![](data:image/svg+xml,<svg display="block" role="presentation" viewBox="0 0 18 18" xmlns="http://www.w3.org/2000/svg"><path d="M 0.148 1.06 C 0.053 0.966 0 0.838 0 0.705 C 0 0.572 0.053 0.444 0.148 0.35 L 0.348 0.15 C 0.442 0.052 0.572 -0.002 0.708 0 L 7.868 0 C 8.003 -0.002 8.134 0.052 8.228 0.15 L 8.428 0.35 C 8.523 0.444 8.576 0.572 8.576 0.705 C 8.576 0.838 8.523 0.966 8.428 1.06 L 4.638 4.85 C 4.547 4.947 4.42 5.002 4.288 5.002 C 4.155 5.002 4.029 4.947 3.938 4.85 Z" fill="rgb(17, 17, 17)" height="5.00155387474404px" id="T1L7vK7HI" transform="translate(4.712 7)" width="8.575800000000001px"/></svg>)
From 50% to 95%: what is driving acceptance rates up
Article
Leon Jungfleisch
CEO & Co-Founder
A product update on multi-agent decisions, Gap Mining, Custom Flows, automatic quality checks, and the next improvements coming to vennie.
Quality you can see in operation
The last few weeks have been noticeable: across all accounts, we are seeing major jumps in acceptance rate—one account, for example, went from 50% to 95%.
That is not a coincidence. It is the result of continuously improving quality through Gap Mining, where we actively review every account and close gaps in vennie’s knowledge, and through new features that make vennie fundamentally smarter.
Here is an overview of what we have released recently—and what is coming in the next few days.
What is live now
vennie makes better decisions—at every step of the process
Anyone who has worked with Gorgias knows the problem: the AI agent slightly misunderstands a customer’s request, picks the wrong next action, and everything that follows builds on that wrong decision. The result is a response that is technically correct but misses the actual point.
We have rebuilt this process from the ground up. vennie now uses a multi-agent approach in which several specialised agents jointly decide what the customer actually wants and what the right next step is. They challenge and verify each other before any decision is made.
Dynamic Memory also feeds into every decision and continuously improves through ongoing processing. The more vennie handles, the better these decisions become.
Autonomous Learning—now also manually trainable
Autonomous Learning is trained in two ways: through the Edit button you already know, and through the Gap Mining & Automation Mining we actively run for you. We regularly check where vennie can still improve and add that information directly.
Custom Flow Node for product information
Flows can now be enriched with detailed product information. You can tell vennie exactly how certain products should be described and which guidelines the description should follow.
Wait Node for Custom Flows
vennie can now pause within a flow and then continue automatically. For example, a customer asks about tracking. vennie checks the status and finds that the package has not shipped yet. Instead of giving an imprecise answer, vennie waits 24 hours, checks again, and only then sends the customer a confirmation with the tracking number. No manual intervention and no follow-up needed.
New Xentral Nodes in Custom Flows
For everyone using Xentral: you can now pull Xentral data directly into your Custom Flows.
What is coming in the next few days
Autonomous Learning—now everywhere
So far, Autonomous Learning has primarily fed into response generation. That was the first step. Now we are embedding it across all relevant parts: information retrieval, tone of voice, product information, processes, and the initial decision of whether a ticket should be handled at all.
↳ Going forward, suggestions that were not accepted will also feed into training, with the same threshold of two occurrences before a memory item becomes active. That closes the last remaining gap in the training process.
Automatic quality check before every response
The more instructions and Custom Flows an account has, the more places there are where a response can be slightly off: the tone of voice may not be quite right, a specific guideline may be overlooked, or the conversation context may not be fully considered.
We have built a dedicated quality-checking agent that reviews every response before it goes out. If something is not right, vennie adjusts the response directly or restarts the process. The result is fewer outliers and more consistent quality—even for complex accounts.
More robust order-search process
vennie is getting a multi-step approach to order searches. If an initial search returns no result, vennie searches again with better use of the conversation context. This makes the entire process significantly more reliable.
Automatic spam handling
vennie already detects spam today. What was missing was a complete process around it. We are building that now so spam tickets can be resolved and closed automatically without anyone having to review them manually.
