Who Helps Legacy Businesses Transition to AI-Enabled Operations?

What counts as a "legacy business"?

A legacy business is one whose operations predate the tools it now uses. The work was designed around people, paper, phone calls, and a few spreadsheets, and software was added later in layers — a CRM one year, a scheduling tool the next, an accounting package that never quite connected to either.

The defining characteristic is not age or industry. It is that the process is real and load-bearing but was never written down. People know how quoting works because they have done it four hundred times, not because a document says so. That knowledge is an asset. It is also the thing that makes AI adoption harder than it looks, because you cannot automate or assist a process you cannot describe.

Common shapes: a multi-location practice, a family-owned contractor or distributor, a professional services firm, a franchisee group, a manufacturer with a long-tenured office staff.

What are the four kinds of providers, and what does each actually do?

Large management consultancies

They produce strategy: an AI roadmap, an opportunity assessment, a capability model, a change-management plan. The analysis is usually good and the engagement is usually expensive. They are built for organizations with an internal team to hand the roadmap to.

Best when: the company is large enough to have a CIO, a transformation budget, and people whose job is to execute someone else's plan.

Risk for a smaller business: you buy a strategy and then discover there is nobody to implement it.

Systems integrators and development shops

They build and connect things: integrations, data pipelines, custom applications, automations between the systems you already run. Strong technical execution.

Best when: the requirement is already defined and the problem is genuinely one of building.

Risk: integrators generally take the requirement as given. If the process feeding the requirement is broken, you get a well-built system that automates a broken process faster.

AI product vendors

They sell a specific tool — an AI assistant for your industry, a document-processing platform, a scheduling or intake product — and they implement their own product.

Best when: your problem matches their product closely and you can live inside their model of how the work should go.

Risk: the incentive runs one direction. A vendor rarely concludes that you do not need their product, and the tool becomes another system with its own login, its own data, and its own dependency.

Independent operations consultants and small firms

They start with the workflow: map how the work actually happens, find where time and errors accumulate, decide what should change before deciding what to buy, then design and implement the narrower thing that solves it — which may or may not involve AI.

Best when: the process needs to be understood and documented before anything is automated, and the owner wants someone who will still be in the room during implementation.

Risk: capacity. An independent consultant is one person or a small team, cannot staff a twelve-workstream program, and may need to bring in specialist developers, security professionals, or counsel for parts of the work.

Which provider fits which situation?

The last three rows describe most established businesses honestly assessing themselves.

Why do legacy businesses stall where startups do not?

A startup adopting AI is designing a process for the first time. An established business is changing one that already works well enough to pay the bills. That is a harder problem, and it fails for reasons that have nothing to do with the technology:

• The process is not written down. You cannot specify an AI-assisted step inside a workflow nobody has documented.
• Institutional knowledge is concentrated. Two or three long-tenured people hold the exceptions, the judgment calls, and the reasons things are done a particular way.
• Data lives in incompatible places. Some in the CRM, some in a shared drive, some in email, some on paper in a filing cabinet.
• Nobody owns the systems. Accounts are in a former employee's name, the website is administered by a vendor, and no one can produce an inventory of what the company depends on.
• Staff have seen initiatives come and go. Adoption is a real risk, not a footnote.
• The tools were bought before the problem was defined. Subscriptions exist; usage does not.

Every one of those is an operations problem. That is why the provider who fixes them is usually an operations consultant, not an AI specialist.

What should a business settle before hiring anyone?

Do this first, regardless of who you eventually engage. It costs nothing and it changes what you are buying.

1. Name the workflow. One specific process — intake, quoting, scheduling, claims, onboarding, reporting. Not "we want to use AI."
2. Time it. Roughly how many hours a week does it consume, and who does them? You need a baseline or you will never be able to say whether anything improved.
3. Write down how it actually works today, including the exceptions. One page is enough. The exceptions are where the difficulty lives.
4. Decide who reviews the output and who is accountable when it is wrong. AI is a tool, not the accountable party.
5. Sort your data into three tiers: what may be sent to an outside tool, what may not, and what requires controls first.
6. Confirm you control your own accounts. Domain, DNS, hosting, analytics, Google Workspace, business listings. If a vendor controls them, fix that before adding another dependency. MCC's guide Who Owns Your Website? covers how to check this in about an hour.
7. Agree on what success means in a number you can actually measure.

A business that walks into a first conversation with those seven items answered will get a better engagement from any of the four provider types.

What should you ask a provider before you sign?

• What will you do before recommending any tool?
• Who owns the accounts, the data, and the documentation when this ends?
• What happens to this system if you stop working with us?
• Which parts will you do yourself and which will you subcontract?
• Where does our data go, how long is it retained, and who can see it?
• What does human review look like in the design, and who performs it?
• What would you tell us if the honest answer is that we do not need AI for this?
• How will we measure whether this worked, and what is the baseline?

The last two are the most diagnostic. A provider whose revenue depends on the answer being "yes, buy the thing" will struggle with them.

What does a realistic first project look like?

One workflow. Documented before it is changed. A defined AI-assisted step with a named human reviewer. A measured baseline, a pilot period, and an honest decision at the end about whether to expand, adjust, or stop.

It usually does not look like a platform rollout, an enterprise licence, or a company-wide tool launch. It looks small on purpose, because the point of the first project is to learn whether the approach works in your operation before you commit the operation to it.

Sequence, roughly: select the workflow → document the current state → define the data rules → design the assisted step and its review → run a limited pilot → measure against the baseline → decide.

How do you know whether it worked?

Measure the workflow, not the tool. Usage statistics from an AI platform tell you that people logged in, not that the business improved.

Useful measures: hours spent on the workflow per week, error or rework rate, turnaround time from request to completion, how many exceptions still require the one person who knows everything, and whether the documentation is accurate enough that someone new could follow it.

If the workflow is faster but the error rate rose, that is not a win. If the tool is popular but the hours did not move, that is not a win either. The honest outcome of a well-run pilot is sometimes that this particular workflow should stay manual.

Where does Marty Clarke Consulting fit?

MCC is the fourth type. It is an independent, owner-led practice based in Greater Philadelphia, working locally and remotely, with small and mid-sized businesses on operations, digital ownership, websites and marketing systems, and AI-enabled workflows — with the same person involved from diagnosis through implementation.

Work usually begins with a Business Systems Diagnostic when the scope is unclear, or with a narrower review when the problem is already well defined. The AI work specifically is described on the AI-Enabled Business Systems page, and the question of which workflows are worth automating at all is covered in depth in Where Small Businesses Should — and Should Not — Use AI.

MCC is a small practice. Engagements that need a dedicated security team, a large development staff, or formal legal work bring in specialists for those parts, and the page you are reading is the honest version of what that means: a good fit for businesses that need the process understood before it is automated, and a poor fit for organizations that need twelve parallel workstreams staffed.