After an AI Consultation: Turn the Advice Into a 30-Day Plan

Within two days of the call, sort the advice into do now, later and not for us, then pick one job and give it a named owner, a baseline number and four weekly stages: set up, supervised trial, normal use, and a day-30 review against the baseline. One job, one owner, one number beats a list of ten ideas.

Most post-consultation plans fail for the opposite reason to the one owners expect. The advice isn't wrong; there's simply too much of it. A good consultant will mention six to ten possibilities, and starting all of them in the same month means none gets the attention it needs to prove itself. The other ideas aren't lost. They wait on a list for the next 30 days.

Follow me on Instagram@sagnikteaches

Sort the advice within 48 hours, before it goes cold

Notes taken during a call are half-sentences, and a week later nobody remembers what "check the PMS fields" meant. Within two days, write every suggestion on its own line and put it into one of four groups: do now, later, not for us yet and need to ask. A suggestion earns "do now" only if it passes four tests: it happens at least weekly, you can count it, a mistake can be caught before a customer sees it, and a named person has around two hours a week to own it.

Connect on LinkedInSagnik Bhattacharya

As an illustration, take a 16-room boutique hotel with a general manager, a front-desk team of four and an owner who isn't involved day to day. The consultation covered eight ideas, and the general manager sorted them like this:

Subscribe on YouTube@codingliquids

You can ask a chat assistant to do the first pass, provided you give it your rules and check its work. Remove guest names from the notes before pasting them in.

Below are my notes from a call with an AI consultant about our
16-room hotel. List every suggestion on its own line and put each
into one group: DO NOW, LATER, NOT FOR US YET, NEED TO ASK.
DO NOW rules: it happens at least weekly, we can count it, mistakes
can be caught before a guest sees them, and one named person has two
hours a week for it. Choose at most ONE do-now job, plus anything
that must happen before it. Give a one-line reason for each.
Do not add suggestions that are not in my notes.
Notes: [paste notes or transcript with guest names removed]

An illustrative first answer put "AI dynamic pricing: could raise revenue significantly" at the top of DO NOW, gave two do-now jobs despite the limit, and listed a "booking chatbot" that nobody had mentioned. All three faults are typical. The model was drawn to the most impressive-sounding idea, ignored a numeric rule and filled a gap with something plausible. Treat the output as a draft sort: move pricing back to "not for us yet", delete the invented item and apply the four tests yourself.

Pick the one job these 30 days are for

When two candidates both pass, compare them with a quick sum and a question about dependencies. For the hotel, the two strongest were review replies and pre-arrival emails:

• Review replies: about 60 reviews a month across Google and the booking platforms. The team answers about 24 of them at around 8 minutes each, which is 3.2 hours a month. The goal is to answer nearly all of them within two days without adding hours.
• Pre-arrival emails: about 35 arrivals a week at 4 minutes each is 140 minutes a week, or roughly 10 hours a month. That's the bigger time saving on paper.

Pre-arrival emails save more time, so why not start there? Because they depend on data the hotel doesn't have yet: arrival times, occasions and room preferences are missing from about half the bookings. Review replies depend on nothing except a good fact sheet about the hotel, and writing that fact sheet is work the chatbot and the pre-arrival emails will reuse later. The first job should be one that can succeed on its own and leaves something useful behind. For the craft of the replies themselves, see replying to hotel reviews with AI without sounding canned.

Rewrite every action so someone could start it tomorrow

Consultation notes turn into vague intentions unless each action names a person, a verb, an output, a deadline and the place the evidence will live. Three before-and-after rewrites from the hotel's plan:

• Before: "Look into AI for reviews." After: "By Friday at 5pm, the front-desk lead drafts replies to the ten most recent reviews using the prompt saved in the shared AI folder; the general manager approves or edits each and logs the minutes spent in the review sheet."
• Before: "Sort out an AI policy." After: "By Wednesday, the general manager writes one page listing what may and may not go into AI tools (no card details, no passport numbers, no health or dietary notes next to a name), and each front-desk member replies 'agreed' by email."
• Before: "Measure the results." After: "On day 1, the general manager exports the last 90 days of reviews into a sheet with platform, star rating, date posted and date replied, then records the reply rate and the median days to reply."

If an action can't be written this way, it isn't ready for the plan. It usually means a question is still open, which belongs in the "need to ask" group.

An AI action plan template that fits on one page

Keep the whole plan on one page so it can sit next to the keyboard. Anything longer stops being read by the second week. Copy this and fill it in:

AI ACTION PLAN: [the one job]        Start: [date]   Review: [date + 30]
OWNER: [name]      APPROVER: [name]      HOURS A WEEK: [number]

BASELINE (measured before day 1)
  Volume: [per week or month]   Time per item: [minutes]
  Quality measure: [e.g. reply rate, error rate, complaints]

TARGET AT DAY 30
  Main number: [...]   Must not happen: [...]

TOOLS AND COSTS
  [tool, plan, monthly cost]   One-off costs: [...]

RULES
  What staff may put into the tool: [...]
  What always needs approval: [...]

WEEK 1  Set up and baseline:  [actions, owner, due date]
WEEK 2  Supervised trial:     [actions, owner, due date]
WEEK 3  Normal use:           [actions, owner, due date]
WEEK 4  Measure and decide:   [actions, owner, due date]

DECISION AT DAY 30: keep / adjust / stop, and why
PARKED IDEAS: [the later and not-yet lists]

Two fields do most of the work. The baseline is what makes day 30 an answer rather than an opinion; the tutorial on setting a baseline before you introduce AI covers the measuring in more depth. "Must not happen" is the line that keeps a quick win from causing a slow problem: for a hotel it might be "no reply promises a refund or discloses anything about a guest's stay beyond what they wrote".

The hotel's 30 days, week by week

The hotel runs Google Workspace Business Standard, which includes Gemini in Gmail, Docs and Sheets and the Gemini app, so the plan needed no new subscription. A business without a built-in assistant would need one; ChatGPT Business, for comparison, is $25 per user per month billed monthly with a two-seat minimum, so about $50 a month. Here is the filled-in plan, with the time each week actually took:

That's about 13 hours across the month, most of it in week 1. Front-loading is normal: the fact sheet and baseline take real time once and then keep paying back.

This is the reply prompt the lead used, saved in a shared document so nobody retyped it:

You write replies to guest reviews for a 16-room boutique hotel.
Use ONLY the facts in the attached fact sheet. Warm, brief, no more
than 80 words. Thank the guest for one specific thing they mentioned.
If they raise a problem, acknowledge it plainly and say what we are
doing about it only if the fact sheet covers it.
Never offer refunds, discounts, upgrades or compensation. Never
mention other guests or booking details the reviewer didn't write.
If the review mentions illness, injury, allergy, theft or a named
staff member's behaviour, write NEEDS MANAGER as the first line.
Review: [paste review text only]

The last two rules weren't in the first version. Both were added after things went wrong, which is exactly what weeks 2 and 3 are for.

Week 2's catch. Two drafts ended with "our team will be in touch about a partial refund". The fact sheet said nothing about refunds, but the reviews had complained about noise, and the model reached for the kind of gesture it had seen in countless hotel replies. The manager caught both at approval. The fix was one sentence in the prompt, and it cost ten minutes.

Week 3's miss. A four-star review mentioned, halfway through, that a guest had reacted to something at breakfast despite flagging an allergy. Because the review was four stars, it went through the lead's check rather than the manager's, and the posted reply thanked the guest for their "lovely stay". The manager spotted it in the one-in-five spot-check the next morning, edited the reply and called the guest. The rule became: sensitive topics go to the manager whatever the star rating, and the prompt now flags them. The weekly check below is how both problems surfaced quickly instead of at day 30.

The owner's 15-minute Friday check

Once a week, whoever is accountable spends a quarter of an hour on the same questions. Here is the hotel's check at the end of week 2, filled in:

1. Is the one job still the only job? Yes. Someone suggested starting the chatbot early; parked until day 30.
2. Did the owner spend the planned hours? The lead spent 2 of a planned 2.5; approvals took the manager another hour.
3. Is the log up to date? Yes: 14 reviews logged with minutes and edit level.
4. Did anything nearly embarrass us? Yes: two refund promises, caught at approval. Prompt updated Thursday.
5. How do the staff feel about it? The lead finds waiting for approval slow. Manager will approve at 11am and 4pm daily.
6. Costs against budget? $0 in new tools; 9 staff hours so far against 10 planned.
7. Anything to ask the consultant or vendor? Whether to reply to reviews older than 30 days (decided: no, only new ones).
8. On track for the day-30 target? Yes, if week 3 holds the same reply rate without manager approval on every draft.

The questions matter less than the habit. A plan that nobody looks at for three weeks tends to be quietly abandoned, and nobody notices until the owner asks at day 30. If you want a fuller approach to approvals that don't slow everyone down, see setting up human review for AI work.

Day 30: reading the numbers against the baseline

At day 30 the general manager repeated the baseline measurements for days 8 to 30 (the trial period) and set them next to the 90 days before:

Look closely at time. Before, 72 replies over three months at 8 minutes is about 3.2 hours a month. During the trial, 41 replies in 23 days is roughly 53 a month, and at 3.5 minutes that's about 3.1 hours a month. So the hotel spends almost exactly the same time as before, but answers more than twice as many reviews, five days faster. If the goal had been "save time", the result would look like a failure. The goal was "answer nearly every review within two days without adding hours", and on that measure it clearly worked. Write the goal down before day 1, or you'll be tempted to pick whichever number looks best afterwards.

One practical wrinkle: Google says owner replies can take up to 30 days to appear on a listing. Count replies from your own log, not from what's visible on the page on day 30.

To turn numbers into a decision, agree the thresholds on day 1 as well. A sensible default:

• Keep if the main number hit its target, no serious error reached a customer more than once, and the person doing the work wants to carry on.
• Adjust if the main number moved but by less than half the target, or more than one draft in four needed rewriting. Run another 30 days with one change, not five.
• Stop if the main number hasn't moved after two weeks of normal use, or serious errors keep reaching customers after fixes, or the team has quietly gone back to the old way.

The hotel kept it, with two adjustments already made: the sensitive-topics rule and fixed approval times. The tutorial on setting pilot success criteria that hold up has more on choosing thresholds when the stakes are higher.

A "stop" at day 30 can be a good result, too. Suppose a tour operator ran the same kind of plan for AI-drafted itineraries and found that 9 of 12 drafts needed heavy edits because each tour's timings changed with the season. At 35 minutes a draft against 30 minutes by hand, the plan cost them about an hour over the month and told them clearly not to renew a $40-a-month tool. Without the baseline they'd have kept paying on a hunch.

Where 30-day plans usually stall, and the early warning signs

Most stalls are visible by day 10 if you know what to look for:

• Three jobs started at once. A campsite owner begins a booking FAQ bot, social posts and pitch-allocation emails in the same week. By day 10 all three are half set up and the log has no entries. The sign: nothing measurable has happened by the second Friday.
• No named owner. A letting agency's plan says "the team will test AI for maintenance requests". At the Friday check nobody can say who spent any hours on it. The sign: the check's second question gets a shrug.
• No baseline. Everyone at a holiday-let company feels the new guest emails are quicker, but nobody timed the old ones. At day 30 the director asks for evidence and there isn't any. Even 20 timed examples before day 1 would have settled it.
• The approver disappears. The manager who approves drafts goes on leave in week 3, drafts pile up, and staff start posting unapproved replies to clear the queue. Name a deputy approver in week 1.

If your plan has already stalled, the tutorial on why AI pilots stall and how to get them live works through recovery. If the whole idea came from a call that you're not sure was worth it, what a free AI consultation gives you helps you weigh the advice you started from.

What happens to the ideas you parked

At day 30, go back to the four groups. The "later" list becomes the candidates for the next 30 days, and the first job often makes one of them easier. At the hotel, the fact sheet written for review replies now answers most of what a website chatbot would need, and the review log showed which questions guests ask most. Pre-arrival emails are still blocked by missing booking fields, so the next plan's week 1 includes getting front desk to fill arrival time and occasion on every new booking.

"Not for us yet" items get a date rather than a deletion: dynamic pricing is revisited when the booking system renews. "Need to ask" items become a short email to the consultant or the software vendor. After two or three cycles you have a string of small, proven changes instead of one large, uncertain project, and each plan starts from a better baseline than the last.

Further reads

• How to Prepare for an AI Consultation and Leave With a Plan — Get more usable advice out of the call in the first place.
• How to Judge Whether Your AI Consultant Delivered Value — Decide whether the advice itself was worth what you paid.
• AI Implementation Roadmap for Small Businesses: 5 Phases in 90 Days — Where a 30-day plan fits inside a 90-day roadmap.
• How to Choose Your First AI Project: 7 Tests Before You Commit — Seven tests for choosing the job your plan is built around.
• How to Write an AI Usage Policy for Your Small Business — Expand the one-page staff rules into a proper policy.
• How to Automate Pre-Arrival and Post-Stay Emails With AI — A natural second job for a hotel after review replies.
• Is AI Dynamic Pricing Worth It for a Small Hotel? — Whether the idea the hotel parked is worth revisiting.
• Is a 1:1 AI Consultation Worth It for a Small Business? — The real cost of a 1:1 AI consultation, how many hours it must save to break even, and three cafés that get three different answers.
• What Happens in a 1:1 AI Implementation Consultation? — What to send beforehand, how a consultant maps your work and scores AI jobs, the tool check, and the one-page plan you should leave with.
• How Much Does an AI Readiness Assessment Cost? — Free, vendor-funded and paid readiness assessments compared, where the days go, your staff's share of the cost, and a way to compare three quotes fairly.
• What Is the Cheapest Way to Get Expert AI Advice? — Free help pages, forums, adoption kits and discovery calls answer most AI questions. When paid advice is worth it, and how to compare quotes.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.