The First 90 Days: The AI Operating Plan I'd Run After Buying a Small Business

An AI operating plan for the first 90 days after buying a small business is a written, week-by-week sequence that uses AI to capture how the business actually runs, protect the customer relationships you just paid for, and free the new owner to spend their hours on the two things that decide whether the acquisition works: keeping the team and keeping the revenue. It is not a plan to automate the business in three months. It is a plan to see the business clearly in month one, stabilize it in month two, and start compounding in month three.

I write about this from the buyer's seat. I run acquisition pipelines for small businesses and real estate, and I have already covered how AI helps find off-market deals, how it supports due diligence before you close, and how it speeds up underwriting without owning the buy decision. This post is the piece that comes after the wire clears. Most acquisition content stops at closing day. Most acquisition failures start there.

Key Takeaways

• The first 90 days split into three blocks: Days 1-30 are for capture and observation, Days 31-60 for stabilization and the first automations, Days 61-90 for compounding systems and a real operating rhythm.
• AI's highest-value job in a transition is documentation: turning the seller's knowledge, the inboxes, the call logs, and the job files into a written operating manual before that knowledge walks out the door.
• Customer-facing communication in the first 30 days stays human. The moment to introduce AI into customer touchpoints is after you understand the relationships, not before.
• Every automation you add in the first 90 days needs a log the owner reviews weekly. You are still learning what normal looks like, so you cannot yet tell a helpful automation from a quiet mistake.
• The seller's transition period is a depreciating asset. Spend it extracting undocumented knowledge with structured AI-assisted interviews, not on tasks anyone can do later.

Why the first 90 days decide the deal

Buy a solid business at a fair price and you can still lose it in the transition. The three standard failure modes are customer attrition when the familiar owner disappears, staff departures when the new owner changes too much too fast, and cash flow surprises when informal processes the seller ran from memory stop happening. None of these are spreadsheet problems. All of them are information problems, and information problems are exactly where AI earns its keep.

Here is the uncomfortable math. If the business does $900,000 a year across roughly 400 customers, the top 40 relationships probably carry half the revenue. Losing five of those accounts in a rocky transition can erase your entire first-year debt service cushion. Meanwhile the knowledge of who those 40 customers are, what they expect, and which of them are one bad month from leaving usually lives in one place: the seller's head. Your 90-day plan is really a race to move that knowledge somewhere durable before the seller's attention, availability, and goodwill fade.

The plan below assumes a manager-run or owner-run service business in the $500K to $3M revenue range, a seller transition period of 30 to 90 days, and a buyer who intends to operate, not flip. Adjust the specifics, keep the structure.

The three-block structure at a glance

The order matters more than the contents. Buyers who start automating in week one are automating a business they do not understand yet. Buyers who are still "observing" in month three never get their time back. Thirty days per block, with a written exit checklist for each, keeps you honest in both directions.

Days 1-30: capture everything, change almost nothing

The first block has one deliverable: a written operating manual for a business that has never had one. AI makes this achievable in 30 days instead of a year, because the raw material already exists. It is just trapped in formats nobody reads.

Point AI at the artifacts first:

• The shared inbox. Have AI read the last six months of the main service inbox and produce a taxonomy: what types of requests arrive, in what volume, who answers them, how fast, and with what boilerplate. This becomes your intake process document and usually surfaces two or three request types nobody mentioned in diligence.
• The job or order files. A pass over the last 200 jobs produces the real service catalog: what the business actually sells, at what average ticket, with what margin spread. Sellers describe the business they think they run. The job files describe the one they do run.
• The calendar and call logs. These reveal the true operating rhythm, including the informal Thursday supplier call and the monthly customer who never shows up in the CRM.
• Vendor invoices. AI extraction across a year of invoices gives you a vendor list with spend, frequency, and the renewal dates diligence may have missed.

Then use AI to squeeze the seller's transition time, which is the single most valuable and most perishable resource you own in month one. Before each seller session, have AI generate structured interview questions from the gaps in the documentation it just built: "The inbox shows warranty claims going to Dana, but there is no written warranty policy. What is it?" Record the sessions, transcribe them, and have AI merge the answers back into the manual the same day. A seller doing three one-hour structured sessions a week for four weeks will transfer more operating knowledge than six months of casual shadowing.

What stays human in this block: every customer conversation. The top 20 accounts get a personal visit or call from you, without an AI-drafted script, because the thing they are deciding in that meeting is whether they trust you. Trust is in the category of work I never automate, and never more so than in a transition.

Days 31-60: stabilize with the first automations

By day 31 you have a manual, a customer map, and a list of what is quietly broken. The second block is where AI moves from documenting the business to operating small parts of it, in a specific order: internal before external, drafts before sends, logged before trusted.

The first three automations I would stand up, in order:

1. The weekly owner's report. An automated Monday summary: revenue booked last week, jobs completed, new inquiries, outstanding invoices past 30 days, and any customer who has gone unusually quiet. This is the single automation with the best ratio of effort to insight, because it replaces the dashboard-checking habit that eats new owners' mornings. Everything in it comes from systems you already mapped in block one.

• Follow-up drafting. Every completed job triggers a drafted follow-up message in the company's existing tone, which a human reviews and sends. Drafts, not sends. You are eight weeks into owning these relationships; the review step is where you keep learning how the customers talk.
• Inbox triage. Incoming requests get classified and routed using the taxonomy from block one, with anything ambiguous going to a human queue. Measure the misroute rate for two weeks before you trust it.

Each automation gets a log, and the log gets ten minutes in your weekly review. This is the discipline that separates an AI-operated business from a business with unattended scripts. In month two you still do not fully know what normal looks like, which means you cannot recognize abnormal. The log review is how you catch the follow-up draft that misread a complaint as a thank-you before it becomes a pattern.

What stays human in this block: pricing decisions, anything involving an employee's pay or role, and every apology. A transition is exactly when a mishandled complaint costs the most, so complaints route to you, with AI supplying the account history behind them.

Days 61-90: compound into an operating rhythm

The third block turns one-off automations into a rhythm the business runs on after the transition ends. The test for everything you add here: would this still be running, unchanged, in month twelve?

• The weekly scorecard. Pick five to seven numbers that describe the health of the business: booked revenue, close rate on quotes, average days to invoice, repeat-customer share, and one leading indicator specific to your industry. AI assembles the scorecard; you read it every Friday. If a number cannot be pulled automatically, that is a systems gap worth fixing now.
• Pipeline and CRM hygiene. A nightly pass that flags quotes with no follow-up in seven days, jobs closed without an invoice, and contacts with no owner. Small leaks, found daily, instead of a painful quarterly cleanup.
• Review and referral requests. Once follow-up drafting has run clean for a month, completed jobs with a positive signal graduate to an automated review request. This is usually the first customer-facing send I fully automate, because the downside of an imperfect message is small and the compounding upside of review velocity is large.
• The decision log. Start recording the operating decisions you make and why. Six months from now, when you consider a manager, this log plus the operating manual is the difference between delegating a system and delegating chaos.

By day 90 the shape should look like this:

Days 1-30            Days 31-60              Days 61-90
CAPTURE              STABILIZE               COMPOUND
─────────            ──────────              ─────────
Operating manual  →  Weekly owner report  →  Weekly scorecard
Customer map      →  Follow-up drafts     →  Review requests
Seller interviews →  Inbox triage         →  Pipeline hygiene
Vendor map        →  Fix broken processes →  Decision log
                                     ↓
                     Owner hours shift to: team, customers, next deal

The point of the diagram is the bottom line. Every block exists to move owner hours away from clerical work and toward the work that only the owner can do. If your calendar in week 13 looks like your calendar in week 2, the plan failed even if every automation runs.

The exit checklist for each block

Blocks only work if they have doors. Before you let yourself move from one block to the next, check the list, in writing.

Leaving Days 1-30 requires: an operating manual covering intake, delivery, invoicing, and the top ten recurring processes, each verified by the seller or the senior employee; a customer map with the top 20 accounts ranked by revenue, each with a completed personal touch from you; a vendor list with spend and renewal dates; and at least eight seller interview sessions transcribed and merged into the manual. If the seller's transition window is shorter than 30 days, this block compresses but never gets skipped. It is the block the whole plan stands on.

Leaving Days 31-60 requires: the weekly owner's report arriving without manual effort for three consecutive Mondays; follow-up drafts running on every completed job with a reviewed-and-sent rate you can state from the log; inbox triage holding a misroute rate under roughly five percent for two straight weeks; and the one or two broken processes the capture phase exposed either fixed or scheduled with a named owner and a date.

Leaving Days 61-90 requires: a scorecard you have actually read for four consecutive Fridays; pipeline hygiene flags going to a person who clears them; the first customer-facing automation live with its log clean for 30 days; and your hours tracking showing clerical time trending down three weeks running. Miss a gate and you extend the block, not the plan's ambitions.

What I would not automate in the first 90 days

The restraint list matters as much as the build list, so here it is plainly.

• Outbound customer communication in month one. Customers are watching for signs the business they trusted just changed. A subtly off automated message is exactly such a sign.
• Anything the employees can see before you have explained it. Staff in an acquired business are already nervous. An automation that appears without context reads as a preview of layoffs. Every automation gets introduced as "this takes a task off your plate," because in this plan, that is what it is.
• Collections. Chasing the past-due list from the seller's era requires judgment about which customers are worth pressure and which are worth grace. AI preps the list and the history; the calls are yours.
• Vendor renegotiation. Month one is for learning why the seller paid what they paid. There is often a reason the cheap supplier was not used.
• Any decision with someone's name attached. Hiring, firing, pay, and public commitments stay human, in the transition and after it, for the reasons I laid out in what AI should not do even when the stakes were only my own money.

How this connects to the deal you underwrote

The 90-day plan is not separate from the acquisition work; it is the continuation of it. The customer concentration analysis from due diligence becomes the top-20 visit list in block one. The add-backs you questioned during underwriting become the expense lines the weekly report watches. The seller's answers in diligence become the first draft of the interview questions in the capture phase. Buyers who treat diligence documents as disposable closing paperwork rebuild all of that knowledge from scratch in month one, at exactly the moment they have the least spare attention.

This is also the honest pitch for building the AI muscle before you buy. If the first time you use AI to read an inbox or normalize a financial statement is day one of owning a business, you will spend your transition learning tools instead of learning the company. Run the same plays on the businesses you evaluate and pass on, and by the time you close, the 90-day plan is muscle memory.

The measuring stick: owner hours, not automation count

The metric that tells you whether the plan worked is not how many automations you shipped. It is where your hours went. Track it weekly with one honest note: hours on clerical and administrative work versus hours on team, customers, and strategy. In week two, expect something like 70/30 clerical. By week thirteen, the target is closer to 30/70. That reversal, not the tooling, is the return on the plan.

A second measuring stick: the business should be more legible than it was under the seller, not just equally functional. If you were hit by the proverbial bus in month four, could a competent operator pick up the manual, the scorecard, and the decision log and run the business? Sellers spend decades building businesses only they can run. The 90-day window is your one chance to break that pattern before you accidentally rebuild it around yourself.

FAQ

Should I start automating before the seller leaves?

Yes, but only the internal, invisible layer: documentation, reporting, and analysis. The seller is your fact-checker for everything AI extracts, and that review is far cheaper while they are still answering the phone. Customer-facing automation waits until you, not the seller, are the person customers associate with the business.

What if the business I bought has no digital systems to point AI at?

Paper-heavy businesses take an extra two to three weeks in the capture phase, not a different plan. Modern AI extraction handles scanned invoices, photographed job sheets, and handwritten schedules well. The capture block simply starts with digitization, and the operating manual you produce is an even bigger competitive upgrade because the previous owner truly had nothing written down.

How much should I budget for AI tooling in the first 90 days?

For a business under $3M in revenue, plan on $200 to $500 a month: an AI assistant subscription, transcription, and modest automation glue. The real cost is your time configuring it, which is why the plan sequences a small number of high-yield automations instead of a platform migration. Do not replace the CRM or the field-service software in the first 90 days unless it is actively losing data.

I'm not technical. Can I run this plan myself?

The capture block, yes: interviewing the seller with AI-generated questions and having AI summarize documents requires no technical skill. The stabilize and compound blocks involve wiring systems together, and that is a reasonable place to bring in help for a defined build. What I would not outsource is the weekly log review and the scorecard reading. The owner who delegates understanding the business has repeated the seller's mistake with extra steps.

Does this plan change for a real estate acquisition instead of an operating business?

The structure holds; the artifacts change. For a rental property or RV park, the capture block reads leases, ledgers, and maintenance history instead of job files, the stabilize block automates rent-roll reporting and maintenance intake, and the compound block builds the occupancy and delinquency scorecard. The restraint list is identical: tenant-facing communication earns automation slowly, and anything touching a person's housing gets a human decision.

What is the single most common mistake in the first 90 days?

Changing customer-visible things before understanding them. New owners want to signal improvement, so they redo the invoicing format, the phone greeting, or the service packages in week three. Every one of those is a cue to customers to re-shop a relationship they had stopped thinking about. The AI operating plan is deliberately boring on the outside for 60 days. The visible improvements land in block three, after the relationships have transferred.

If you are working through an acquisition and want the AI side of the transition built with you rather than figured out alone at midnight, that is exactly the work I do with a small number of clients. Request a Strategic AI Consulting Conversation and bring the deal you are closing.

Current Search Intent Check

Recent Search Console data shows people arriving through "cost segregation rv parks texas". That changes the bar for this post: it needs to answer the operator question directly, name the workflow being improved, and give the reader a practical decision rule instead of another broad AI opinion.

Recent Search Console data shows people arriving through "audrey ashworth". That changes the bar for this post: it needs to answer the operator question directly, name the workflow being improved, and give the reader a practical decision rule instead of another broad AI opinion.