How to Automate Client Approvals Without Chasing People

To automate client approvals without chasing people, send the client one artifact with one clearly stated decision. Give it a real deadline that’s tied to what happens next, and let an escalation ladder you agreed in advance handle silence. In practice that’s one reminder before the deadline, one on the day, then a single escalation to a named person. Don’t keep sending more nudges. Let automation do the remembering, logging and drafting. Anything that changes scope, price or the relationship stays with your account lead.
Here’s how this usually plays out for agency owners. The homepage mockups went to the client nine days ago, and the dev sprint is booked for Monday.
Since then your account manager has sent “just following up” three times. The client replied once, in a different thread, asking if the blue could be “a bit warmer.” Nobody knows whether that’s an approval with a note or a request for another round. So the sprint slips, and the account manager spends Friday afternoon chasing.
This guide covers why approvals stall, the four-part approval packet we design around, when reminders and escalation should fire, a worked example, and the parts that should stay human.
Why do client approvals stall in the first place?
Usually the client doesn’t know exactly what they’re being asked to decide, or by when, or what happens if they don’t decide. The chasing is a symptom of that.
Look at a typical approval email. It links three files, says “let us know your thoughts,” and doesn’t include a date. “Thoughts” invites feedback instead of a decision. And with no date, it sits behind everything that has one.
The wider data points the same way. In March 2025, BetterBriefs, with the WFA and the IPA, published the BetterIdeas Project, a survey of more than 1,000 agency and in-house marketers across 54 countries. It found that it now takes five rounds of rework or extra creative development, on average, to get an idea approved, up from three in 2007. The words respondents used most often to describe their approval process were “painful, slow, subjective and inconsistent.”
That study is about creative sign-off. Content calendars, ad sets, timelines and change orders aren’t the same thing. Still, the logic carries over: the vaguer the ask, the more room there is for another round.
Coordination overhead is the other half of the problem. Asana’s Anatomy of Work Index (2021 edition) found that knowledge workers spent 60% of their day on “work about work,” meaning status updates, searching for information and chasing, rather than skilled work. That figure is a few years old, but chasing approvals is a textbook case of it.
What should an automated approval request contain?
When we design approval flows inside an AI operating system, we use a structure called the One-Ask Approval Packet. It’s a design pattern we use, and we don’t claim it guarantees any particular result.
We built it by working backward from the failure modes above. A vague ask, no date, no stated consequence and nobody to escalate to are four separate gaps, and any one of them can stall a decision. So the packet makes all four mandatory before anything gets sent.
- One artifact. A single link to the exact version being approved, with its version number. A folder or “the latest in Figma” doesn’t count.
- One decision. Phrase the question so it has a small set of answers: approve, approve with listed changes, or reject with reasons. “Thoughts?” isn’t allowed.
- One deadline, with a consequence. A date and time, plus what it starts or delays. For example: “Approve by Thursday 5pm so development starts Monday. If we don’t have a decision by then, launch moves by a week.”
- One escalation path. Agree it at kickoff so it isn’t a surprise. If the approver goes quiet, a named sponsor on the client side hears about it once.
If the system can’t fill all four fields, it doesn’t send the request. It sends it back to the account lead instead. That one rule is meant to stop vague “here are the files” emails before they reach the client.
Here’s the pre-send check as a checklist your team or your system can run.
- One link to the exact versioned artifact
- One decision with fixed answer options
- A deadline with date, time zone and what it starts
- A named approver and a named escalation contact
- A plain note of what is out of scope for this approval
When should reminders and escalation fire?
Each step should send fewer messages and go to someone more senior. Two reminders to the approver are enough. After that, the question moves one level up, once.
Here’s the default timeline we start from for a five-business-day approval window. Adjust the timing to fit your clients and contracts.
The rules behind the timeline matter more than the exact days.
| Rule | What it does | Why |
|---|---|---|
| Any reply stops the sequence | A person reads it and decides the next step | Automation shouldn’t talk over a real conversation |
| Max two reminders to the approver | No third “just checking in” | More nudges teach people to ignore you |
| Escalation goes up one level, once | The account lead writes to the sponsor, not the bot | A senior contact gets a response that volume won’t |
| Out-of-office pauses the clock | Reminders shift to the return date or a backup approver | Chasing someone on leave looks careless |
| Escalation is drafted, never auto-sent | The system writes it and a person approves it | It touches the relationship |
| Silence isn’t treated as approval by default | A missing decision shifts the timeline instead | See the legal note below |
How should the system read the client’s response?
Most of the mess comes after the reply. “Looks good, but can the blue be warmer?” is an approval with a change, and the system needs to log it that way.
We sort every response into one of five outcomes, and only a clean approval moves forward without a person looking at it.
Logging pays off later. Every approval is stored with the artifact version, the approver’s name, their exact words and a timestamp. Two months on, when someone asks “who signed off on this?”, you can answer in seconds.
What does a worked example look like?
This example is hypothetical. We wrote it to show the method, so it isn’t a client result, and the names are invented.
A 14-person web agency is redesigning a dental group’s website. Version 3 of the homepage mockups is ready on a Tuesday, and development is booked to start the following Monday. That leaves a three-business-day window, so the system compresses the default timeline.
The packet. The system pulls the v3 link from the project folder. From the kickoff record, it gets the approver (Priya, the group’s marketing manager) and the sponsor (Sam, the operations director). Then it drafts:
“Hi Priya, homepage mockup v3 is ready for sign-off: [link]. We need one decision: approve as is, approve with the changes you list, or send it back with reasons. Please reply by Friday 5pm ET so development can start Monday. If we don’t have a decision by then, launch moves back one week. Copy and photography are approved separately, so this is only about layout and design.”
The account lead reads it, changes one word, and sends it. That takes about two minutes.
Thursday morning. No reply yet, so Reminder 1 goes out automatically. It’s two lines with the same link and the same deadline.
Friday 9am. The deadline-day note goes out, saying what moves if there’s no decision by 5pm.
Friday 4pm. Priya replies: “Looks good. Can the blue be a bit warmer?” The sequence stops. The system classifies her reply as approval with small changes and drafts a confirmation: “Thanks, logging v3 as approved with one change: a warmer primary blue. We’ll show you the new shade in the first dev build. Shout if you meant anything bigger.” The account lead approves it and it goes out.
What got logged. Version v3, Priya as approver, the timestamp, her exact words and the one change. Development starts Monday as planned.
If Priya had stayed silent. There’s no third reminder. On Monday the system drafts one note from the account lead to Sam, and it’s framed around the schedule, without blaming anyone: “We’re holding the dev slot for the homepage but don’t have a decision yet. Could you help us get one by Wednesday, or should we move launch to the 24th?” A person reviews it and sends it.
What should stay human?
Automation is good at remembering dates, putting the packet together, sending reminders and keeping the record. It’s bad at judging intent and reading a relationship.
Keep these with a person:
- Anything that changes scope or price. “Can we also add a blog page?” is a change order, so it goes through your scope process.
- The escalation note. It goes to someone more senior on the client side, so tone matters.
- Ambiguous replies. If the system can’t tell which of the five outcomes applies, it asks the account lead.
- Sensitive accounts. If a client is in a dispute, up for renewal or wobbling, a person reviews every message.
A legal note, in plain language. In the US, the ESIGN Act says a signature or contract can’t be denied legal effect just because it’s electronic. That doesn’t make every “looks good” email a binding approval, and it doesn’t make silence count as consent. Whether an email approval or a “deemed approved after X days” clause holds up depends on your contract, the facts and the jurisdiction. If approvals involve real money or liability, have a lawyer review your approval and change-order wording. This isn’t legal advice.
How do you set this up in an agency?
You don’t need a new platform to start. Most agencies can run this with the project tool and email they already use, plus a shared approval log.
- Add approval rules to kickoff. Get a named approver and sponsor, and agree the reminder and escalation rules in writing.
- Build a packet template. Include the four required fields plus “what’s out of scope for this approval.”
- Set your reminder defaults. Two reminders, one escalation, and the sequence stops on any reply.
- Start a log. One row per approval: artifact, version, approver, decision, exact words and timestamp.
- Automate one part at a time. Start with the log, then the reminders, then drafting and classification. Review each step before the next one goes live.
Here’s plain-language wording you could adapt for a kickoff document. Have it checked against your contract before you use it:
“Each approval request will name one decision and a reply-by date. We’ll send up to two reminders. If we don’t have a decision by the date, we’ll contact [sponsor name] once, and affected dates will move by the length of the delay.”
To see whether it’s working, track two numbers for a month before and a month after: days from request to decision, and messages per approval. Our guide on measuring ROI on AI automation explains how to set that baseline honestly. Approval and QA steps often sit side by side, so the same setup carries over to automating delivery QA.
FAQ
Can I just use a proofing or approval tool?
Yes. For creative-heavy work, a proofing tool helps with versions and comments. But a tool won’t write a clear decision for you, and it won’t agree an escalation path with your client. Set the packet and the rules first, and let the tool carry them.
Is it OK to say “if we don’t hear back, we’ll treat it as approved”?
Be careful. Whether that holds depends on your contract and jurisdiction, and some clients will find it aggressive. The safer default is to say what moves in the schedule if there’s no decision. Get legal advice before you rely on deemed approval.
How many reminders is too many?
We start with two to the approver, then one escalation to the sponsor. Beyond that, more messages to the same person rarely help. Change the channel or the person instead.
Will clients find automated reminders impersonal?
Less so if each reminder restates the specific decision and deadline in a couple of lines and stops the moment they reply. Reminders feel robotic when they’re generic and keep going on a schedule regardless of what’s already happened.
Does this work for law, accounting or advisory practices?
The structure carries over to engagement letters, draft returns and document reviews. But regulated approvals often come with formal sign-off and record-keeping rules, so check your professional obligations before automating any step.
Who wrote this?
This guide is by Satya Phanindra Reddy, founder of Magic Teams AI. We install local-first, human-in-the-loop AI operating systems for founder-led agencies and small professional practices, usually in a one-week intensive.
The One-Ask Approval Packet comes from that design work, where we decide which steps need a machine’s memory and which need a person’s judgment. If you’re also curious how guides like this get found in AI search, see our explainer on answer engine optimization for B2B.
If approvals in your business still depend on someone’s inbox and memory, and you want a connected workflow your team can review, a fit call with Magic Teams is a good place to talk it through.