Short answer
Twelve workflows show up repeatedly in small companies that automate well: onboarding, lead generation, invoicing, reviews, cash flow, reporting, documents, contracts, HR admin, marketing, customer service, and internal knowledge search. None of them is optional reading here — each has its own article covering the mechanism, the failure modes, and where the line to a human has to stay. This page exists to show how they fit together, which ones touch revenue directly, and the order that decides whether the first one earns you permission to build a second.
What this list is for
Most lists of "AI workflows to automate" are written to be impressive, not useful — twenty ideas with no relationship to each other, in no particular order, aimed at a company that could plausibly deploy one a month for two years. That is not how this works in a 10 to 100 person company. Three or four workflows running well, watched by someone, beats twelve running unattended and unmonitored.
So this page does something narrower. It groups the twelve workflows that actually recur across our own engagements by what they touch — revenue, back office, or the judgment-heavy work that needs a human checking it — and links each one to the article that covers how to build it without the failure modes that make automation projects quietly expensive. Nothing on this page reexplains a mechanism that already has its own article; it tells you which one to read and why it matters before you open it.
How to read the table below
The table is not a menu where every row is equally worth building. It is a shortlist, already filtered down from the much longer list of things a model could technically do. What earned each of these twelve a place is the same four-part test we use to score any candidate — frequency, predictability, cost of an undetected error, and visibility of the output — covered in full in how to decide what to automate first. If a process in your company does not clear that bar, it does not belong on this list either, no matter how close it looks to one of the twelve.
The third column is the signal, not a instruction — it describes what it typically looks like inside a company when that workflow has become worth building, so you can check your own operation against it rather than against a stage of company size or headcount.
Twelve workflows, and the signal for each
| Workflow | What it automates | Signal it is the right time |
|---|---|---|
| Client onboarding | Account setup, welcome sequence, first-meeting scheduling | Sales quotes a start date the team cannot reliably hit |
| Lead generation | Prospect list building, first-touch outreach, replies routed to a human | A rep spends more hours prospecting than talking to qualified people |
| Invoicing and payment reminders | Invoice generation, chase sequence, escalation to a human | Late payment gets chased by whoever notices first, not on a schedule |
| Review management | Review requests, monitoring, drafting (never posting) responses | Reviews arrive faster than anyone reads them |
| Cash flow forecasting | Rolling near-term forecast built from live transaction data | The forecast is a spreadsheet nobody has opened since last month |
| Automated reporting | Pulling numbers from every tool into one recurring report | Someone loses a morning a week copying numbers into a deck |
| Document management | Classification, extraction, routing, and archiving | A document takes longer to find than it took to read the first time |
| Contract management | Extracting renewal and obligation dates, alerting before the deadline | A renewal was missed and nobody knew until the invoice arrived |
| HR administration | Onboarding paperwork, leave requests, policy questions | HR answers the same handful of questions by hand every week |
| Digital marketing | Content drafts, scheduling, campaign reporting across channels | Marketing spends more time reporting on campaigns than running them |
| Customer service chatbot | First-line resolution and escalation for repeat questions | One question type keeps resurfacing at the top of the queue |
| Internal knowledge search | Answering questions from company documents in plain language | Someone asks a colleague something the handbook already answers |
Workflows that touch revenue directly
Four of the twelve sit close enough to revenue that a mistake is felt immediately, by a client or by the bank balance — which is exactly why they need the tightest checking, not the loosest.
- Client onboarding is the only process most clients watch happen in real time, which makes it the wrong place to optimize for completion rate over how the client actually experienced it. Automating onboarding without losing the client covers the ordering problem and the one question you must never ask a client twice.
- Lead generation is the workflow most likely to backfire, because automation multiplies whatever targeting already sits underneath it — bad targeting included. The safe sequencing, and how to avoid burning a sending domain in the first month, is in automating lead generation without burning your domain.
- Invoicing and payment reminders is usually a process problem wearing a client problem's clothes. Automating the chase covers how to keep it firm without the tone that damages a relationship over a late payment that was never intentional.
- Review management now feeds a second audience beyond prospective customers: it is a citation source that AI answer engines pull from directly. What to automate, what never to automate, and where the legal line sits is covered in automating review management without breaking FTC rules.
Back-office and finance workflows
Five of the twelve never touch a client directly, which is why they are usually the easiest to deploy — and the easiest to leave unmonitored, because nobody outside the company notices when they quietly stop working.
- Cash flow forecasting moves from a static monthly snapshot to a forecast that refreshes as transactions land, which is the part that actually changes the decision it supports. What actually changes, and what still needs you covers the shift in full.
- Automated reporting replaces the recurring morning of copying numbers from four tools into one deck. The stack that actually works for a small business covers what to connect first and what a working setup costs to build.
- Document management turns a folder structure nobody trusts into something that answers a plain-language question. What actually gets automated draws the line between classification and archiving, and the retrieval work covered separately below.
- Contract management is not about finding a contract — it is about a renewal date that passed unnoticed. Turning documents into dates covers why the alert has to be calculated backwards from the deadline, not forwards from signing.
- HR administration splits cleanly into an administrative half that is safe to automate and a decision half that changes the rules the moment a system starts ranking people. What to automate, and what never to covers exactly where that line sits.
Marketing, support and judgment-heavy workflows
The last three sit closest to the "assist, do not automate" box: a model produces a draft or an answer, and a human still decides or is still one escalation away.
- Digital marketing touches email, content, social and reporting all at once, which is precisely why deploying it everywhere in month one is the mistake. Where to start, and what to skip covers the channel to automate first and the one that costs a brand voice if you get it wrong.
- A customer service chatbot that cannot resolve or escalate is a queue with extra steps, not an improvement on one. Where chatbots help and where they hurt covers how to scope one so it deflects real work instead of manufacturing new tickets.
- Internal knowledge search — retrieval-augmented generation over your own documents — answers the question a colleague currently answers from memory. The hard part is never the model; it is the corpus, as covered in making an assistant answer from your files.
What holds the twelve together
None of these twelve is a single algorithm bolted onto a process. Each one is a system: a deterministic backbone that moves data and triggers steps on schedule, and a model doing the narrower part that needs to read or write in plain language. Where a workflow crosses from executing a rule into making a judgment call — approving a refund, deciding which lead to prioritize — the failure modes change, and that line is covered on its own in rules, judgment, and the line between.
Most of the twelve can be built without a developer, on a no-code orchestration tool wired to a model API. What actually survives contact with production once that tool is chosen — and what quietly does not — is the subject of business process automation without developers. If the tool underneath most of these is n8n specifically, this practical guide covers when self-hosting earns its keep and which decisions are hard to reverse once workflows are running on it.
The order that decides whether it sticks
Deploying is the easy part. What decides whether a company still trusts automation a year from now is the order these twelve get built in, and what happens after the first one ships.
Pick the first workflow for what it proves, not for its size on paper. It has to run a full month without babysitting, fail loudly when it breaks, and move a number you recorded before you started — the full method for choosing and scoring that first candidate is in deciding what to automate first.
Before building any of the twelve, it is worth reading the seven decisions — not bugs — that sink most automation projects, most of which have nothing to do with the technology: AI automation mistakes to avoid. Chief among them is skipping measurement: decide the metric before you build, not after, and use the method that survives a finance review rather than a headline number, covered in how to measure the ROI of AI automation honestly. For the checklist that turns a chosen workflow into a shipped one — scope, data rules, pilot, review, in the order that decides whether it survives — see the AI implementation checklist small businesses actually need.
Frequently asked questions
What counts as an AI automation workflow, and what does not?
A workflow qualifies when a model is doing something a fixed script cannot — reading unstructured input, drafting language, or making a narrow judgment call — inside a process that otherwise runs the same way every time. A nightly export or a scheduled email is automation, but it is not an AI workflow; nothing in it decides anything. Most of the twelve on this page combine both: a deterministic backbone that moves data and triggers steps, with a model doing the one part that needs to read or write in plain language.
Which of these should a small business deploy first?
Not the one that sounds most impressive. The right first workflow is the one that runs often, follows a predictable path, and produces an output someone would notice was wrong — the criteria are the same regardless of which of the twelve you are scoring, and are covered in full in our article on deciding what to automate first. In practice that is usually invoicing, onboarding, or reporting, because all three happen weekly at minimum and their failure is visible immediately.
Do all twelve of these need a developer to build?
No. Most of them run on no-code orchestration tools connected to a model API, and can be assembled by someone who understands the process without writing code. A few — cash flow forecasting with a genuine predictive model, or a document search system with real retrieval — benefit from technical help to get the data layer right, even though the day-to-day workflow still runs through the same no-code tool as the rest.
Is there a point where deploying more workflows stops being worth it?
Yes, and it arrives earlier than most companies expect — a rule of thumb from LYVIA's own engagements, not a published benchmark, but a consistent pattern: once three or four workflows are running, the bottleneck stops being which process to automate next and becomes whether anyone is watching the ones already live. A workflow nobody monitors degrades quietly. Deploying a twelfth workflow while the third one has been silently failing for a month is a net loss even though the count went up.
If you want help deciding which of these twelve applies to your company, and in what order, that is the first conversation we usually have with a new client. Book a call.
