How Winnetka ETA Operators Can Use Claude Opus 4.7 to Grow Revenue After Closing

Winnetka, Ill. , April 19, 2026. You just wired the money. The seller handed you the keys. You walk into the office on a Tuesday morning, and the first thing you notice is that the break-room coffee maker is older than the longest-tenured employee. The second thing you notice is that the QuickBooks file hasn't been reconciled since February, the CRM is a shared Excel sheet on a network drive nobody uses anymore, and the "marketing plan" is a printed PDF taped to the wall.

Welcome to Entrepreneurship Through Acquisition.

If you came out of an MBA program at Kellogg, Booth, Stanford, or Harvard and you have spent the last eighteen months on a search, you already know the deal. ETA is not about discovering unicorns. It is about buying a boring, profitable business, the kind of $2M-to-$5M EBITDA company on the North Shore that nobody in Silicon Valley would ever look at, and turning it into a bigger, more profitable version of itself.

The problem is that the previous owner ran the whole thing out of their head. Tribal knowledge is everywhere. The longest-tenured employee has been there for 22 years and is planning to retire in eight months. And you, the brand-new CEO, have maybe $300K in working capital and a covenant-heavy SBA note telling you that you have about six quarters to prove the thesis before your bank gets nervous.

So when Anthropic released Claude Opus 4.7 on April 16, three days before this article went live, the date matters. This is not an incremental update. It is the biggest practical jump in usable business-capability AI that North Shore operators have seen in the last twelve months. The window to deploy it before competitors is right now.

What follows is what actually changed in Opus 4.7, and three specific, boring, revenue-positive things an ETA operator in Winnetka can do with it this month.

What Actually Shipped in Opus 4.7

Opus 4.7 is Anthropic's newest flagship model. It sits at the top of the Claude product lineup, above Sonnet and Haiku. Any firm already using Claude through Claude Cowork, the Claude.ai web interface, or a custom integration will be automatically moved onto 4.7 over the next few weeks. Opus 4.6 is being deprecated on the API in June, according to Anthropic's release notes.

Four things changed that matter for a Main Street business owner.

Coding and structured-task capability jumped. SWE-bench Verified, an industry benchmark that measures how often a model can solve a real, previously-unseen software engineering ticket end to end, went from 80.8% on Opus 4.6 to 87.6% on 4.7. SWE-bench Pro, the harder variant, went from 53.4% to 64.3%. In plain English, the model is now reliably good enough to do first-draft technical work that, twelve months ago, required a developer.

Vision resolution tripled. Opus 4.7 can now process images up to 2,576 pixels on the long edge, roughly 3.75 megapixels, according to Anthropic. That is about three times what previous Claude models could handle. On the "visual navigation" benchmark, which tests whether the model can look at a screenshot of a software interface and understand what is on the screen, Opus 4.7 scored 79.5% compared to 57.7% for 4.6. That is the single number to remember. It means Claude can now look at the Excel dashboard your controller emails you every Friday and actually read it correctly on the first try.

Task budgets are new. This beta feature lets a user tell Claude, at the start of a long job, roughly how many tokens it is allowed to consume before stopping and asking for permission. For an owner paying per-token, this prevents a "Claude got stuck in a loop and burned $40 of API credit" situation. It also means overnight automated workflows can run safely without babysitting.

The /ultrareview command. Inside Claude Code, the developer-facing version of Claude, a new slash command called /ultrareview spawns a multi-agent code review. No ETA operator is ever going to type this command directly. But the consultant, freelance developer, or AI agency you hire will, and it matters because it means a single developer can now do the work that used to require a small team.

Pricing did not change. Five dollars per million input tokens, twenty-five per million output. This is the single most underrated part of the release. Anthropic is providing a meaningfully better model at the same price, which means any ROI calculation an operator ran on Opus 4.6 just got better.

For independent verification: the Help Net Security coverage and the GitHub changelog announcement published the same day.

Why This Release Matters Specifically for ETA Operators

There is a particular kind of math an ETA operator has to do in the first 90 days after close.

The business was bought for, let us say, 4.5x EBITDA. 70% was financed with an SBA loan carrying a personal guarantee. The first principal-plus-interest payment is usually three months after close. The lender is looking at monthly financials and mentally modeling the debt-service coverage ratio. If that DSCR falls below about 1.25x for more than a quarter or two, the lender starts asking uncomfortable questions, the kind that put a personal residence, signed over as collateral, into the conversation.

At the same time, any seller note is usually due to start amortizing in month 13 or month 24. Equity investors, if outside capital came in, expect 15% to 25% annual EBITDA growth for the five- to seven-year hold. Missing the first year makes every subsequent year harder.

So in the first 90 days, an ETA operator has to do three things simultaneously: stop the bleeding on anything the previous owner was doing badly, find the revenue the previous owner was leaving on the table, and build operating capacity to grow without proportionally growing headcount.

Traditionally, this is why ETA operators hire a business-improvement consultant or bring on a 2IC, a second-in-command, in month three or four. That is $120K to $180K in fully-loaded cost for the 2IC, or $25K to $75K for a short consulting engagement.

Claude Opus 4.7, deployed intelligently, is not a substitute for a great hire. But it is a substitute for a lot of the work that a junior analyst, a part-time marketing coordinator, a financial analyst, or an operations associate would do. In the first 90 days, that substitution is often the difference between proving the thesis on schedule and triggering a covenant.

Three specific use cases illustrate the point.

Use Case 1: Finding Revenue in Dormant Customer Data

The seller's bottom-80% customers represent the most under-priced asset in most ETA acquisitions.

A pattern shows up across almost every post-close situation in Chicago's North Shore, based on 97 AI projects worth of observation. The seller had a customer list. It has between 400 and 4,000 customers on it, depending on the business. The top 20% drove 80% of revenue for the last three years, and the seller was entirely focused on keeping them happy. The bottom 80%, who collectively represent somewhere between 10% and 25% of historical revenue, have not been touched in two years.

That is the first pool of found money.

The problem with reactivating dormant customers is not that nobody knows they exist. The problem is that re-engaging 2,000 customers you have never met, each of whom has a different purchase history and a different reason for going dormant, is a genuinely hard, time-consuming task. To do it well by hand would take a full-time junior hire the better part of four months.

The Opus 4.7 workflow replaces that.

The operator exports three files: a CRM contact list (usually from HubSpot, Salesforce, or more often in an ETA acquisition a shared Excel sheet), an invoice history from QuickBooks or whatever accounting software the previous owner used, and a year of email archive with customer communications. All three go into a Claude project. Opus 4.7's expanded context window, up to 1 million tokens on the enterprise tier, lets the model load substantially more raw data than previous versions. The operator gives it one prompt.

On Opus 4.6, a prompt like that would have worked maybe 70% of the time. It would often miss fields, misread dates, or get confused by inconsistent formatting in the CRM export. On 4.7, with the improved instruction following and the higher-resolution vision reading spreadsheet screenshots when needed, this works on the first pass the vast majority of the time.

The output is 50 personalized notes, each with the customer's history embedded, that the new owner can send over the course of three weeks. Reactivation rates on this kind of outreach typically run 8% to 14% for a well-run campaign. If the average dormant customer represented $3,500 in annual revenue at the acquired business, reactivating 5 of them pays for a year of Claude at the enterprise tier. Reactivating 15 of them is the kind of result that belongs on a first quarterly investor update.

Two things make this work specifically on 4.7. First, the instruction-following improvement. Opus 4.7 now refuses to silently generalize a request from one customer to another, which means the notes are actually personalized rather than five variations of a template. Second, the vision upgrade means if the operator needs to show Claude a screenshot of the old POS system or a PDF invoice the previous owner kept in a drawer, the model reads it correctly.

Use Case 2: Making Sense of the Inherited Reporting Stack

Every ETA operator has the same story about their first month. They asked the controller for a "financial dashboard" and what they got was a 12-tab Excel file with links to three other workbooks, two SQL queries that run against the accounting system, and a line in the seller's handover notes that says "ask Linda about the inventory adjustments."

In the old world, the two options were to spend six weeks learning the spreadsheet yourself, or to hire someone to rebuild it. The first option steals time from revenue work. The second costs money and takes months.

There is now a third option.

The operator takes screenshots of every tab in the inherited Excel file. Screenshots of the two SQL query results. Drops them into Claude, along with a one-paragraph description from the controller of what the business actually does. Then asks Opus 4.7 to map the entire reporting stack: where every number comes from, what the dependencies are, and where the reporting logic could be simplified.

This specific workflow did not work reliably on Opus 4.6. The vision resolution was too low. The model could not read the numbers in a typical Excel screenshot, so manual transcription was required, which defeated the point. On 4.7, at 3.75 megapixels, an entire Excel worksheet can be screenshotted and the model will read it correctly.

The output is a document that does three things. It explains, in plain English, what every metric on the dashboard actually measures. It flags every place the seller's spreadsheet is doing something questionable, including double-counting, rolled-up assumptions, and hard-coded values that should be formulas. And it gives the operator a prioritized list of the five to ten things to fix first to have a financial picture that can actually be trusted.

That output is worth, conservatively, the first month of a $90K-per-year financial analyst. The ETA operator does not have to hire the analyst in the first 90 days. They can wait until month 6 or month 9, when cash is more comfortable, and bring someone in who inherits a clean reporting environment rather than spending their first three months untangling what Linda left behind.

Paired with Claude Cowork's scheduled-task feature, the operator can go further: set up a recurring job that every Friday at 5 PM reads the updated Excel dashboard, compares it to the previous week, writes a two-page variance explanation in the operator's voice, and emails it for Monday morning review. That is the work a junior financial analyst does, at closer to $50 a month than $7,500.

Use Case 3: Underwriting the Next Deal Without a Tuck-In Team

Most ETA operators do not plan to own just one business for seven years. The real plan is to buy one company, stabilize it, and then do two to four tuck-in acquisitions of smaller competitors or complementary operators over the next three to five years. That is where the real equity value is created, and it is how the model has evolved on the North Shore. The area is, structurally, full of owner-operator businesses at a generational inflection point, as covered in this guide on what Lake Forest business owners should know about AI in 2026.

The problem with tuck-ins: underwriting one is genuinely hard, and most ETA operators do not have a full corp-dev team. The operator is a CEO, a controller, and maybe a part-time ops person. Every deal looked at takes 20 to 40 hours of time that is not going into growing the core business.

Task budgets in Opus 4.7 change this math.

Here is the workflow. A CIM, or confidential information memorandum, arrives from a broker about a tuck-in target. Historically, handing that CIM to Claude produced a decent summary, but a proper underwriting pass required running Claude against the full data room once it opened: financials, customer contracts, employee records, the whole thing. That analysis could run for hours and consume a frankly unpredictable amount of API spend.

With task budgets, the operator can set up the job like this:

The model runs. It works through every document in the data room. It produces an underwriting memo that, a year ago, would have cost $15,000 to $35,000 from a boutique transaction advisor. It does it in an afternoon, not three weeks. And because of the task budget, the operator knows the API spend in advance: in this case, about $5 in input tokens and maybe $8 in output tokens.

Is the memo as good as what a partner at a lower-middle-market advisory firm would produce? No. It is, however, usually better than what the operator would produce alone in the same amount of time. It gives a starting framework to iterate on with a CFO or financial advisor.

For the first two tuck-in deals after a platform acquisition, this is more than enough. Once the operator is doing larger add-ons or more complicated carve-outs, human advisors come back into the loop. But for the first 18 months of the hold period, when the pressure is to move fast on small deals, Opus 4.7 meaningfully extends the range of opportunities that can be evaluated.

How to Get Started in the First Two Weeks

An operator does not need to rebuild the tech stack. Two or three small pilots run over 14 days will reveal which ones return capital fastest. The sequence below is the one that has worked at similar firms.

What This Does Not Replace

A direct note about something that goes wrong regularly.

Claude Opus 4.7 does not replace hiring. It does not replace a great controller, a great operations manager, or a great salesperson. An operator who thinks of Opus 4.7 as a headcount substitute will miss the point and frustrate the team.

What it does is extend the capability of the people already on the payroll. The controller gets a smart analyst sitting next to them, whispering in their ear. The operations manager gets a pattern-recognition engine that reads every piece of data that comes through the business. The new owner gets a junior associate who will do the boring parts of any analysis at 2 AM without complaining.

The businesses on the North Shore that are going to outperform in this cycle are not the ones that fire their teams and "replace them with AI." They are the ones that use AI to let the team they have focus on the specifically-human work, the client relationship, the judgment call, the negotiation, the mentorship, while the model handles the paperwork, the drafting, and the first-pass analysis.

For searchers still in the search phase who have not closed a deal yet, the same argument applies during diligence. The deal that wins is the one that gets under an LOI fastest and executes cleanest. A single searcher working with Opus 4.7 can now do the diligence and underwriting that used to require a team of three associates.

A Final Note on Timing

Opus 4.7 dropped on April 16. The window in which using it produces a genuine edge over the competing business down the road is approximately six months. By October or November, every serious operator on the North Shore will have figured some version of this out. The operators who move now are the ones who get to compound those gains for the rest of their hold period.

For operators who want to think through what the right first workflow is for a specific business, or want a second set of eyes on how to structure the roll-out so it sticks, a free 30-minute AI audit is available. In person in Winnetka or on video. No obligation, no pitch deck. The output is a one-page plan.

The technology shipped. The question is what an operator does with it in the next 90 days.

Frequently Asked Questions