The 42 AI workflow patterns we’ve shipped, scoped, or learned from.

Reads incoming customer questions, answers the most common 60-70% of them automatically using grounded responses from the client's knowledge content, and routes anything uncertain to a human with the context already attached. The customer feels supported faster, the support team's queue stays focused on the genuinely hard questions, and answers stay accurate because the pattern grounds every response on real client documentation rather than its training data.

Sits inside the support agent's interface and drafts replies for complex or research-heavy tickets. Reads the incoming ticket, searches past resolved tickets and internal documentation for precedent, and produces a draft response the human agent edits and sends. Never sends autonomously. The pattern accelerates skilled agents on hard cases rather than handling easy cases without them, which is why it shows up in firms where the support workload is complex by nature and full deflection isn't realistic.

A translation layer wrapped around the support workflow. Incoming messages in any supported language are translated for the agent's working interface, the agent replies in their native language, and the outgoing message is translated back to the customer's language before sending. The pattern preserves tone and product-specific terminology by maintaining glossaries the agent or the team can edit. Used by firms with international customers but a single-language support team.

Listens to or processes recorded customer-facing calls, produces accurate transcripts, and generates structured summaries: who said what, what was agreed, what action items emerged, what the sentiment was, and what topics came up. The output flows into the CRM, the support system, or wherever calls are normally documented. Typical use: sales calls, support calls, customer success check-ins. The pattern's job is to make voice conversations searchable and actionable without requiring humans to write notes in real time.

Generates outbound emails that feel personally written but are produced from structured inputs about the recipient and the sender's goal. Common uses: sales outreach to a defined account list, customer success check-ins on a cadence, milestone messages tied to product or contract events. The pattern grounds each email on real facts about the recipient (their role, their company's recent activity, their account state) rather than generating generic copy. Every email is human-reviewable before send, and the pattern is deliberately constrained to one email at a time per recipient — no automated multi-step sequences without human checkpoint.

Watches a shared inbox or distribution alias and classifies each incoming message: what kind of request it is, how urgent, who should own it, what (if anything) the sender wants. Routes the message to the right destination — a person, a team channel, a ticket queue — with a one-paragraph summary so the recipient doesn't have to re-read the original to know what's needed. The pattern's value comes from compressing the human triage step (currently done by someone scanning the inbox several times a day) into something that happens within seconds of arrival.

Given a deal in progress, the pattern drafts the first version of a proposal or quote: scope description, pricing pulled from the rate card or product catalog, terms drawn from approved templates, and any context the sales rep has captured in their notes. Sales rep edits and finalizes. The pattern compresses the 2-4 hours typically spent on proposal first drafts into something closer to ten minutes of review, and reduces the time deals sit waiting on a document the rep hasn't gotten to yet.

Continuously cleans and enriches the records inside the CRM: standardizes formats, deduplicates contacts and companies, fills missing fields from public sources, flags stale records, and maintains relationship links (who works where, who reports to whom) as people move. The pattern's job is to keep the CRM trustworthy without burdening reps with data entry. SMBs with neglected CRMs typically see the system flip from a frustration to a useful tool within 60 days of this pattern going live.

Watches activity across the systems where leads show up — website visits, content downloads, email opens, support inquiries, product trial signups — and produces a continuously updated score per lead: how likely they are to be a fit, how engaged they are, what they appear to be interested in. Surfaces high-scoring leads to sales in priority order and flags accounts where engagement patterns suggest active buying intent. Replaces the gut-feel triage of marketing-qualified leads with something measurable.

Before a scheduled customer or prospect meeting, generates a concise briefing document for the person taking the meeting: who's on the other side, what their company has been doing recently, what the relationship history with them looks like, what was discussed in prior meetings, what's currently outstanding. Replaces the 15-20 minutes of frantic research that most reps either do badly or skip entirely. Lands in the rep's morning email or calendar event so they read it before the meeting, not during it.

Continuously watches signals across the customer base — usage patterns, support ticket trends, NPS responses, payment behavior, key contact changes — and flags accounts where renewal risk is rising. Outputs a ranked list to the customer success team with the evidence behind each flag, so the team can intervene early rather than discovering the problem 30 days before renewal. The pattern's value comes from catching the slow churn signals (declining logins, champion leaving) months before they show up in renewal conversations.

For firms that respond to formal RFPs, RFIs, or tenders, the pattern turns a 200-question vendor questionnaire into a first-pass completed response in hours rather than days. Works by maintaining a structured library of approved answers, security artifacts, and case studies, then matching incoming RFP questions to the right answer with appropriate customization. A human reviewer goes through the result before submission. Compresses the most time-consuming part of competing for procurement-led deals.

A single search interface that lets anyone in the company ask questions and get answers drawn from across the firm's knowledge sources — wikis, shared drives, past project archives, Slack threads, ticket history. Returns a synthesized answer with citations linking back to source documents. The pattern's value is less about finding documents and more about finding the answer across documents that don't always sit in the same system. Replaces the typical 'I know we wrote that down somewhere' expedition with something that takes ten seconds.

A guided assistant for new hires or for any employee learning a new role. Walks through the firm's onboarding curriculum at the user's pace, answers questions as they come up grounded on company documentation, tracks progress, and flags topics where the user seems stuck. Sits alongside the structured onboarding plan rather than replacing it — handles the 'where do I find that?' and 'what does this acronym mean?' load that otherwise falls on managers and buddies. Particularly valuable in firms hiring at any meaningful rate, where each new joiner currently absorbs time from senior staff.

Captures internal meetings — recorded or transcribed in real time — and produces structured notes: who attended, what was discussed, what was decided, what action items emerged with owners, what was explicitly deferred. Different from A4 (which is customer-facing calls): this pattern is tuned for internal meetings and feeds an organizational decision archive that becomes searchable over time. The pattern's compounding value: in two years, anyone can ask 'why did we decide X?' and find the meeting where the decision happened.

Observes how people actually do their work — by capturing screen recordings, step-by-step walkthroughs, or asking the person to narrate what they're doing — and generates a draft Standard Operating Procedure that documents the process. The drafted SOP goes to a human for review and approval before becoming the official version. Particularly valuable for firms where senior operators have undocumented tribal knowledge: the pattern lets you capture that knowledge in hours rather than the days it would take someone to write the SOP from scratch.

A focused Q&A assistant grounded specifically on the firm's policies, contracts, and compliance documentation. Answers questions like 'what are our payment terms with vendor X', 'what's our PTO policy', 'what does our MSA say about IP'. Different from C7 (general knowledge search) in that it's tuned for the precision and citation discipline that policy and contract questions require: every answer must cite the specific clause, with the exact text visible. The pattern's value is taking the 'I have to ask legal' and 'I have to read the contract' load off the team.

Reads structured and semi-structured documents that arrive in unstructured formats — invoices, purchase orders, bills of lading, claim forms, contracts, scanned paperwork — and extracts the fields the business cares about into structured records. Routes each extracted record through validation (does it match what we expected, do the totals add up, is the supplier known), then delivers to the system of record. High-confidence extractions land directly; low-confidence ones route to a human reviewer with the source document and the proposed values shown side by side. The pattern absorbs the data-entry work that currently consumes hours of skilled time every day in operations-heavy SMBs.

Sits between an inbound stream of work items — orders, service tickets, requests, repair jobs — and the teams that handle them. For each item, the pattern reads the content, classifies it against the firm's working categories, predicts urgency and skill required, identifies any special handling needed (VIP customer, regulatory case, known-difficult product), and assigns to the right queue or person. The pattern's value is taking the manual triage step — usually done by a dispatcher or team lead — and making it instant and consistent. Different from A6 (which triages messages) because this pattern handles operational work items that drive downstream activity, not just communication.

Looks at images or video of physical items moving through a production, packing, or fulfillment process, and flags items that fail the firm's quality criteria: damaged packaging, missing components, incorrect labels, manufacturing defects. Sits alongside human QC rather than replacing it: the pattern handles the high-volume routine cases so humans can focus on the genuinely ambiguous ones. Different shape from the other D patterns because it processes images, not text or structured records, and because the kill metrics are different (false negatives let bad product through; false positives slow throughput). Used in light manufacturing, fulfillment, and food production.

For firms with field workers, technicians, or mobile service teams, optimizes the daily schedule: which jobs go to which workers, in which order, accounting for skills, travel time, customer windows, priority, and any constraints (e.g. some jobs need two people, some require specific equipment). Replans during the day as new jobs come in, jobs run long, or workers become unavailable. The pattern's value is replacing a dispatcher's manual scheduling — which works fine for a few jobs but breaks under volume — with continuous optimization. Different from D13 (which classifies and routes inbound work) because this pattern actively constructs and modifies schedules.

Predicts demand for inventory items at a useful granularity — per SKU, per location, per week or per day depending on the business — and surfaces reorder recommendations: what to buy, how much, when. Accounts for seasonality, trend, promotions, weather, known disruptions. The pattern's value is replacing the rule-of-thumb reordering most SMBs use (which over-stocks slow movers and under-stocks fast movers) with something that learns from actual sales patterns. Output is a recommendation, not an automated purchase: humans approve buying decisions and the pattern improves over time as it sees what's bought and what sells.

Continuously checks activity across systems against the firm's compliance and policy rules: are expense claims within policy, are PR submissions following the approval workflow, are customer communications meeting regulatory requirements, are access grants getting proper review. Flags potential violations as they happen rather than waiting for periodic audits. Routes flagged items to the right reviewer with the evidence. The pattern's value is shifting compliance from a quarterly cleanup exercise to an always-on signal: violations get caught and fixed within hours, not months, and the audit story becomes much simpler.

Continuously watches signals about the firm's vendors and suppliers — financial news, public filings, security incidents, certification expirations, delivery performance, payment behavior on their side — and flags rising risks before they become operational problems. Different from B5 (which monitors customer churn risk) in that it watches the supply side and the risks are different: supplier going bankrupt, security cert expiring, key vendor being acquired, delivery quality degrading. The pattern's value is moving vendor risk from an annual review exercise to a continuous signal, so issues get addressed during the relationship rather than discovered when they cause downstream impact.

Reads transactions from the firm's bank and credit card feeds, matches each one against the bookkeeping system's open items (open invoices, expected receipts, recorded bills), and categorizes the unmatched ones to the right GL account based on description, amount, vendor pattern, and historical categorization. Surfaces the small set of genuinely ambiguous transactions for a human to decide. Replaces the slow, manual back-and-forth between bank statements and the books that bookkeepers traditionally do, and shifts the work from data-entry to exception handling. Builds the categorization model from the firm's own historical books, not a generic chart.

Watches the firm's accounts receivable, identifies invoices coming due or already overdue, and drives the follow-up process: drafts collection emails at the right tone for the right stage, schedules calls for high-value or delinquent accounts, notifies sales when customer relationships are at risk, and tracks every touch in a single timeline. Different from generic dunning automation because it adapts to the firm's tone, the relationship with each customer (don't dun a customer in active expansion talks), and history (some customers always pay on day 45). The pattern's value is consistent, timely AR follow-through without burning the controller's time or annoying customers who pay reliably anyway.

Reads expense submissions — receipts uploaded by employees, card transactions to be coded, mileage logs — and produces complete, policy-checked expense entries with the right GL category, project, and approval routing. Catches policy violations (out-of-policy meal, missing receipt, duplicate submission) at submission time rather than during approval. Routes flagged items back to the submitter with specifics. Different from E19 (bank reconciliation) because this pattern works with employee-driven submissions and policy enforcement; different from D12 (document extraction) because the document type and downstream workflow are narrower and more standardized. The pattern's value is compressing the expense cycle from days to hours while reducing the controller's role in chasing down corrections.

Takes the numbers from the monthly or quarterly close — P&L, balance sheet, cash flow, KPI dashboards — and writes the narrative that goes around them: what moved this period, why it moved, what's worth flagging to leadership. The pattern reads not just the numbers but the supporting context (sales pipeline, headcount changes, project milestones) to explain the why behind the what. Drafts go to the finance lead for review and editing before going to the audience. Replaces the slow, manual writeup that delays monthly reporting and often gets compressed at month-end into something less thoughtful than it should be.

Maintains a rolling cash flow forecast — 13-week or longer — that updates daily based on what's actually happening: AR aging changes, AP commitments, payroll runs, expected revenue from pipeline. Surfaces what's coming, what could go wrong, and what scenarios would matter. Different from E22 (which writes commentary on past periods) because this looks forward and updates continuously. The pattern's value is moving cash forecasting from a quarterly board exercise to an always-on view: the CFO or controller sees liquidity risks weeks before they'd otherwise notice, and the team can answer 'can we hire' or 'should we accelerate this payment' with current information rather than month-old spreadsheets.

For roles with high application volume, reads each resume against a structured definition of what the role actually requires, scores the fit, and produces a shortlist with reasoning for each ranking. Different from generic ATS scoring because the pattern works against an explicit role specification co-authored with the hiring manager, not against vague keyword matching, and because every shortlist comes with the why — the specific signals that put each candidate where they are. Pre-screens for disparate impact at the cohort level and refuses to make 'hire/no-hire' calls; final decisions stay with humans. The pattern's value is letting hiring teams give every applicant a thoughtful first look, even when volume would normally make that impossible.

Handles the operational work of moving candidates through the interview pipeline: finds interview slots that work for everyone, schedules and reschedules as needed, sends clear and timely communications to candidates at every stage, drafts personalized rejection notes when needed. Different from generic scheduling tools because the pattern keeps the candidate experience tone consistent with the firm's brand and adapts to the role and stage. Reduces the operational drag of hiring that often falls on already-busy recruiters or hiring managers.

A focused assistant employees can ask about HR policies, benefits, time off, expense rules, conduct policies — anything in the employee handbook and surrounding policy documents. Different from C7 (general internal search) and C11 (legal/contract Q&A) in that it's tuned for employee-facing questions: it knows when to give a direct answer and when to route to a People Ops person, it handles sensitive questions (compensation, performance issues, harassment) with appropriate care, and it gives benefit-specific answers that depend on the asker's plan, country, or tenure. The pattern's value is offloading the predictable 70% of HR questions from People Ops while making sure the other 30% reaches a human safely.

Helps managers write structured performance reviews by synthesizing the evidence already captured throughout the period: project outcomes, peer feedback, 1:1 notes, goals progress. Drafts a first version of each review section the manager edits to their voice and judgment. Catches common pitfalls (one-incident bias, recency effect, vague language, missing examples) and prompts the manager to add specifics. Never makes the rating decision — that stays entirely with the human. The pattern's value is reducing the slog of review-writing while improving review quality, particularly for managers with many direct reports who would otherwise write thin reviews under time pressure.

An end-to-end pipeline that takes a content brief — topic, audience, format, tone — and moves it through research, drafting, internal review, editorial polish, and publication. Different from a tool that just generates text in that it works from the firm's actual content principles, references the firm's own prior writing and research, routes drafts to the right humans at the right stages, and never publishes anything without explicit human approval. The pattern's value is compressing the time from brief to published from weeks to days while keeping the firm's voice and quality bar intact.

Continuously analyzes what people search for in the firm's space, what the firm has published, and what the competition publishes — and surfaces the highest-value gaps: topics with real search demand the firm doesn't cover, existing content losing rank, opportunities to consolidate underperforming pieces. Different from generic SEO tools because the pattern works against the firm's specific commercial priorities (which topics actually drive qualified pipeline) rather than just search volume. Output is a prioritized list of content actions: write this, update that, retire this old piece. Humans make decisions; the pattern does the analysis.

For firms running paid acquisition, generates ad copy variants for different audience segments, channels, and angles — then watches performance and continuously feeds back into what works. Different from generic ad-generators in that the pattern grounds variants on the firm's actual messaging principles and approved claims library, refuses to make claims it can't substantiate, and tracks performance back to inform future variants. Humans approve every variant before it goes live. The pattern's value is letting marketing teams test more angles more often without each test consuming hours of copywriting time.

Pulls product feedback from everywhere it gets mentioned — support tickets, sales call transcripts, customer success notes, public reviews, social channels — and synthesizes it into structured themes the product team can act on. Different from running a survey: this pattern works on unsolicited feedback in its natural form, identifying signals across thousands of mentions that no human would catch by reading individually. Output is themed feedback with weight (how many sources, how recent), severity (how badly does it block customers), and direct links to specific source mentions. Humans interpret and decide what to do with it; the pattern surfaces what's worth interpreting.

Continuously watches where the firm and its products get mentioned across public sources — news, social, review sites, podcasts, industry forums — classifies each mention by sentiment, topic, and influence, and surfaces what needs attention. Different from a generic mention-tracker because the pattern weights mentions by source authority and audience reach, distinguishes real reputational signals from noise, and routes the few mentions that genuinely need response to the right team. The pattern's value is ensuring that the firm sees what's being said about it without anyone needing to scroll through hundreds of low-signal mentions to find the meaningful ones.

Lets non-analysts ask data questions in plain English and get answers from the firm's data warehouse or reporting database. Translates the question into a query, runs it, and presents the result with the underlying numbers and the SQL behind them. Different from generic text-to-SQL because the pattern works against the firm's actual data model with documented semantics — it knows what 'active customer' means in your business, not what 'active customer' might generically mean. Refuses to answer questions it can't ground in the schema, surfacing uncertainty rather than guessing. The pattern's value is collapsing the 'I'll ask the data team and hear back next week' loop into something that happens in the meeting where the question came up.

Continuously analyzes the firm's metrics for changes worth noticing — unusual movements, emerging trends, anomalies, segments performing differently than expected — and pushes a digest of what matters to the people who should care. Different from a dashboard because the pattern decides what's worth surfacing rather than letting humans scan everything; different from generic anomaly detection because the pattern speaks in business language ('your premium segment churn is rising while standard is stable') rather than statistical jargon. The pattern's value is replacing the 'I look at the dashboard every Monday and hope something jumps out' habit with proactive flagging of what's actually moving.

When a leader is weighing a decision — should we open a new region, raise prices on this tier, hire ahead of plan, kill this product line — the pattern builds a structured model of the decision and lets the leader explore scenarios: what assumptions would have to be true for this to work, what does sensitivity look like, what's the expected vs. downside-case impact. Doesn't make the decision; structures the thinking. Different from generic forecasting because the model is constructed per decision rather than from a fixed template, and integrates with the firm's actual data so scenarios start from reality. The pattern's value is replacing the slow consultant-style analysis cycle with something a leader can iterate on in an afternoon.

Takes a request — formal ('weekly board report'), informal ('I need to see how this region is doing') — and produces a tailored report or dashboard: the right metrics, the right time periods, the right comparisons, the right visualizations, in the format the audience prefers. Different from H33 (one-off questions) and H34 (pushed insights) in that this pattern is for recurring or substantial information products: things that someone would otherwise spend hours constructing in a BI tool. The pattern produces drafts the analyst or report owner refines, then schedules and delivers them on cadence. The value is the difference between everyone reusing the same five generic dashboards and everyone getting reports actually tuned to what they need to see.