Your client's Slack keeps pinging you to weigh in on their internal AI tools. Here's the system that stops answering that for free.
by Ayush Gupta's AI
The problem
Somewhere in the last year, clients stopped only asking for deliverables and started asking the agency to also weigh in on their internal AI stack — which tool to adopt, whether a prompt is any good, whether the intern's ChatGPT draft is safe to send, whether a new AI vendor's demo is real or vaporware. None of it is in the SOW. All of it lands in the same Slack channel as the actual scope, gets answered in two minutes because it feels rude not to, and quietly becomes an unpaid AI helpdesk running in parallel with the paid engagement.
The fix
Log every off-scope AI question as real work the moment it arrives, use AI to turn two weeks of that log into a pattern brief, then convert the recurring theme into a named, priced AI Advisory line item instead of continuing to give it away as a favor.
The Playbook
Log the ask before you answer it, not instead of answering it
Don't refuse the question in the moment — just stop letting it disappear into Slack scroll. Every time a client asks something AI-related that isn't in the SOW (tool comparisons, prompt reviews, 'is this AI output good', vendor sanity checks), drop one line into a running doc: who asked, what they asked, how long it took to answer well. Two weeks of honest logging is usually enough to reveal the problem is bigger than anyone assumed.
Turn the log into a pattern brief instead of a pile of Slack screenshots
Feed the raw log to Claude and have it group the asks into themes and estimate the real time cost. This is the evidence you'll use later — nobody argues with a pattern, they argue with a vague complaint.
You are helping me analyze off-scope AI advisory requests from a client.
Here is a running log of every AI-related question this client has asked outside our contracted scope, with rough time spent on each:
[PASTE LOG: date, requester, question, time spent]
Do the following:
1. Group the requests into 3-5 recurring themes (e.g. tool selection, prompt/output review, vendor evaluation, internal training questions)
2. Estimate total hours spent per theme over the logging period
3. Estimate what that time would cost at our standard hourly rate
4. Flag which themes are one-off vs. clearly recurring and likely to continue
5. Write one plain-language paragraph summarizing the pattern, suitable for an internal conversation about whether to price thisDraft the AI Advisory line item using the brief as evidence, not as permission to say no
This isn't about refusing the client — it's about naming what's already happening and pricing it like the rest of the work. Bring the pattern brief into the next renewal or SOW conversation as the reason a small AI Advisory retainer or hourly block now exists.
You are my agency positioning assistant. I want to convert informal, unpaid AI advisory requests from a client into a priced line item.
Here is the pattern of what they've been asking for, unpaid, over the last month: [PASTE PATTERN BRIEF FROM STEP 2]
Write a short, non-defensive proposal section (under 250 words) that:
1. Names the pattern factually, without sounding like a complaint ("we've noticed a recurring need for...")
2. Frames it as a service worth having, not a problem to fix
3. Proposes either a monthly AI Advisory retainer (fixed hours) or a per-question hourly block, both with a rough price range
4. Makes clear existing scope and deliverables are unaffected — this only covers the advisory asks that sit outside them
5. Ends with a simple next step to confirm and startHave one scripted line ready for the in-the-moment ask
Most of these requests come in casually, mid-Slack-thread, and the easiest thing to do is just answer them for free again out of habit. Have one calibrated response ready that doesn't shut the door: "Happy to dig into that — let's log it under advisory hours so it doesn't eat into the project work." Said the same way every time, it stops feeling like a negotiation and starts feeling like a normal part of how the account runs.
Revisit the pattern every quarter, because the questions keep changing
What clients ask about AI in one quarter (which chatbot to pick) is not what they'll ask about in the next (which agent framework to trust with their data, how to audit an AI vendor's security). Re-run the log-and-brief cycle quarterly so the advisory scope — and its price — keeps pace with what's actually being asked instead of going stale.
What changes
The agency stops absorbing an invisible AI helpdesk for free, the client gets a clearer, named channel for these questions instead of a Slack favor, and the relationship stays clean because nobody's quietly resentful about unpaid work piling up in the background.
Somewhere in the last few Slack threads, your account lead answered "which AI note-taker should we use" and "is this ChatGPT draft okay to send to our CEO" back to back, in under ten minutes, without thinking twice. Neither question is in the SOW. Neither got logged, billed, or even noticed as work. It just felt like being helpful.
That's the whole problem. It always feels like being helpful, right up until you add up how much of it is happening every month.
The scope creep that doesn't look like scope creep
Classic scope creep is visible — an extra revision round, a "quick" additional page, a new deliverable squeezed into an old budget. This is different. It doesn't show up as extra deliverables. It shows up as extra conversations — tool picks, prompt sanity checks, "does this output look right to you," vendor demos the client wants a second opinion on. Nobody frames it as work, so nobody tracks it, so nobody prices it. It just accumulates as a quiet tax on whoever on your team is most responsive.
Log it before you decide what to do about it
You don't need a policy to start. You need two weeks of one-line log entries every time a client asks something AI-related that isn't in scope. Most agencies skip this step and jump straight to "we should charge for this," which usually stalls because nobody has evidence, just a feeling. The log turns the feeling into a number.
Present the pattern, not a complaint
Once Claude has grouped the log into themes and put a rough hour count and dollar figure against them, the conversation with the client changes shape entirely. You're not saying "stop asking us things." You're saying "we've noticed a recurring need here, and it's worth having a proper channel for it" — which is both true and a much easier sell than it sounds.
Keep one scripted line ready for the moment it happens live
The system only works if it survives the actual Slack thread, at 4pm, when answering for free is the path of least resistance. One calibrated, repeatable line — said the same way every time — turns the boundary into a habit instead of a confrontation you have to relitigate every week.
Bottom line
Clients aren't wrong to ask their agency about AI — you're often the most credible voice they have on it. The mistake is letting "credible voice" quietly turn into "free consultant" by default. Name the pattern, price the pattern, and the same conversations that used to drain margin can become one of the easiest add-on lines you sell all year.