billing and order data, refreshed by a sync. That was fine when one human handled one ticket and a second contact took hours. Agents changed the timing. A customer can open three channels in a minute, three agents can act in the same few seconds, and each one checks a copy that hasn’t received the others’ credits yet.

With the Tacnode Context Lake™

Billing
Billing system
Tacnode Context Lake™
Recorded, and saved for the next agent
AI agents
Chat, email, voice
  • The credit was recorded before it was sentEach agent records its credit against the customer’s cap, then sends it to billing.
  • The agents saw each otherThe cap is held in one place, so two agents can’t both take what’s left of it.
  • Each action was saved for the next agentThe chat agent’s credit was what the email and voice agents read.
  • Every decision can be auditedWhich agent decided what, against which cap and when, and the state as it stood at that moment.

Questions to ask your team

  1. 1Can more than one AI agent act for the same customer at the same time, on different channels?
  2. 2When an agent issues a credit, refund or discount, how long until another agent can see it?
  3. 3How often does reconciliation find goodwill above policy for a single incident?
  4. 4After the fact, can you see what each agent read when it decided?