Three AI marketing bots, each with one job: a build guide
The most common AI automation mistake is buying one general assistant and giving it five jobs. The bot blends them. Your buyer gets a support answer written in a sales tone. You get a price the bot inferred from a page it should never have touched. And because the bot touches four different numbers, none of them tells you whether it’s actually working.
The fix is a bench: three bots, each with one job, one set of inputs, and one measurement number. Here is how to build it.
Why one bot per job
A scoped bot handles one repeatable task end to end and passes everything else to a person. It reads from a short list of tools and writes to an even shorter one. You can test it, measure it, and fix it because there is only one thing it was ever supposed to do.
A general assistant fails the measurement test. When a bot touches four different outcomes, you can’t isolate which one is broken. You end up guessing instead of fixing.
️ Bot one: the speed-to-lead responder
This bot owns the first sixty seconds after someone raises a hand. That window matters more than most teams admit. Replying in seconds beats replying well two hours later.
What it does
It greets the new lead, asks three to five short qualifying questions, and books a meeting on the right calendar. It stops at the meeting. Pricing, scoping, and anything bespoke belongs to your team.
What it needs from you
- Your form and chat events, delivered the moment they fire, not on a nightly sync.
- Your qualifying questions, written out in the order a human would ask them.
- Calendar access for the reps who take first meetings, with their real availability loaded.
- Your routing rules, so the right region and segment reach the right person.
How you know it works
Read four numbers weekly: median time to first reply, the share of leads that answer the bot, the share that book a meeting, and the share a rep had to rescue. Compare each against your own numbers from the month before you turned it on. That is the only baseline that means anything for your team.
Before you build anything, pull your actual median reply time to a form fill. Measure it, don’t assume it.
Bot two: the quote and negotiation bot
This bot owns the back and forth on price, but only inside limits you set in advance.
What it does
It answers questions about your published pricing, applies your standard discounts, and assembles a quote the buyer can read. It escalates anything below your price floor, anything with unusual terms, and anything involving a named enterprise account.
What it needs from you
- Your current price book, with every bundle and term your reps can actually sell.
- A hard floor per product, set in the tool itself so a clever prompt cannot move it.
- Your discount ladder, tied to volume, term length, and whatever else you already reward.
- The escalation list, naming who signs off each kind of exception and their backup.
Write those limits down once and keep them where a reviewer can find them. A bot that closes more deals by giving more away has moved your margin, not your performance.
How you know it works
Track the share of quotes that close, your average discount, time from request to quote, and how often a person had to take over. Watch the discount number most closely. That is the signal that drifts first.
Bot three: the marketplace handoff bot
This bot owns the moment a buyer moves from your site into somebody else’s checkout.
What it does
It passes the product details, the current stock position, and the delivery promise into the marketplace flow and keeps them current. Those three details break more orders than anything else on that path. A buyer who sees a different delivery date at checkout abandons it.
What it needs from you
- A live stock feed. An hourly file turns into oversells on your busiest day.
- Clean product data, matched to each marketplace’s own fields and rules.
- Your delivery logic, so the date the buyer sees on your site matches the date they see at checkout.
- Order status fed back, so your service team can answer questions about orders placed elsewhere.
How you know it works
Read the share of handoffs that complete, abandonment rate at that step, mismatches between your data and the marketplace’s data, and revenue by marketplace. Investigate every mismatch. Each one is a buyer who got two different answers from the same brand.
Which bot to build first
Let your own numbers pick. Slow replies to inbound leads point at the responder. Quotes taking days and discounts drifting point at the quote bot. Marketplace orders failing at the last step point at the handoff bot.
Build one, run it for a quarter, then start the second. Which leak costs you most this month?
✅ The pre-launch checklist
Before any bot talks to a customer, write down the following on one page per bot:
- One job, written in a sentence your whole team would agree with.
- One number, with your current figure recorded and dated.
- Named inputs, each one live and each one owned by a person.
- A handoff rule, stating exactly when the bot stops and who picks it up.
- A kill switch, with the person who flips it named in the same document.
Then run it in shadow mode for two weeks. The bot drafts, a person sends, and you read the gap between what it proposed and what actually went out. That gap tells you more than any benchmark.
Pro tip: watch the intervention rate
Across all three bots, the number to watch over time is how often a person has to step in. A rising intervention rate means the bot has drifted out of its lane. That is the earliest signal to catch before it becomes a customer problem.
Also keep a register of every live bot with its job, its inputs, and its owner. Read it before you add the next one. Two bots messaging the same person with overlapping scopes is a support ticket waiting to happen.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
