Most operators who have an agent connected to Cabgo's MCP use its capabilities in one direction: reports. Shift status, inactive drivers, cash balances, rating metrics. That covers the part of the work where the operator needs information, but leaves untouched another dimension that's equally important: the part where the operator needs to act on demand. Creating first-trip coupons, activating driver bonuses for a weekend with an event, reactivating passengers who haven't ridden in three weeks. Those tasks remain manual, done when the operator remembers to do them, designed from the dashboard without the operational context the agent has available in real time. The thesis of this article is that chat is a better design surface for a campaign than a dashboard form — not because the agent is faster, but because the conversational process forces the operator to articulate the problem before creating the solution.
This article is for the operator who already has the agent running with Cabgo's skill installed and wants to extend its use beyond monitoring. We cover what marketing tools the MCP server has available, three campaign patterns that work in 50 to 150-driver operations, how to build the prompt that produces the right design on the first response, and what to review before confirming execution. That last point matters: creating a coupon passengers see in the app is a Tier-3 action — it leaves the system toward the outside and there is no undo button once it's active. The agent proposes, you confirm, the server executes. The flow doesn't change, only the nature of what you're creating does.
Why operational marketing stays the most postponed task
Creating a coupon in Cabgo's dashboard takes three minutes. The question is why most operators only do it when a passenger complains or when a competitor runs a visible promotion in the city. The answer isn't lack of time or lack of tools — it's lack of context at the right moment. To decide whether a reactivation coupon makes sense this week, the operator needs to know how many passengers have gone more than 14 days without a trip, whether driver availability can handle a 20% demand increase, and whether there's a relevant local event or date that could amplify the reach. That information exists in the MCP server, but no dashboard screen combines it in the «create promotion» view. The result is that the decision gets pushed back until an urgent reason appears, and urgent campaigns are rarely well-calibrated because they're designed without that context.
The agent resolves that disconnect because it has access to all that data in the same conversational context where you create the campaign. Instead of opening three different views to get the context you need, you describe the problem to the agent and it fetches the data, proposes the parameters, and waits for your confirmation before executing. Design and execution happen in the same conversation — which means every campaign you create via chat has more deliberation behind it than ones created directly in the dashboard form without that prior step.
What the agent can do in operational marketing
Before designing the first marketing prompt, it's worth being clear about what MCP server tools are available for these tasks. The marketing actions the agent can execute in Cabgo cover the full cycle from creation to post-campaign analysis:
- Create discount coupons: fixed amount or percentage, with total usage and per-user limits, expiration date, and optional restriction by geographic zone or service type
- Create driver bonuses: by number of trips completed in a period, by weekly average rating, or by availability during high-demand time windows
- Query past campaign history: coupons created, redemption rate, estimated cost to the operation, and the segment that used them most
- Check the status of an active campaign: users who have it available, redemptions made so far, and discount budget consumed
- Deactivate a campaign before its expiration date if the cost exceeds the target or if operational conditions change
Three campaign patterns that work in regional operations
The campaign types with the most impact in 50 to 150-driver operations repeat with variations across almost all of LATAM's regional market. The first is **dormant passenger reactivation**: any user who completed at least one trip but hasn't requested one in the last 21 days. This segment responds better to a fixed-amount coupon than a percentage one — a «50-peso discount on your next trip» is clearer than «20% off», especially in markets where average prices vary little across zones. Before creating the coupon, the agent can check how many users fall into that segment this week and whether the available driver volume can handle the expected demand. An operation with 200 active passengers and 40 dormant has a 20% inactivity rate worth addressing; the same rate in an operation with 800 actives indicates the segment is manageable without urgency.
The second pattern is the **weekend driver bonus**: an incentive for drivers who complete more than X trips between Friday at 7pm and Sunday at 11:59pm. This pattern is harder to calibrate without data because the right value of X depends on recent weekend history. A bonus with too low a threshold is reached by almost every driver without any real change in availability; one set too high nobody reaches and supply doesn't move where it matters. The agent can query the last four weekends of history and return the 70th-percentile trip count per driver — that number is the right threshold for a bonus that rewards the top 30% of fleet performers without giving the incentive to everyone.
The third pattern is the **new zone coupon**: a discount for passengers who request a trip with origin or destination in a neighborhood the operation recently activated, where volume still hasn't built momentum. This type of coupon requires a geographic restriction that in the dashboard means manually configuring the zone's coordinates. In chat, the operator types the neighborhood name and the agent resolves the coordinates from the server's zone catalog, proposes the coverage radius based on the area, and has everything ready to confirm without the operator needing to open an external map.
The prompt that produces the right design on the first response
A well-structured marketing prompt has four parts: the operational context the agent needs to calibrate the campaign, the demand problem you're solving, the business constraints, and the format of the proposal you expect. The fourth part is what most operators leave out. Without it, the agent proposes reasonable but generic parameters that need several rounds of adjustment. With it, the first response includes exact parameters ready to review and confirm.
- **Operational context**: include the active fleet size, average trips per driver per day, and availability percentage during peak hours — without this, the agent estimates those values and may miscalibrate
- **Demand problem**: describe the specific pattern you want to change — passenger inactivity, low availability on Friday nights, low penetration in a new neighborhood — not the type of campaign you want to create
- **Business constraints**: maximum discount budget, maximum percentage of the fleet that can reach the bonus threshold, campaign start date and expiration date
- **Proposal format**: ask the agent to return the exact parameters as a list — type, amount, usage limit, expiration, geographic restriction — and the estimated cost if 50% of eligible users redeem the campaign
- **Explicit prior verification**: add «before proposing parameters, query the last 4 weeks of history to calibrate the values» — this prevents the agent from generating numbers without grounding them in the real operation
What the campaign history reveals that the dashboard doesn't show
Creating the campaign is half the work. The other half is understanding what worked and calibrating the next one. The MCP server's campaign history has more detail than what appears in the dashboard view: not just the total redemptions, but how many distinct users used it, in what zone and at what time it was most active, and whether the drivers who covered that demand were in the productive or low-activity segment of the fleet. The agent can query that history and deliver it in a format the operator can compare week over week — the same kind of structured output built with audit prompts, extended to include the campaign segment.
An operator who reviews campaign history with the agent once a week starts seeing patterns that change how they design the next ones. The weekend bonus that worked with an 18-trip threshold in March doesn't necessarily produce the same result in June if driver availability has shifted. A 50-peso reactivation coupon that achieved a 35% redemption rate in the north zone may see less than 10% in the south zone if average trip revenue there is different. That data exists in the history and the agent pulls it out with the same kind of prompt you use for weekly audits — extended to include the campaign block. The review takes no more than 5 additional minutes and closes the loop between creating, measuring, and adjusting.
What to review in the confirmation card before executing
When the agent presents the confirmation card to activate a campaign, there are three fields worth reading before approving: the impact radius (how many users will see the campaign active in their app), the estimated maximum cost if all eligible users redeem it, and the expiration date. The first two protect the budget; the third protects against coupons that stay active weeks after the campaign ended and no one remembers they're there. A coupon expiring June 30th that the team forgets to review can keep generating discounts in July — not because the agent fails, but because nobody deactivated it. Adding a review task to the original prompt («at the end, add: VERIFY ON [date] that the campaign is still active or was deactivated») turns the confirmation card into a closing reminder in addition to an approval request.
The first campaign I built with the agent I did without a saved prompt — I just described the problem: too many passengers without trips in the week before the local festival. It asked how many days I considered «without a trip», queried the last three weeks of history, and proposed a 40-peso coupon with a 200-use limit expiring in five days. I spent less time reviewing the confirmation card than it would have taken to open the dashboard form. It wasn't that the agent was faster — it was that I had already thought through the problem before reaching the form.
The agent's role in operational marketing isn't to replace the operator's intuition about their market — it's to give that intuition calibrated data at the moment the decision is made, not 20 minutes after the operator opened the dashboard, gathered numbers separately, and returned to the «create promotion» screen. Campaigns designed in that conversation have more deliberation behind them than ones created from the form without that step, not because the agent is smarter than the operator, but because the conversational format requires the problem to be articulated before the form appears.
If you've been using the agent only for reports and audits, extending its use to marketing requires no additional configuration — just a different kind of prompt. The next time you spot a demand drop or an important weekend event coming up in the city, don't open the dashboard first: open the chat, describe the pattern you're seeing, and ask the agent to propose campaign parameters based on the last four weeks of history. The first time you review that confirmation card with already-calibrated parameters and an estimated cost included, the resistance to using the agent for marketing disappears.
