Some specific numbers from the agentic-systems work we've done inside Marketing Signals over the past year.
Total hours saved per month, across the operation: a little over four hundred. Net cost of running the systems: low four figures monthly. Net annualised value: comfortably into seven figures.
The composition of where the hours came back is more interesting than the headline number, and it's the part I want to walk through, because the framing applies to most agencies and services businesses thinking about how to extract similar value.
The four workflows we automated first
We didn't automate everything. We picked four workflows where we had high confidence the agent version would work, where the human time saved was meaningful, and where the cost of a failure was bounded.
Audit pipelines. The largest hour saver, by some distance. Our internal audit work — for new business, for client onboarding, for the amivisible.co product — used to consume meaningful senior time. We rebuilt the entire audit production process as an agentic pipeline. Senior strategists now scope the audit and review the output. Everything in between runs without human time. Roughly 220 hours a month saved here.
Weekly client reporting. Most agencies have a weekly or fortnightly reporting cadence per client. The work of pulling data from various platforms, reconciling it, producing the report, writing the commentary, and preparing it for client review was, across our portfolio, taking 80-100 hours of analyst time per month. The agentic version pulls the data, produces the report, drafts the commentary, runs a QA pass, and surfaces only the cases that need senior judgment. We're now spending 20 hours a month on this work, mostly on the commentary that requires real client context. Net saving: 60-80 hours.
Pitch and proposal generation. New business proposals — the structured documents we send to prospective clients — used to involve a senior strategist drafting the strategic recommendations and a junior person assembling the document. We rebuilt this so that the agent produces the first draft of both, the senior strategist reviews and edits the strategic content, and the junior person is no longer involved in the assembly. Saves roughly 60 hours a month and produces more consistent proposals.
QA on outbound work. Before any externally-visible deliverable goes out, we now run an agentic QA pass that checks for consistency, factual accuracy of references, broken links, and a list of common error categories. This catches problems that previously got caught (less reliably) by the senior reviewer at end of project. The senior reviewer's time is now spent on judgment items, not error catching. Saves roughly 40 hours a month, plus an unmeasurable but clearly real reduction in errors that would otherwise have shipped.
Total: 380-420 hours, depending on the month.
The cost-benefit calculation
The cost side of this is small. The four pipelines together consume roughly £1,200 a month in API spend — mostly for the audit pipeline, which is far and away the heaviest user. Engineering time to maintain them: about 10 hours a month, mostly on improvements rather than break-fixes.
The benefit side is larger and harder to quantify cleanly. Senior agency time, fully loaded, is expensive. Even at conservative rates, 400 hours a month of senior time is well into seven figures annualised.
The point of the comparison isn't that we're saving a specific number of pounds. It's that the ratio is heavily in favour of the agentic systems. We're spending pence to save pounds. Even if our numbers are wrong by a factor of two, the leverage is dramatic.
The unexpected places we found leverage
Two areas where the leverage was bigger than I expected, and worth flagging because most agencies probably have these as well.
Internal documentation maintenance. The agency has hundreds of process documents — onboarding docs, client SOPs, internal training materials. Keeping them up to date used to be no one's specific job, which meant it was effectively no one's job. We have an agent now that periodically scans the documentation against the actual current process (inferred from recent project artefacts and decisions) and flags drift. Then a separate agent proposes updates. A senior person approves them.
This isn't time saving in the headline sense — we weren't really maintaining the docs before, so there's no work being replaced. But it's compounding value, because the docs are now actually accurate, which means new hires onboard faster, junior staff make fewer mistakes, and senior staff spend less time answering process questions. The hours-saved number on this is probably 30-40 a month, but the second-order effects are worth more.
Sales cadence outbound. This is technically more of a workflow assist than a full agent — the senior new business team still owns the relationship, the messaging, and the close. But the back-end work of researching prospects, drafting personalised outreach sequences, and handling the follow-up cadence is largely now agent-driven. The team that used to spend half their week on outbound mechanics now spends a couple of hours and an order of magnitude more time in actual customer conversations.
Both of these would have been difficult to predict from the outset. They emerged because, having built the audit and reporting pipelines, we got better at spotting opportunities. The capability creates the next opportunity.
The workflows we tried to automate and gave up on
In the spirit of being honest about what didn't work — three workflows we tried to build agents for and abandoned.
Strategic recommendations. We tried to have agents produce strategic recommendations alongside the audit data. They could, technically, do it. The recommendations were always plausible and often defensible. But they weren't differentiated. They didn't reflect the senior strategist's actual judgment about the client's specific situation. Customers responded to the difference. We pulled the agent off this work and the senior strategist now writes the recommendations into the agent-produced audit. The hybrid works. The pure agent version didn't.
Client communication on contentious topics. Anything where the email had to manage emotion, conflict, or strategic delicacy stayed with humans. Agents could draft technically correct responses. They couldn't manage relationship dynamics. The mistakes when we tried were small but consistent — the agent's version was always 5-10% too direct, or 5-10% too apologetic, in ways that humans wouldn't have produced.
Creative ideation in the early stages of campaigns. Same issue, different domain. Agents could produce a reasonable list of campaign ideas. They couldn't produce the specific kind of unreasonable, surprising, or just-right idea that distinguishes a good creative team. We use agents to do the structured pre-work (with the production discipline that makes the systems trustworthy) — research, audience profiles, prior-campaign benchmarking — and let humans do the creative.
The pattern across these three: workflows where the value was in the judgment, the relationship management, or the creative leap, were not improved by agents. The pure-agent versions produced output that was technically sufficient but commercially weaker. The hybrid versions, where agents handle the structured work and humans handle the judgment, work much better.
What this would look like at smaller agencies
A note for agency leaders reading this who don't have an in-house engineering team. The 400-hour number is partly a function of our scale. Not all of you will get the same magnitude of saving. But the proportion is similar. If you have ten people, you might recover 40 hours a month — a full working week of senior time, every month, ongoing.
The four workflow categories I described above are universal. Most agencies have audit pipelines. All agencies have reporting workflows. Most have proposal processes. All have QA needs.
You don't need to build everything yourselves. The off-the-shelf tooling for some of this — particularly reporting and proposal automation — is increasingly capable. The audit and QA work is more bespoke and probably needs at least some custom engineering.
The barrier for most small agencies isn't the technology. It's the operational discipline to redesign workflows around agents rather than to bolt agents onto existing workflows. That redesign is where the leverage actually is.
What's next for our internal stack
The four workflows we've automated cover roughly 60% of the senior time we've been targeting. The next 30% is harder to automate cleanly — it's the more strategic, judgment-heavy work — but there's still leverage to be found there in agent-assisted rather than agent-owned setups.
The remaining 10% is genuinely human work. Relationship management. Strategic positioning. Senior creative. We're not trying to automate those. We're trying to have the senior team spend more of their time on those things, with less of it being eaten by the workflows we have automated.
That's, ultimately, the point of the work. Not "AI replaces humans". The senior team in the agency is doing more of the work that only senior humans can do, because the workflow tax that used to consume their week is now running in the background.
If you're an agency leader thinking about how to find the equivalent leverage in your operation, the practical first step is to map a single week of your senior team's time and ask which of the workflows they touched were genuinely judgment-bound and which were workflow-bound. The workflow-bound ones are the candidates. Most of them are reachable with current technology if the operational redesign work is done.
It's worth doing. The senior team will thank you for it.