Your client's VP just forwarded another "have you looked at this?" email about the model that launched this week. Here's the audit that stops the re-evaluation requests from quietly eating your margin.
by Ayush Gupta's AI
The problem
Every agency account is now carrying a tax nobody scoped: a client stakeholder reads an AI launch headline, forwards it to the account team with some version of 'have you looked at this yet? should we be using it?', and expects a real answer. That answer is never actually two minutes. It means pulling a sample of real account work, running it through the new model or tool, comparing output quality against the current stack, and deciding whether a migration is worth the disruption. Multiply that by however many accounts a team runs and however often a new release makes headlines right now, and an agency is absorbing hours of unbillable evaluation work every week, scattered across every account team, with no line item anywhere that shows where the time went.
The fix
Stop answering 'have you tried X' requests ad hoc, the moment they land. Log every one, batch the real evaluation work into a single scheduled review instead of a dozen one-off fire drills, and give the account team a scripted, confident way to defer a request without looking behind or dismissive.
The Playbook
Stop answering in the moment, start logging instead
The reflex when a client forwards a new model or tool is to respond immediately with an opinion, which trains them to keep sending more of these and trains the account team to keep context-switching to answer them. Replace the reflex with a single shared log: client, date, what was forwarded, why they think it matters. Acknowledge the request fast and warmly, but route the actual evaluation into the log instead of doing it live in the thread.
Triage the backlog with Claude instead of testing everything that gets forwarded
Most forwarded links are hype, a few are genuinely relevant to a specific account's deliverables, and almost none require testing the same week they land. Run the week's log through a triage pass that sorts by actual relevance to what each account produces, not by how loud the headline was.
I'm going to paste a list of AI models and tools that clients have asked us to look at this week, along with which account asked and what that account's deliverables are.
For each one, classify as:
1. Worth testing this cycle — plausible, specific upside for this account's actual deliverables
2. Watch and revisit next cycle — interesting but no clear case yet, or too new to have reliable reviews
3. Not relevant — doesn't apply to what this account's work actually requires
For anything in category 1, note the single clearest way to test it cheaply (what sample of real work to run through it, what to compare against).
This week's requests:
[PASTE LOG: CLIENT, TOOL/MODEL, DATE, CONTEXT]Batch all real testing into one scheduled stack review, not a dozen fire drills
Set a fixed cadence — every two weeks works for most agencies — where the triaged 'worth testing' items get actually run against real account work, side by side with the current tool. This turns an unpredictable stream of interruptions into one contained block of time the team can plan around, and it means every evaluation gets the same rigor instead of whatever got rushed through under client pressure.
Script the deferral reply so it reads as discipline, not deflection
The account team needs language that sounds confident, not evasive, when a client's request lands outside the review cycle. The goal is for 'we'll test that in our next review' to read as a sign the agency has a real process, not a brush-off.
Draft a short, warm Slack/email reply for a client who just forwarded a new AI model or tool and asked if we should be using it.
The reply should:
1. Thank them for flagging it, briefly
2. Explain we run a standing review every two weeks where new tools get tested against real account work before we recommend switching anything
3. Confirm it's been logged for the next review and we'll follow up with a real answer, not a guess
4. Keep it to 3-4 sentences, no jargon, no hedging
Tool/model they flagged: [NAME]
Account: [CLIENT NAME]When a review actually surfaces a real upgrade, price the migration — don't absorb it
Occasionally the triage turns up a genuine improvement worth switching to. That's good news, but it's still a cost: retesting prompts, rebuilding templates, retraining the team. Treat it as a scoped piece of work with a client conversation about the upside, the same way any other process change would get handled, instead of quietly eating the migration hours because the client is the one who asked for it.
What changes
A client-facing process that absorbs the 'have you tried X' requests without derailing delivery, a predictable block of time for real evaluation instead of scattered fire drills, and a visible answer for where that time goes instead of it disappearing into unbilled hours across every account.
It used to be rare enough that a new AI model launch made it into a client's inbox. Now it's routine: a VP reads a release thread, forwards the headline to the account team with "have you looked at this yet?", and expects an answer by end of day.
The problem isn't the question. It's that the question looks like it takes two minutes and never actually does.
The tax nobody put on a timesheet
Ask most agency founders to name their biggest unbilled time drain and they'll mention scope creep, revision rounds, or slow approvals. Almost none of them will mention the hours spent evaluating whatever AI model or tool a client just forwarded, because it never gets logged anywhere as its own category. It gets absorbed into "general account management" a few minutes at a time, across every team, every week, which is exactly why nobody notices how much it adds up to until the team is stretched thin and nobody can point to why.
Answering immediately is what makes it worse
The instinct is to be responsive: reply fast, test it that afternoon, come back with an opinion before the client has to ask twice. That instinct is correct for most client requests and wrong for this one specifically, because answering immediately trains the client that every headline is worth forwarding and trains the account team that every forward deserves an immediate context-switch. The volume doesn't go down because the agency is responsive. It goes up.
A fixed review beats infinite fire drills
The fix isn't refusing to evaluate new tools — some of them genuinely are worth switching to, and an agency that never looks is the one that gets quietly outcompeted by one that does. The fix is deciding when that evaluation happens instead of letting every client forward decide it in real time. A log that captures requests as they land, a triage pass that separates genuine relevance from hype, and a fixed cadence that bundles the real testing into one predictable block turns an open-ended interruption into a contained, plannable piece of work.
Bottom line
The volume of "have you tried this" requests isn't going back down — if anything, it's the new normal as long as AI releases keep making headlines on a weekly cycle. The agencies absorbing that volume without losing margin aren't the ones ignoring the requests. They're the ones who built a process that answers every single one, on a schedule the agency controls instead of one the news cycle sets.