AI Belongs in Your BDC. Just Not Where the Vendor Says
What to automate first in the BDC, what must stay human, and the three-question test that tells you which is which.
Every BDC demo eventually arrives at the same slide. The AI takes the conversations, your people take the keys, and the rep delivers the line like a gift: imagine never staffing a phone again.
It is exactly backwards. The conversation is the one thing in your BDC a machine should not own. Everything wrapped around the conversation, the seconds before it and the minutes after it, is where AI actually earns its invoice. I have spent 20 years in this business and the last several building these systems myself, and I can tell you the vendors are pointing at the wrong end of the headset.
Here is the sorting that works on a real floor, plus the test for everything the list does not cover.
Automate first: the four workflows that pay
Start where the work is repetitive, the judgment is low, and a mistake is cheap to fix. In a BDC, that is four places.
After-hours response. Leads do not keep showroom hours. The 9 PM Sunday lead is often the most serious shopper of the week, and in most stores it waits until the morning huddle ends while a competitor answers in two minutes. A machine can acknowledge instantly, answer the easy questions about availability and hours, ask one qualifying question, and book the morning callback. Speed is the whole game in that window, and speed is what machines are for.
Lead routing. Most CRMs still deal leads like a poker hand: round robin, regardless of source, language, store, or who is actually at a desk. A routing layer that reads the lead and lands it with the right human in seconds is not glamorous. It pays anyway, every single day, because the lead finally meets the person most likely to win it.
Follow-up sequencing. Every store has a long-term cadence on paper and almost none execute it by hand, because day 11 of a 90-day cadence is exactly the kind of task humans drop. Let your people write the playbook and the messages. Let the machine own the calendar, because the machine never forgets day 11 and never decides a lead feels dead.
Call summaries. Your best agents hang up and start typing. Your worst agents hang up and start dialing. Automatic transcription and summarization into the CRM means every call leaves a record, every manager can scan the morning's conversations before the save-a-deal meeting, and coaching finally has raw material better than memory.
Leave it human: the conversations a machine should never own
The first live conversation. Trust in this business gets decided in the first thirty seconds of the first real exchange. A customer who realizes mid-conversation that they have been talking to software does not feel impressed, they feel handled, and handled customers buy elsewhere. Let the machine open the door at 9 PM. A person walks through it at 8 the next morning.
Negotiation. Numbers, trade values, payment structure. These are judgment calls with money and accountability attached, and accountability is the one thing you cannot delegate to a model. When a deal goes sideways, nobody in your store wants to explain that the software quoted it.
Escalation. An upset customer is not a ticket, it is a retention moment with a review attached. Nothing accelerates a one-star rating like a bot apologizing in a loop. The moment heat shows up, a human takes the conversation, with the machine's summary of how it got there already on the screen.
Notice what these three share. They are the moments where the customer decides how they feel about your store. Feelings are not a workflow, and no vendor roadmap changes that.
How to tell the difference
You do not need a vendor to sort this for you. Three questions handle almost any BDC task.
Does it repeat? If the task arrives in the same shape hundreds of times a month, it is a candidate. Routing repeats in shape and volume. The first conversation repeats in volume but never in shape, because every customer walks in carrying a different story.
Does it require judgment? Reading a payment situation, sensing that the trade-in story does not add up, calming someone who has been wronged: judgment. Logging a call, stamping a source code, sending day 11 of the cadence: process. Automate process. Staff judgment.
What does a mistake cost? A follow-up text that lands a day early is recoverable. A fumbled negotiation or a botched escalation can cost the deal, the review, and the referral behind it. Automate where errors are cheap and reversible. Keep humans where errors are expensive and permanent.
High repetition, low judgment, cheap mistakes: automate it. Anything else stays human until the machine proves otherwise on your data, not on the vendor's slide.
I have operated this lever
This is not theory I picked up from a whitepaper. From 2024 to 2026 I was VP of Marketing at Strolid, and the job sat directly on this lever: we lifted monthly opportunities from an average of 12 to 50+ rooftops per month, drove a 40 percent lift in MRR, and posted a record Q2 2025 doing it. While I was there I also built an internal meeting-intelligence pipeline in TypeScript and Python, the same unglamorous capture-summarize-route work I am telling you to automate. The pattern held every time I touched it: machines on the wrapper, humans on the conversation, and the numbers follow.
The core play: take back the night
Run this inside the next thirty days.
Pull the baseline first. Median response time and contact rate for every lead that arrived between 7 PM and 8 AM over the last full month. Most stores have never isolated that window, and the number is usually ugly enough to end the internal debate on its own.
Write the baseline on the board where the team can see it. The point of this play is not just the automation. It is teaching the store to demand a before-and-after for every tool that ever follows.
Then automate exactly that window and nothing else. Instant acknowledgment, answers to the easy questions, one qualifying question, a booked morning callback, every word logged to the CRM under the right source. Your humans keep the first live conversation. They just start it warm instead of cold, eleven hours ahead of the competition.
On day 30, compare contact rate and appointments set against the baseline. If it wins, you have a proven workflow and a template for the next one. If it loses, cancel, and you will still have spent one month learning more about your BDC than most stores learn in a year.
The conversation stays yours
The vendors are not wrong that AI belongs in your BDC. They are wrong about the seat. Buy the wrapper. Keep the voice.
If you want help sorting your store into the automate column and the human column, see how I work at /work or what an engagement costs at /pricing. Bring your after-hours numbers. We will both know where to start within an hour.
