Notes from week 12 of our first Dash-Pilot rooftop
Twelve weeks of running Dash-Pilot on a real F&I desk. What we measured, what surprised us, and what we are still trying to understand as the pilot continues.
Why we are writing this now
We picked one dealer partner. One rooftop, one F&I desk, one DMS. Twelve weeks in, the patterns are starting to harden into something worth writing down. This post is honest about what is working, what is not, and what we are still trying to understand. None of this is a marketing claim. It is a field note.
If you are evaluating Dash-Pilot, or thinking about what an AI F&I Co-Pilot would actually do at your store, this is the closest we can get to the truth without taking you on site.
What we set out to measure
When the pilot started, we wrote down four things we would track every week. We did not change the list.
- Time-to-first-lender-callback on a new deal.
- Number of deal jackets needing rework after submit.
- Stipulation kicks per hundred deals.
- F&I manager subjective load, captured Fridays by short survey.
We chose these because they are concrete, the F&I director already cared about them, and they were measurable without instrumentation we could not ship in week one. We deliberately avoided “hours saved” as a top-line metric. That number is too easy to make look good and too hard to believe.
What is working
The biggest surprise is also the most boring one. Dash-Pilot does not change what an F&I manager does. It changes what they spend time on.
Pre-flight on a new credit application now happens in seconds. OFAC and Red Flags clear, identity confidence lands above threshold, the lender call sheet gets ranked. The F&I manager opens the deal and reviews work that is already done, instead of doing it from scratch. That sounds like a small thing. In practice it pulls a handful of minutes out of every deal on average, and over a busy Saturday those minutes compound.
The second thing that is working is the audit trail. Every recommendation Dash-Pilot makes is replayable. Model version, inputs, prompt, output. We did not realize how much this would matter to the dealer principal until week three, when an OFAC flag came back from a lender and the F&I director was able to show, in twenty seconds, exactly what Dash-Pilot saw and recommended. That conversation went very differently than it would have a quarter ago.
What is not working yet
Two things are honestly not where we want them.
First, lender API uptime is the constraint, not Dash-Pilot. Some non-trivial share of the time, the slowest lender in the routing list is the one we want to call first, and we are waiting on them. We have a fallback path that ranks the next lender, which the team uses well, but the experience is choppier than we would like. We are not going to fix lender uptime. We can compress the time we spend waiting for it, and that is the next quarter of work.
Second, the DMS write-back is more brittle than we expected. The DMS UI changes on a cadence we are still calibrating to. Our drift detector catches most of it within minutes, but “most” is the wrong word here. We are building toward “every,” and we are not there yet.
What we are watching
Three things we do not yet have enough data on, but we are tracking.
- Sales team sentiment. The F&I director loves it. The sales team is mostly indifferent so far, which is the goal at this stage. If sales-team sentiment turns negative, we have a problem.
- Customer experience at the desk. The customer should not be able to tell Dash-Pilot is in the loop. So far they cannot. We will revisit this with structured customer interviews in the next month.
- Long-tail compliance. We have not had an audit yet during the pilot window. We are designing the deal-jacket retention with auditors in mind, but we will only really know how that plays the first time it gets used.
What this is not
A few things this post is not, and we are saying so on purpose.
It is not a case study. The pilot is twelve weeks old. We do not have enough deals through the system to publish hard numbers we would defend in front of a CFO. We will, in time.
It is not a sales pitch. Dash-Pilot is not for every dealer today. If you have a high-functioning F&I desk on a modern DMS, you are not our first ten customers. We are looking for groups where the F&I close-cycle is the bottleneck and the appetite for an agent in the loop is real.
It is not the architecture. The interesting parts of how this works are in the read layer, the canonical model, and the write-back agent. Those are worth their own post, and we will write it.
What you can do with this
If you read this far and it sounds like your store, the next step is a thirty-minute call. We will not ask for a deal pull on the first conversation. We will ask what your F&I director’s Friday looks like and listen for the parts of it that an agent could compress.
If you are skeptical, good. The dealer principals we have learned the most from were skeptical first. The contact form is on the contact page, or email us directly.
We will write the next field note at week twenty-four.