Skip to content

Compared with a chatbot connected to your ad accounts

A general AI assistant connected to your ad and email platforms can do a lot, quickly. But it mostly acts when asked, it holds your account access, and its limits are instructions in a prompt.

What it does well

An AI assistant on your ad and email platforms is useful, and improves every few months. Ask well and you get reports, ad drafts and setting changes faster. But a general assistant isn’t made to run your budget and customers’ inboxes alone, day after day.

  • Answering questions about your accounts.
  • One-off tasks: pulling a report, drafting an ad, changing a setting.
  • Knowing its own tool well, when it’s built into that tool.
  • Getting better quickly as models improve.

Where it stops

  1. 1It acts when someone asks. What it remembers between sessions can’t be enforced.
  2. 2Its limits are instructions in a prompt, which it follows most of the time.
  3. 3It holds your account access and makes changes itself.
  4. 4An assistant built into one tool sees only that tool.
  5. 5The chat is the only record of why a change was made.

Side by side, with cost and time.

A chatbot on your accounts compared with Heraldic, row by row
QuestionA chatbot on your accountsHeraldic
Who sets the limitsA promptYou, and the software holds them
Holds your credentialsYes, while it’s connectedAgents never do. Only the executor, the part that makes changes, holds them
Stops at your limitsIf the prompt holdsChecked by software before every change, and again on approval
What it optimizesWhatever you asked forYour conversion events, each valued by what it’s worth to you
Across channelsOne request at a timeOne set of rules, one contact policy, one record
Video and large-scale creativeWhat its tools can draftStill images from your approved library; video you upload sits beside them
Explains itselfThe chat logA record of every action, with its reason
What it costsA subscription to the assistant, plus your time to ask and check.Not set yet. Ad spend goes straight to the ad platforms
Time to a first resultImmediate for a one-off task. Ongoing work mostly waits for someone to ask.Set by your funnel: at a store with daily orders, a test can be called in two weeks; a lead-generation funnel takes longer.

When to choose it instead

If you want a faster way to do the work yourself, one task at a time. For work that runs every day without you, you need standing briefs, hard rules and a record.

Using both

They can work side by side: ask an assistant questions about your accounts, and let agents do the ongoing work within your limits.

Questions about this comparison

Isn’t this just a chatbot on a schedule?

No. Agents propose changes, and a policy engine with no AI in it decides. When you approve a change, every rule runs again before anything happens. Agents never hold your account credentials.

Which AI model does it use?

The model for each step is a setting, chosen by testing models from several providers. Your rules, history and library stay the same whichever model is in use.

Can’t I give an assistant standing instructions?

You can, and a good one will mostly follow them. A standing instruction is still text the model reads, so it holds only as long as the model follows it. Here, limits are settings checked on every proposal, and again when you approve.

Get a team that works to your rules.

Join the beta, or try the demo first.