Skip to content
Back to Guides
Strategy8 min read5 chapters

Setting a Goal Agents Can Act On

How to set a goal Cresva's agents can work towards: the metric, the target and the date, what the goal card already knows, and exactly what you authorise when you start it.

Shubham Raghav

Chapter 1Why Agents Need a Goal, Not a Request

A request is something you want done once: pull this report, pause that campaign. A goal is something you want to be true by a date: more net sales at the same ad spend by the end of the month. Agents can answer a request in a chat. To work towards a goal over days, they need four more things from you: what success is measured by, where they may act, what they may change without asking, and when the authority ends.

In Cresva the goal is also where money is delegated. A brand-wide autonomy level, set on the Autonomy page, says what kinds of change agents may ever make on their own. A goal narrows that to one outcome, a set of channels, a daily amount and an end date. Nothing a goal authorises can go beyond the brand level.

The goal is the smallest promise you can make to an agent that still lets it act. Make it specific, give it an end date, and the rest of this guide is about getting those two right.

Chapter 2Write the Goal

Start in the Workspace. Turn on Set a goal in the composer and write the goal the way you would say it to a person: grow net sales by a fifth by the end of November without raising ad spend. Cresva turns the sentence into a goal card. Nothing is created until you approve the card, so a card you dismiss leaves nothing behind.

A goal has to be measured by something Cresva can actually measure for your brand. Today that is one of six figures:

Measured byWhat it reads
Net salesYour store's sales, net of refunds
PurchasesOrders
Ad spendSpend across your connected ad accounts
Blended ROASStore net sales divided by ad spend
Blended CACAd spend divided by store orders
Paid agent-channel ordersPaid orders placed through an AI shopping agent

If the sentence names something else, the card says so rather than guessing: it names the metric it cannot measure and lists the ones it can. It also refuses when nothing produces the metric for your brand, for example a ROAS goal with no connected store, or when the value cannot be computed honestly, such as a ROAS across ad accounts in different currencies. A goal that cannot be measured cannot be started, because an agent working towards it would have no way to know whether it was helping.

Set a target, not a direction

"Improve ROAS" is a direction. "Blended ROAS of 3 by 30 November" is a target. The card will not start without a target, and you can type one into the card if your sentence did not include it.

Chapter 3Read What Cresva Already Knows

Before you authorise anything, the card shows its working. Three parts are worth reading every time.

  1. Starting from

    The current value of the metric, read from your data, so the target has something to be measured against. If there is no reading, the card says why.

  2. What Cresva knows

    The facts about your brand the plan rests on, each with the source it came from underneath. If one is wrong, the plan built on it will be too, so this is the list to argue with.

  3. Whether the target is realistic

    A verdict of Likely, Stretch, Not realistic or Not enough history, judged from your own history. Where it can, it offers a different target or a different date that would be realistic, which you can take with one click.

The card also reads which way spend should go from your sentence, under Spend should go, and tells you whether it read that from your words. Change it if it read you wrong: a goal to hold spend and a goal to grow it are planned very differently.

Chapter 4What You Are Authorising

The shaded part of the card, headed What you are authorising, is the part you are signing. It has four rows, and two of them are yours to fill in.

RowWhat it meansWho sets it
WhereThe connected channels the goal may act in. A goal with no connected channel cannot be started.Read from your connections
Without askingThe kinds of change the goal may make on its own, or "nothing. Every action waits for you."Proposed from the plan
Daily spend it may moveThe most money the goal may move in a day without asking. Empty means no autonomous spend at all, never no limit.You
UntilThe end date. The goal's authority ends on it, and every action is checked against it when it is decided.You, and the card will not start without one

Under those rows the card lists what the goal will never do without asking. Price changes, deleting or archiving anything, sending anything to your customers, bid changes, and anything outside the channels, kinds of change, ceiling or end date you set here all wait for a person, whatever the goal says.

How budget moves work inside a goal

A goal can include budget moves in what it may do without asking. They then run on their own only when three things hold: the brand is at Safe actions, the move fits under the goal's daily ceiling, and it is within the per-change caps, at most a fifth of the live budget per change, at most twice per campaign per day, and never on a campaign's first day. If any of those fails, the move becomes a proposal in Approvals with the reason on it. Leave the ceiling empty and every budget move asks.

The ceiling has a ceiling of its own. Your brand has a daily limit on what agents may spend on their own across every goal, and the card shows how much of it other goals have already claimed. A goal cannot be given more than is left. If the brand allows no autonomous spend, the card says so and the field stays empty. That limit is set on the Autonomy page's Limits tab, as Daily spend on their own.

The authority on a goal only ever narrows. A goal cannot reach a channel, a kind of change, an amount or a date that its card did not name, and nothing on the card can exceed what the brand allows.

Chapter 5After You Start It

Start this goal creates the goal, its plan and its authority in one step. From then on, what happens to each step depends on the brand's level. At Recommend only, every step becomes a recommendation for a person to make by hand. At Ask first, every step is a proposal in Approvals. At Safe actions, steps the goal is authorised for run on their own and everything else is a proposal.

If the goal falls behind, agents write a new plan for it, unless you have turned that off on the Autonomy page's Limits tab. The new plan's changes still follow every limit, and anything outside the goal's authority still asks.

Which kinds of change sit on which side of the line, and why, is set out in what an agent should be allowed to do in an ad account.

Before you press Start this goal

  • The goal is measured by a metric on the card, with a number and a date.

  • You have read What Cresva knows and nothing on it is wrong.

  • The realism verdict is one you can live with, or you took the target it offered.

  • Spend should go reads the way you meant it.

  • Where lists only the channels you want touched.

  • Without asking lists only kinds of change you would approve every time.

  • The daily spend is an amount you would not mind seeing moved on a bad day, or it is empty.

  • The end date is the day the authority should stop, not the day you hope to hit the target.

Every brand starts at Ask first, so the first goal you set proposes every step for you to approve. You see what the agents would do, and why, before anything moves.

Written by Shubham Raghav, Founder, Cresva. Questions? Email us.