FlowsBook a callSign inCreate free accountSign up
Business and places data

When an AI agent for competitor monitoring beats a fixed alert, and when it does not

A fixed flow sends the same alert every week; an AI agent answers the question you have today. How to choose, combine both, and keep the agent's answers honest.

Part of the guide Location intelligence automation

By the xMap team · · 5 min read

An AI agent for competitor monitoring is worth having when the questions change: "what opened near our Baixa hotel this month?", "which rival near the airport has the most reviews?", "is the new place on our street a chain?". It looks things up live with its tools and answers in a chat. When the question is the same every week (tell me every new rival within 1 km), a fixed flow is cheaper, faster and more predictable. Most teams end up with both: a fixed alert for the routine signal and an agent for the follow-up questions it raises.

This article is for managers deciding how to watch competitors without hiring an analyst. It compares the two approaches, shows how they fit together, walks through a made-up hotel group in Lisbon, lists the ways an agent can mislead, and explains how to build one that answers your team on Slack or WhatsApp.

Fixed flow or agent: how to choose

Fixed flowAI agent
Best forThe same question on a timetableNew questions as they come up
How it startsA schedule or a change near your sitesSomeone writes to it
What you getThe same list format every timeAn answer in words, shaped by the question
PredictabilityHigh: same steps each runLower: it chooses which tools to call
Cost per answerLow and steadyHigher: thinking steps plus each lookup
Where it shinesNever missing an opening or closureExplaining one, comparing, digging in

How a scheduled flow and an AI agent differ for watching competitors.

A useful rule: if you can write the question once and it will still be the right question in six months, make it a fixed flow. If the question starts with "why", "which" or "compare", give it to the agent.

How the two fit together

The fixed flow is the early warning. It runs every morning or every week and reports new rivals, closures or ratings that moved. The agent is what the manager talks to after reading the alert. "A new boutique hotel opened 300 m from us" leads naturally to "how is it rated compared with us, and is it part of a group?". The manager writes that in Slack; the agent looks the place up, compares it with the hotels around, and answers in a few lines.

Because the agent can remember each person's earlier messages, a manager does not have to restate which hotel is theirs every time. And if your other AI employees already collect findings across the company, a weekly brief built from them answers the broad question ("what moved in our markets this week?") while the agent handles the narrow ones.

A worked example, made up

A made-up group runs five hotels in Lisbon. Its revenue manager receives a fixed weekly alert of new hotels and guest houses within 1 km of each property. One Monday the alert lists two new places near the hotel in Alfama. She asks the agent in the team's Slack channel:

  • "Of the two new places near Alfama, which is closer to us and how are they rated?" The agent looks both up, gives the distance of each from the hotel and their ratings with the number of reviews behind them, and notes that one has too few reviews to judge yet.
  • "How many hotels are within 500 m of Alfama now, and how many belong to a chain?" It counts them live and answers with the split.
  • "Same question for our Belem hotel." It remembers the format and answers for Belem.

Three questions, three answers in a few minutes, each figure from a lookup recorded in the run. Without the agent, each would have been a small research task for someone. The hotels, places and answers here are invented to show the exchange.

Ways an agent can mislead, and how to prevent them

  • Answering from memory. If its tools fail or it is not given one, an agent may fall back on what it learned in training. Tell it in its instructions to answer only from its lookups and to say when it could not look something up.
  • Mixing up places with similar names. Two hotels can share a word in their names. Ask it to give the address or distance with each place so the reader can spot a wrong match.
  • Inventing a reason. "It opened because the area is booming" is a guess, not a finding. A Check the work step with a rule such as "no claims that are not in the data" catches this before the answer goes out.
  • Running wide searches on every question. Give the agent a sensible default distance and a cap on tool calls per message, so a casual question does not turn into a city-wide search.

Building it in xMap Flows

The agent that answers your team:

  1. Chat, or an app event such as a new Slack message in your competitor channel.
  2. AI agent: instructions naming your properties and their addresses, the kinds of rival you care about, a default distance, and to answer only from its tools. Memory per conversation, so it remembers each person.
  3. Tools linked to it: Nearby places for rivals around a point, with their ratings, reviews and whether they belong to a chain; Map business scan when you want a place's live status.
  4. Check the work: rules such as every place named has a distance, and no figures that are not in the data.
  5. Reply: posts the answer back where the question was asked.

The fixed alert beside it is a separate, short flow: Schedule, your properties from My data, Nearby places, Only new and an email. The template New hotels opening near mine is that alert ready made, Weekly competitive intelligence for my cafes is a weekly summary for a cafe group, and A weekly brief from your AI employees rolls up what all of them found. The guide to location intelligence automation shows how these pieces sit with the rest.

Questions

Can an AI agent watch competitors on its own without being asked?

Yes, with a schedule in front of it and a fixed brief, but for a routine watch a plain flow is usually simpler and more predictable. Agents are best for questions that vary.

Which channels can the agent answer on?

A chat page, the editor's chat, and messages from apps such as Slack, WhatsApp, Telegram and Teams. Reply answers on the same channel the question came from.

Will the agent remember which locations are ours?

Put your locations in its instructions so every answer uses them. With memory per conversation it also remembers what each person asked earlier.

How do I know where an answer came from?

Every run records the tools the agent called and what each returned, so you can open a run and see the lookups behind each figure.

Templates that do this

Read next

More on business and places data · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.