// Resources
How Much Time Does Campaign Automation Save Compared to Doing It Manually?
What the randomised studies actually measured, why the hours-saved claims in the comparison articles disagree with each other, where a campaign's hours really sit, and how to measure the saving in your own business before you buy anything.
By Victor Xu, Founder · LinkedIn
Last updated · 2026-10-20
The short version
On the writing and drafting part of campaign work, the saving is large and it has been measured properly: in a randomised experiment published in Science in July 2023, 453 college-educated professionals given ChatGPT finished short writing tasks with average time down 40 percent and output quality up 18 percent. On the rest of a campaign, the assembling, the approvals, the list work and the reporting, almost nobody has measured anything, and that is where most of the hours in a real campaign sit. So the useful answer is not a number from somebody else's business. It is a count of your own: which steps in your campaign take a person, how long each one takes, and which of them still need a person once the drafting is fast. Teams do report large savings. In HubSpot's 2026 State of Marketing survey of more than 1,500 global marketers, about a third of marketing teams said AI saves them 10 to 14 hours a week and another third said more than 15. That is recollection rather than measurement, and it is still the clearest signal in the category, as long as you treat it as a reason to measure rather than a number to budget against.
How much time does campaign automation save compared to doing it manually?
Expect the biggest measured gains on drafting and the smallest on everything that involves other people, which is why the honest answer depends on the shape of your campaign rather than on an industry average. A campaign that is mostly copy and creative has more automatable hours in it than one that is mostly waiting for a stakeholder to approve something.
Two randomised studies carry the weight here. Shakked Noy and Whitney Zhang's experiment, published in Science on 14 July 2023, assigned occupation-specific incentivised writing tasks to 453 college-educated professionals and gave half of them ChatGPT. Average time taken fell 40 percent and output quality rose 18 percent as judged by blinded professionals in the same occupations. Marketers were one of the six occupations in the sample, alongside grant writers, consultants, data analysts, HR professionals and managers, and the tasks were press releases, short reports, analysis plans and delicate emails.
Read the authors' own limitation before you apply that number to your week. The tasks, they write, "are relatively short, self-contained, and lack a dimension of context-specific knowledge, which may inflate our estimates of ChatGPT's usefulness". A campaign is not short, self-contained or free of context.
The second study is the field experiment with 758 consultants at Boston Consulting Group, reported by Dell'Acqua and colleagues in 2023 and summarised in ITIF's June 2024 review of AI productivity evidence. Inside the range of tasks the model handled well, consultants using GPT-4 completed 12 percent more tasks, finished them 25 percent faster, and produced work rated 40 percent higher in quality.
Put those together and the shape of the answer appears. The drafting step gets much faster, and the measured figure for it is 40 percent on short professional writing. The thinking step does not move. And the coordination steps, approvals, sign-off, the wait for a product detail, do not move at all until you change who has to be involved.
For the broader version of this question across a whole business, how much time and money AI automation saves sets out the same reasoning across every process rather than campaign work alone.
What do the vendor numbers actually measure?
Vendor numbers measure a clean task on a clean brief, which is why they rarely survive contact with your calendar. The clearest evidence of that is how badly they disagree with one another. Search this question and the hours-saved claims on the first page run from about five hours a week, to 6.5 hours a week, to 13 hours a week, to roughly two and a half hours a day, with a separate figure of 4.2 hours a week for email specialists. Several are attributed to a named report without a link to it, and at least one per-campaign figure that circulates widely has no findable primary source at all.
What those numbers have in common is a missing baseline. A saving is a difference between two measurements, and almost none of these say what the manual version took, who did it, or whether the output was equivalent. A figure in hours per campaign is unusable without knowing whether that campaign was one email or a quarter's worth of paid media.
HubSpot's 2026 State of Marketing report is the most useful of the survey evidence, because its base is stated: more than 1,500 global marketers. In its own words, about one-third of teams say AI saves them 10 to 14 hours per week, another third say more than 15 hours, and just under a third say one to nine. Read as a measurement it is weak, because it is a recollection rather than a count. Read as a signal it is strong, because a large majority of teams saying something similar is still information.
The pattern to take from all of it: self-reported savings run high, measured savings on narrow tasks run high too, and measured savings across a whole process are mostly unpublished. Treat any single figure as a hypothesis about your own business, then test it.
Where do the hours in a campaign actually go?
Most campaign hours go on assembly and coordination rather than on writing, which is the single most useful thing to know before buying anything. The table below splits a typical campaign into its steps and marks what the evidence supports.
| Campaign step | What it involves | What automation changes | What the evidence says |
|---|---|---|---|
| Deciding the offer and the audience | Judgement, pricing, margin, who to talk to | Little. The model does not know your margins or your customers | No measured gain. Noy and Zhang's tasks deliberately lacked company context |
| Drafting copy and variants | Emails, ads, landing page, subject lines | Large. Drafting and variant generation is the measured sweet spot | Average time down 40 percent on short professional writing, Science 2023 |
| Building and loading the campaign | Lists, segments, templates, tracking, QA | Large, and almost unmeasured in public research | No published study found. Count your own hours here |
| Approvals and sign-off | Waiting on people | Nothing, until you change the process | No evidence, and no vendor claims it either |
| Running and monitoring | Checking delivery, spend, errors | Moderate. Alerts beat checking a dashboard | No measured study found |
| Reporting | Pulling numbers from several systems into one view | Large, repeatable and the easiest to justify | No published study found, and the easiest to measure yourself |
| Follow-up on responses | Replying, routing, booking | Large, and this is the one that affects revenue | Lead response research covers the value rather than the hours |
The money is in the reporting row and the follow-up row, and neither is what campaign tools are sold on. Reporting is pure assembly: the same numbers from the same places in the same shape every week. Follow-up decides whether the campaign produced anything at all, and it is work that arrives on your customers' schedule rather than yours.
So the campaign step with the most measured evidence behind it, drafting, is often not the step costing you the most hours. Count before you buy.
How do we measure the saving in our own business?
Measure it by timing the manual version first, because without a baseline you will be arguing about impressions in six months. The method is plain and takes one campaign cycle.
Write down every step in your current campaign, in order, with the name of whoever does it. Then, for one full cycle, have each person record the actual minutes rather than their estimate: start time, end time, what interrupted them. Estimates of your own work run low on the parts you dislike and high on the parts you notice, which is why the recording matters.
Next, mark each step with what would have to be true for software to do it. A step needs clean inputs, a defined output, and a rule for what happens when something is unclear. Steps that fail that test are not automation candidates yet, whatever a demo shows.
Then automate one step, keep both versions running for a cycle, and compare the same numbers: minutes per step, errors caught, and the output a reviewer would accept without changes. The third one is the number people forget. A step that takes 20 minutes instead of 90 but needs a full rewrite afterwards has not saved anything.
How to measure AI automation savings sets out the general version of this with the arithmetic for annualising it. The campaign-specific warning is about seasonality: measure a normal cycle rather than the launch week, or your baseline will flatter the result.
One of our own numbers for scale. At Universidad Maimonides, the research unit we built for got 15 hours a week back, measured against the manual process it replaced, and the system has been running on our infrastructure ever since.
Why do our savings look smaller than the case studies?
Savings look smaller because a case study reports the step and your calendar reports the process, and the gap between those two is real work. Three specific reasons come up in every campaign review we do.
The first is review time. A draft produced in two minutes still needs somebody who knows the brand to read it, and if your standards are high that review can cost more than the drafting did. The BCG experiment is the measured version of why review cannot be skipped. On a task built to fall just outside what the model handles well, MIT Sloan's summary of the study reports that consultants given GPT-4 scored 13 percentage points worse on correctness than the control group, and those given GPT-4 plus a guide to using it scored 24 points worse. The lead author's explanation is that people switch off and follow the recommendation.
The second is the coordination that did not change. If three people have to approve a campaign and one of them travels, the campaign still takes a week. Automation moves the work and not the waiting, and most teams discover their real cycle time was never about production.
The third is the setup cost nobody counts. Templates, brand rules, approved claims, suppression lists: the first automated campaign carries the cost of defining all of it, and the saving only appears from the second one onwards. Judge the investment across a quarter rather than a campaign.
The savings that compound come from the repeatable assembly: the weekly report, the list hygiene, the follow-up sequence that runs whether or not anybody is in the office. That work happens on a cadence set by your customers, which is exactly the work a person should not be doing by hand. Where AI creates leverage in a business covers how to find those processes across the whole operation rather than in marketing alone.
Frequently asked questions
What does it cost to automate our campaign work?
Scope drives it rather than any list price: how many systems the work has to touch, how much variation is in your inputs, whether your lists and records are clean, and how much of the judgement stays with a person. A single reporting assembly that pulls from three systems is a contained piece of work. A full campaign pipeline with segmentation rules, approval routing, suppression logic and response handling is a larger one. The number you want in hand first is your own: the hours the manual process takes now, multiplied by a loaded hourly cost, annualised. That figure is what any build has to beat, and it is the one most teams have never worked out.
How does campaign automation actually work?
Something external triggers it, rather than a person opening a tool. A date, a form submission, a tag in your CRM, a report that is due. The system then gathers what it needs from the places that hold it, drafts or assembles the output against rules you set, writes the result where the work lives, and records what it did. A person reviews what the rules flagged as uncertain. The engineering is mostly in the connections and in the error handling rather than in the writing, which is why the hard part of a campaign build is rarely the part the demo shows.
Is campaign automation better than hiring a marketing coordinator?
Those two buy different things, so compare them on coverage. A coordinator handles judgement, relationships, the thing nobody specified and the week everything changes. Automation covers the work that repeats on a schedule and has to happen whether or not somebody is at a desk. Most teams that get this right end up with both, with the person doing the deciding and the system doing the assembling. Count which of your current hours are judgement and which are assembly, and the answer usually becomes obvious.
Can our email campaigns be automated end to end?
The production can, the decisions should not be. Triggering, segmenting, assembling, sending, measuring and following up are all mechanical once the rules exist. What stays with a person is the offer, the claims you are willing to make, and anything going to a customer for the first time. The design that works has the system drafting and a person approving until the pattern is proven, then narrows the approval to exceptions. Anything with compliance exposure, pricing or regulated claims in it stays reviewed.
How long before we can tell whether it worked?
Two campaign cycles, because the first one carries the setup. In the first cycle the numbers look worse than the manual process, because somebody is defining templates and rules that did not exist. From the second cycle onwards you can compare minutes per step against the baseline you recorded. Watch three numbers rather than one: time per step, the share of output a reviewer accepts unchanged, and the share of responses that got a reply within the hour. The last one is usually where the revenue moved.
Find out which of your campaign work is worth automating
We count where the hours in your campaigns go, put a number on each step, and tell you which ones are worth building and which should stay with a person. Then our engineering team builds what the roadmap calls for and we run it for you from there. We own the outcome, not the deliverable.