Last reviewed October 6, 2026
AI saves your team hours. Then it takes a lot of them back. Here is what 11 public sources show about where delivery time goes. Add your numbers to build the first benchmark made only for agency delivery.
That gap is the work around the work. Re-explaining context. Checking output. Fixing it. Switching tools. The data below shows each piece. Our benchmark measures the same pieces inside agencies.
Glean Work AI Index 2026 · 6,000 digital workers · source
Grouped by the 5 bottlenecks the survey asks about. Not AGL data.
of agencies say each account manager handles fewer than 10 clients.
Databox account manager survey · 48 agencies · 2023, updated Jul 2025 · source
of agencies put 15 or more clients on 1 account manager.
Databox account manager survey · 48 agencies · 2023, updated Jul 2025 · source
accounts is typical for an account manager who does not also do the delivery work.
Sakas & Company · Agency consultant view · Article · source
of managers who use AI load context by hand before it is useful.
In Parallel manager survey · 247 managers, 5 countries · Reported Aug 2026 · source
spend 3 to 10+ minutes per session re-explaining what the team already knows.
In Parallel manager survey · 247 managers, 5 countries · Reported Aug 2026 · source
say their AI assistant usually knows enough.
In Parallel manager survey · 247 managers, 5 countries · Reported Aug 2026 · source
a week go to feeding AI context.
Glean Work AI Index 2026 · 6,000 digital workers, US, UK, AU · Dec 2025 to Jan 2026 · source
say the information they need is not reachable from their AI tools.
Glean Work AI Index 2026 · 6,000 digital workers, US, UK, AU · Dec 2025 to Jan 2026 · source
hours saved by AI are lost to fixing its output.
Workday, Beyond Productivity · 3,200 employees and leaders · Nov 2025 · source
of agencies added more human review of AI-generated work.
AgencyAnalytics 2026 Marketing Agency Benchmarks · 494 agency professionals · Feb to Apr 2026 · source
of US desk workers got low-quality AI work from a colleague last month. Each case took 1 hour 56 minutes to resolve.
BetterUp Labs and Stanford Social Media Lab · 1,150 US desk workers · Sep 2025 · source
of agencies save 5 or more hours a week with AI.
AgencyAnalytics 2026 Marketing Agency Benchmarks · 494 agency professionals · Feb to Apr 2026 · source
average revision rounds before creative approval, 2007 versus today.
IPA, WFA and BetterBriefs, BetterIdeas Project · 1,034 respondents, 54 countries · As reported by Marketing Week · source
of agencies say the approval process works well.
IPA, WFA and BetterBriefs, BetterIdeas Project · 1,034 respondents, 54 countries · As reported by Marketing Week · source
of marketing and creative pros say approval takes more effort than the creative work.
StreamWork approval survey · 500 US marketing and creative pros · Oct 2025 · source
of agencies say clients expect faster turnaround than a year ago.
AgencyAnalytics 2026 Marketing Agency Benchmarks · 494 agency professionals · Feb to Apr 2026 · source
of AI users run 4 or more AI tools every week.
Glean Work AI Index 2026 · 6,000 digital workers, US, UK, AU · Dec 2025 to Jan 2026 · source
rerun the same prompt across tools when the first answer falls short.
Glean Work AI Index 2026 · 6,000 digital workers, US, UK, AU · Dec 2025 to Jan 2026 · source
app switches a day. Just under 4 hours a week spent reorienting.
Harvard Business Review, app toggling study · 137 users, 3 Fortune 500 firms · Aug 2022 · source
of digital agencies have implemented AI across the business. 28% are implementing it now.
Promethean Research, 2026 State of Digital Services · 119 agency leaders · Feb 2026 · source
Example numbers loaded. Change them to yours. Then add them to the benchmark.
Under 10. About 70% of agencies sit here (Databox).
576 hours a year at 48 weeks. Glean puts context feeding at 2.3 hrs per person each week.
4 or more. Glean: 33% of AI users run this many.
Small sample. Self-reported. Reporting-software vendor.
Expert opinion, not a survey.
Reporting-software vendor. Self-reported.
Independent agency research firm. Full report is paid.
All digital workers, not agencies. AI vendor.
Managers in general, not agencies. Vendor.
Firms with $100M+ revenue. Vendor.
Self-reported. Vendor co-authored.
Measured behavior. Small sample. Not agencies.
Creative approvals. Skews to larger brands.
Approval-software vendor. Self-reported.
Agency owners, COOs and delivery leads at agencies with a delivery team.
Answers stay anonymous. We combine them, and no single agency is shown. All numbers are self-reported.
The report by email when it is published.
No.
After 50 qualified responses.
Agile Growth Labs.
No. It is run by AGL and self-reported.
11 published studies and articles. Each number names its source, its sample and its date, and links to the original. None of them is AGL data.
So you can compare your agency today. Our own results publish after 50 qualified responses.
No. They will sit next to it, so you can see agency numbers beside the wider market.
