Charlie from xSquads

One thing that bites a lot of teams with HubSpot routing is the order of checks. Put your exceptions first: existing customers and named accounts go to their account owner before any territory or round-robin rule runs, otherwise a new contact from a key account lands with a random SDR. Then do territory, then rotation. Two practical tips. Add a fallback owner (or a shared queue) for anything that matches no rule, and set a daily view of contacts with no owner so gaps get caught fast. Also normalise country and company size values with dropdowns or a cleanup workflow, because free-text "USA" vs "United States" quietly breaks branch conditions. Finally, test with five or six dummy contacts covering each branch before turning it on, and add a rule that skips reps marked out of office so leads don't sit for days.

Useful audit. One practical way to run it: export your subscription list from your accounting tool, then put each tool in a sheet with columns for who uses it daily, which other tools read or write to it, and the last time anyone changed its workflow. Your four questions then become simple yes/no cells, and the "core" tools usually stand out within an hour. If you want a quick first pass, paste the sheet into your AI assistant with something like: "For each tool, answer the four questions from this list, flag anything with two or more yeses, and name who currently owns the integrations." Then check the answers with whoever actually uses each tool. The ownership gap is often the most revealing part, since the glue work tends to sit with one person.

The "say I don't know" rule is the part most support teams skip, so it's worth testing explicitly. A simple check before launch: pull 50 real tickets from last month, including 10 where the answer isn't in your help center, and run them through the agent. Score three things: correct answer, correct refusal, and whether the handoff included context. For the handoff, make the agent write a summary the human can act on without re-asking the customer. A sample instruction: "If confidence is low or the customer mentions a refund, legal issue or cancellation, stop and send: customer ask, what you already tried, account details found, suggested next step." Then track the handoff rate and the reopen rate weekly. If reopens climb, your knowledge base has gaps, not your prompt.

Nice breakdown. One practical addition for the monthly check in Step 6: keep it as a small spreadsheet so it stays comparable over time. Use 10 to 15 questions a real customer would ask, one per row, and one column per assistant (ChatGPT, Perplexity, Gemini, Copilot, Claude). For each cell, note three things: were you mentioned, which pages or sites were cited instead, and what the answer said about you. After two or three months, look at the cited sources column. If the same directories or comparison sites keep showing up, that's your to-do list for where to get listed or mentioned. Also save the exact wording of each question, because small changes in phrasing can change who gets named. Thanks for flagging that the exact selection rules aren't published.

I like the "one task that wastes time every week" advice. One thing that makes it stick: before you try a tool, time the task three times and write down what a good result looks like, for example "reply under 5 minutes, correct price, no promises we can't keep". Then after a few weeks you're comparing against a real baseline, not a feeling. For the follow-up area, a simple setup is to let AI draft the check-in from your notes, but have a person click send and own anything about price or dates. Also keep a short log of the drafts you had to rewrite and why. After two weeks that list tells you exactly where the human has to stay in charge, and it's usually the same three or four situations.

The three clocks idea is the part I would steal first. One add-on for the month-end problem: put a freeze calendar on your stuck-work board, and tag every waiting item with the tester's name and a "needs testing by" date. When work lands in testing, send a five-minute test script instead of a vague "please test": three steps, the expected result, and a yes or no reply. Anything older than 10 days goes into your weekly leadership update with its age shown. Short, specific asks still get answered during quarter-end, and your data makes that conversation easy.

Nice breakdown. One thing that helped me with the RISEN-style cold email example: add a "don't" line and a source line. For instance, "Use only the facts I paste below about the prospect; if something isn't there, leave a [MISSING] placeholder instead of guessing." That stops the model inventing a pain point or a stat, which is the most common way these emails go wrong. Also, rather than asking for one result, ask for three variants with a one-line reason for each, then pick and edit. Finally, keep a short "good outputs" note with two or three emails you actually sent and liked, and paste them as examples next time. Tweaking one variable per run, as you suggest, makes it easy to see which change helped.

The missing information column is the one I would start with. After a few weeks, sort the blanks by how often they appear, then turn the top one into a required field on the enquiry form or a single automated follow-up question. That is a small, reversible first automation. One pitfall: logging the outcome reason as free text, because you end up with fifty phrasings of the same thing. Use a short dropdown such as no budget, no response, went with a competitor, bad timing, plus an Other box, and review Other monthly. Also record the timestamp of the first human reply, since response time often explains lost leads better than lead quality does. Then run the new step alongside the manual one for two weeks and compare results before switching over.

This matches what I see with small support desks: the win is retrieval and timing, not clever replies. One practical way to start this week: export your last 200 tickets and ask ChatGPT or Claude to cluster them into the top 10 question types, with a count for each. Usually 3 or 4 types (where is my order, return window, sizing, address change) cover more than half the volume. Then for each of those, write one approved answer template with placeholders, and give the model this prompt: "Using only the policy text below and the order data provided, draft a reply. If anything is missing, list what to ask the customer instead of guessing." That last line cuts invented policies a lot. Keep a human click before sending for refunds and anything angry, and review ten drafts a week to tweak the templates. Thanks for keeping it grounded for three-person shops.

A trick that's helped me at work: turn this guide into one reusable template so you stop rewriting prompts. Try this for any business task: "Act as a [role, e.g. sales ops analyst]. Context: [who the audience is, what happened, constraints]. Task: [one specific deliverable]. Here is an example of the style I want: [paste]. Format: [table with columns X, Y, Z / 5 bullets / under 150 words]. Before answering, ask me up to 3 questions if anything is missing." That last line is the big win. It makes ChatGPT surface the gaps instead of guessing. Then save the version that worked in a notes doc as your team's standard prompt, and swap only the context each time. For recurring work like weekly reports or customer replies, that consistency saves more time than any clever wording.