AI in an Established Business: How to Start

A month ago, the owner of a restaurant chain with three branches in central Israel sat across from me: 12 years in the market, 80 employees, stable cash flow. He asked me a question I hear in almost every meeting this past year: "Everyone is talking about AI. But what works for me, works. Why should I take the risk?"

That is not a cowardly question. It is the right question.

Since 2008, 18 years in the field, I have guided dozens of established businesses through different waves of digitization: moves to ERP, CRM implementation, entering e-commerce, and now AI. Again and again I see the same dilemma, and again and again the same pattern: businesses that do it right gain. Businesses that jump in because "everyone is doing it" lose money and operational peace of mind.

In this article I break down what an established business really needs to know before implementing AI. It is fundamentally different from what young AI vendors sell you, because they sell technology, and I look at businesses that have already been running for a decade or more.

Why established businesses avoid AI (and why that actually makes sense)

The logical fear: "If it works, don't break it"

Every owner of an established business has built, over the years, a chain of processes that works. The restaurateur who knows exactly how many sea bream to order on Thursday. The importer from China who personally knows the three suppliers who met critical deadlines. The head of education who developed a hiring method that cuts turnover by 40%.

This knowledge is not written in any SOP. It is in people's heads. Any change that threatens to replace it is, rightly, frightening.

The confusion created by AI vendors' marketing

Open LinkedIn today. You will see dozens of posts about "how AI will change your business within 30 days". Most are written by companies that have existed for less than two years, trying to sell a tool to SMBs and showing examples from Fortune 500 use cases.

An established business that reads this is left with two conflicting feelings: "I have to do something" and "I don't understand what". This is a dangerous combination, because it leads to impulsive purchases that do not fit the business.

No clarity about real ROI

A question that comes up in every meeting: "What will I get for these NIS 100,000?" The average AI vendor will answer in buzzwords ("efficiency", "insights", "scale"). An experienced consultant will answer in numbers: what the baseline is today, what the measurable target is, and what happens if we do not reach it.

If your vendor cannot answer the third question, do not sign.

The fundamental difference between a startup and an established business in the AI context

There is a fundamental difference between how a young company approaches AI and how an established business should approach it. This difference is not cosmetic. It changes the entire chain of decisions.

A young company builds from scratch. It has no operational processes to break, no veteran customers who expect certain conduct, no employees who worked one way for 15 years. The risk built into its model is high anyway, and an added layer of technology risk is standard.

An established business, by contrast, stands on three assets that are hard to quantify: Stable cash flow, customer relationships built over years, and the gut knowledge of a veteran team. Any AI implementation that shakes one of these three, no matter how "innovative", is a business risk, not an opportunity.

The practical conclusion: AI in an established business does not replace what works. It extends it.

Here is an example. A year ago we worked with an importer from China in the Shfela region, 20 years in the field. He received an offer from an AI company for a comprehensive platform at NIS 180,000 that was supposed to "change his entire decision chain". After we went through his use cases, we understood he needed exactly two things: an agent that analyzes currency data and flags windows of opportunity, and an agent that runs an automatic cross check between order and invoice. We built both at a cost of NIS 70,000: 85% of the value, at 40% of the price.

The difference? He did not need a "transformation". He needed two focused tools that extend the working method that had worked for him for 20 years.

5 stages of gradual AI implementation in an established business

This is the heart of the answer. After 18 years of technology implementations, I am committed to one simple framework: An established business implements AI in five stages: no fewer, and not in a different order.

Stage 1: Assessment, identifying the real pain (not the tool)

The first stage does not deal with AI at all. It deals with your business.

In the assessment meeting we ask three questions: Which recurring task "burns" the most hours on the team? Where is there a gap between the knowledge of veteran and new people? What decision do you often make with partial data?

The answers produce the list of use cases. Only after there is a list do we check which tool fits. The opposite (starting from a tool and looking for a problem for it) is the fastest way to burn NIS 100,000 with no result.

Stage 2: A focused pilot: one task, one team, 30 days

After choosing a use case, we start narrow. Not in the whole organization, not in all teams, not in all processes.

One use case. One team of 2 to 4 people. Clear measurement for 30 days. That is it.

Why narrow? Because if the pilot fails (and some pilots fail, that is normal), the damage is limited. If it succeeds, you have internal proof that the team can adopt it. There is no better way to convince veteran employees to adopt a change than testimony from a peer on the team.

Stage 3: Measurement, how to really measure AI ROI

Before the pilot you must write down three numbers: Time spent on the task today, error rate today, and direct cost (hours × wage). These are the baseline.

After 30 days of the pilot, you measure the same three numbers. The difference is the ROI. Not a feeling, not "it looks like it works", not a vendor presentation. Numbers before and after.

If your AI vendor is not willing to commit to measurable numbers, you are on the wrong side of the negotiation.

Stage 4: Scale, moving from pilot to operational routine

Scale is the critical moment. This is where most good implementations fail, not because the tool does not work, but because the move from a pilot (with an enthusiastic team that volunteered) to routine (with other employees who did not volunteer) requires completely different work.

The rule: Before scaling, build a written SOP, train 2 to 3 more employees face to face, and define an "internal support team" of 1 to 2 people who can solve problems without calling the vendor every time.

Stage 5: Organizational culture, how the team adopts it without panic

This is the stage most AI companies do not mention, because it is not their product. But without it, the best technology in the world will stay unused.

Three principles: do not present AI as "it will replace you" (because that creates stubborn resistance); let veteran employees adopt at their own pace, not by force; and celebrate small wins (who saved an hour? who caught an error?) openly.

Want to know how these 5 stages look in your business? Schedule a free AI diagnostic call: 30 minutes, no obligation.

Schedule a call

Or directly: phone 052-331-3334 · WhatsApp

Case study: how an established restaurant chain implemented AI without risking operations

Back to the restaurant chain from the opening. Here is how it looked in practice.

The context: A chain of three branches, 12 years in the market, about 80 employees. The main pain was food waste of 8% of turnover, about NIS 480,000 a year thrown in the trash. Manual inventory management, and demand forecasting based on the gut feeling of the branch managers.

The assessment (weeks 1 to 2): Instead of offering an "AI based restaurant management platform", we sat down with the specific pain. We understood the problem was not in the whole decision chain but at one point: daily raw material ordering for the weekend. On Sunday through Wednesday the manual system worked very well. The gap was on Thursday through Saturday.

The solution: A point AI agent that analyzes sales data from the last 18 months, expected weather and local events (holidays, sports events), and recommends order quantities for Thursday and Saturday.

The implementation (30 day pilot): We started in the smallest of the 3 branches. A team of 2: the branch manager and a chef. The agent's recommendations arrived as a daily email at 08:00, and the manager decided whether to adopt them (with an "approve" or "change" button). The recorded baseline: 8.2% food waste, 14 monthly work hours spent on ordering.

The results after 3 months: Food waste dropped to 6.7% in the pilot branch, an 18% improvement. A saving of 7 management hours a month. A decision to roll out to the whole chain in the scale stage.

The key lesson: We did not start from the tool. We started from the pain: food waste on Thursday through Saturday. When the pain is focused, the solution is focused, the cost is focused, and success is measurable.

(Note: the data is anonymous and rounded for the sake of the example.)

The 3 most common mistakes I see in AI implementation

Mistake 1: starting from the tool instead of the problem

The sentence I hear most: "We bought ChatGPT Enterprise, now how do we use it?" That is exactly backwards. A tool is an answer to a question. If you do not have the question, the tool will not help.

Mistake 2: not involving from the start the team that will actually use the tool

The owner decides. The CTO implements. The employees in the field, who were not involved, quietly sabotage. The tool is not adopted. The project "fails", but the real failure was in the decision process itself.

Mistake 3: expecting magic instead of understanding that AI is a process, not a product

AI is not bought as a package. It is built gradually, with data cleanup, integrations, training and adjustments. Businesses that expect that "we will sign and on Tuesday everything will work" are disappointed.

How to start: 3 questions that will clarify whether you are ready

What is the recurring task my team performs manually that "burns" the most hours?
Who on my team is most curious to learn new tools (not the most available, the most curious)?
Do I have one KPI I can measure before and after the pilot?

If you answered 2 out of 3, you are ready for an assessment.

At Forward, since 2008, we have helped established businesses in 6 different industries (restaurants, importing from China, medical devices, real estate, business development in Europe and the Emirates, and education) make this move without breaking what works.

Schedule an AI diagnostic call: 30 minutes, free, no obligation. Together we will understand where the pain is, mark a first use case, and make sure you are ready for a pilot before you spend a shekel.