Common AI Pattern Combinations by Department

Every department asks the same question: where do we start with AI?

The wrong answer is "everywhere at once." The right answer depends on which workflows generate the most volume, which data you already have, and how much process disruption your team can absorb in year one. McKinsey's State of AI 2025 survey, covering 1,993 organizations across 105 countries, found that IT and marketing and sales remain the two functions with the highest AI adoption rates, though knowledge management is rapidly closing the gap. The pattern is consistent: functions with high data volume and repetitive decision cycles get there first.

This article maps the most proven pattern combinations for seven business functions, along with the sequencing that gets you to a real win in 60-90 days rather than a six-month pilot that produces a deck.

How to read this article

Each section covers:

• Primary combination: which patterns work together in this function and why
• Data prerequisites: what you need to have in place before each pattern can run
• Sequencing: which pattern to start with, and what to add next
• First-win milestone: the specific metric that proves value early

Skip to your department.

Key Facts: Department AI Adoption

• Sales and marketing leads all functions in AI adoption, with about 20% of sales activities already automatable using current tools, per McKinsey's State of AI 2025 survey.
• HR teams using AI for hiring achieve a 75% reduction in time-to-hire, with Unilever's AI-driven platform producing a 50% reduction in time-to-fill and a 16% increase in new-hire diversity.
• AI in finance operations achieves a 90% increase in processing accuracy and 70% reduction in processing time for structured document tasks like invoice and loan processing. (PwC AI Predictions, 2026)

---

Sales

Primary combination: Meeting Intelligence + Scoring and Routing + Workflow Copilot + RAG Assistant

Why these together: Sales is a high-cadence function with well-defined inputs (calls, leads, emails) and measurable outputs (pipeline, close rate, cycle time). The data is mostly already being generated. Meeting Intelligence turns call recordings into CRM notes automatically. Scoring and Routing ensures reps work the right leads in the right order. Workflow Copilot gives reps the next-best action at the moment they open a deal record. RAG (Retrieval-Augmented Generation) Assistant answers product and competitive questions without the rep needing to leave the CRM. This exact combination is what makes an AI Sales Operator work at the role level.

Data prerequisites:

• Meeting Intelligence: call recordings stored and accessible via API; meeting metadata (participants, deal association)
• Scoring and Routing: minimum 6 months of historical CRM data with outcome labels (closed-won, closed-lost, qualified, disqualified)
• Workflow Copilot: active CRM integration with real-time context read; Meeting Intelligence must be producing structured outputs
• RAG Assistant: maintained product documentation, battlecards, and objection-handling guides in a vector database

Sequencing recommendation: Start with Meeting Intelligence. It has the lowest risk (no outbound actions, no routing decisions), the fastest time-to-value (reps see better call notes in week 1), and it builds the structured data that makes Workflow Copilot genuinely useful. Add Scoring and Routing when pipeline volume makes manual triage impractical. Layer in Workflow Copilot after Meeting Intelligence has been running for 30 days and the CRM context is rich enough to generate relevant suggestions. RAG Assistant can run in parallel from the start. The AI Sales Ops implementation roadmap walks through this exact sequencing with week-by-week milestones.

First-win milestone: CRM update rate improvement within 30 days.

"Sales teams that deploy Meeting Intelligence as their first AI pattern see CRM field completion rates rise from under 50% to 85-90% within 30 days, because the notes write themselves. That single metric improvement compresses the ROI argument for every subsequent pattern in the stack." (Rework Sales AI Implementation Data, 2026)

If your reps are currently updating CRM fields on 40% of calls, Meeting Intelligence should move that to 85% or higher. That's a measurable, time-bounded, operationally significant win that builds the case for the next pattern.

---

Customer Support

Primary combination: RAG Assistant + Scoring and Routing + Workflow Copilot + Autonomous Agent

Why these together: Support is high-volume with a predictable distribution: most tickets are variations of a small number of issues. RAG ensures agents always have access to relevant resolution history. Scoring and Routing sends tickets to the right queue without manual triage. Workflow Copilot helps agents draft responses faster on the complex tickets. Autonomous Agent handles the structured, repeatable Tier 1 cases (standard refunds, password resets, account questions) without human involvement.

Data prerequisites:

• RAG Assistant: historical resolved tickets with resolution notes; product documentation; current policy documents
• Scoring and Routing: labeled ticket history (type, urgency, correct queue for each ticket type); sufficient volume to train the classification model
• Workflow Copilot: integration into the helpdesk interface; RAG Assistant must be running to provide resolution context
• Autonomous Agent: confirmed API access to payment processor, helpdesk system, and email; rollback capability for every action type; hard scope constraints defined before deployment

Sequencing recommendation: Start with RAG Assistant. It requires no outbound actions, carries low risk, and delivers immediate value to agents who are currently searching manually for resolution history. Add Scoring and Routing when ticket volume makes manual triage a meaningful time cost. Layer in Workflow Copilot after RAG is running and agents are comfortable with AI suggestions. Defer Autonomous Agent to year 2. It has the highest prerequisite requirements, the highest governance investment, and the highest consequence of error. Build credibility with the simpler patterns before taking on autonomous resolution.

First-win milestone: Tier 1 deflection rate. What percentage of Tier 1 tickets is the combination handling without full human resolution effort? Track this monthly. An improvement from 15% to 40% in 90 days is a strong signal the stack is working.

---

Finance and Accounting

Primary combination: Vision Extract + Anomaly Agent + Document Review

Why these together: Finance is documentation-heavy and exception-driven. Vision Extract automates the data entry that currently consumes AP and expense team time. Anomaly Agent monitors transaction and expense streams for policy violations and fraud signals. Document Review accelerates contract and audit workflows where manual review is the bottleneck.

Data prerequisites:

• Vision Extract: access to source documents (invoices, receipts, contracts) in digital form; target system of record with write access confirmed; document type training examples
• Anomaly Agent: minimum 60-90 days of clean historical transaction data for baseline; consistent collection cadence with no gaps
• Document Review: sample documents representing the range of what needs to be reviewed; standard or template to compare against; known exception categories

Sequencing recommendation: Start with Vision Extract on accounts payable invoice processing. The ROI case is straightforward: invoices per hour processed manually vs. with Vision Extract, error rates compared. It doesn't require a training period or baseline. Deploy Anomaly Agent in parallel if fraud detection or expense policy monitoring is a known problem. Document Review is appropriate when contract review is a documented bottleneck with measurable cycle time. These three patterns are largely independent of each other, so the sequencing is driven by where the pain is highest, not by technical dependencies.

First-win milestone: AP processing time and error rate. If manual invoice processing takes 8 minutes per invoice and Vision Extract brings that to under 2 minutes with a lower error rate, that's a concrete, auditable win that finance leadership can present to the CFO without needing to explain what an AI pattern is.

---

HR and People Operations

Primary combination: RAG Assistant + Scoring and Routing + Meeting Intelligence

Why these together: HR serves two very different customers: employees asking policy questions, and the recruiting function evaluating candidates. RAG Assistant handles employee policy questions at scale, reducing the load on HR Business Partners (HRBPs) for routine questions. Scoring and Routing handles high-volume recruiting by triaging applications and routing candidates to the right recruiter. Meeting Intelligence captures interview notes and reduces the reliance on handwritten feedback that gets lost between rounds.

Data prerequisites:

• RAG Assistant: current HR policy documentation (benefits, leave, compliance requirements); refresh process owned by HR, not IT
• Scoring and Routing: minimum 6 months of historical hiring data with outcome labels (hired, progressed to next round, rejected); job description and application data in a consistent format
• Meeting Intelligence: recorded interviews (with candidate consent); structured output schema mapped to the applicant tracking system (ATS)

Sequencing recommendation: Start with RAG Assistant for employee policy Q&A. The time-to-value is the shortest of any pattern in the HR stack: employees start asking questions the day it's deployed, and HRBPs see their repetitive inquiry volume drop within weeks. No model training, no historical data requirement, just a well-maintained knowledge base. Add Scoring and Routing for high-volume recruiting roles when application volume makes manual screening impractical. Meeting Intelligence is valuable but requires candidate consent processes to be in place before deployment.

First-win milestone: Time-to-screen reduction. For high-volume roles, how many days does it take to screen 100% of applications to the first-round decision? If manual screening takes 5 business days and Scoring and Routing gets it to under 24 hours, the recruiting team's throughput has fundamentally changed.

---

Product and Engineering

Primary combination: Workflow Copilot + Meeting Intelligence + Generative Research + Autonomous Agent (engineering only)

Why these together: Product and Engineering are both high-output functions where individual contributor leverage matters more than triage efficiency. Workflow Copilot improves individual productivity for both PMs writing specs and engineers writing code. Meeting Intelligence captures product discovery and user research sessions. Generative Research compresses competitive analysis and market research from days to hours. Autonomous Agent is specifically valuable for engineering coding tasks where the test-fix loop is well-defined and the tool boundaries are clear.

Data prerequisites:

• Workflow Copilot: integration into the primary tool (IDE for engineers, document editor for PMs); low-latency inference
• Meeting Intelligence: recorded product discovery calls and user research sessions with participant consent
• Generative Research: web access or internal corpus; citation tracking for research integrity
• Autonomous Agent (coding): GitHub repository access; test runner with structured output; tool registry with tested schemas; PR creation as the output boundary (no auto-merge without human review)

Sequencing recommendation: Start with Workflow Copilot. Both engineers and PMs see results immediately, adoption is voluntary and natural (the copilot is a tool, not a process change), and it requires no training data or baseline period. Add Meeting Intelligence for product teams running regular customer interviews. Generative Research for PMs doing competitive analysis. Autonomous Agent for engineering teams that want to automate the test-fix-revise loop on well-scoped tasks, but only after the team has established governance for what the agent is and isn't authorized to do.

First-win milestone: Time-to-first-draft for specs and PRDs. If a PM currently takes 3-4 hours to write a first-draft PRD from a set of user interview notes, Meeting Intelligence (to structure the notes) combined with Workflow Copilot (to assist the writing) should bring that to under 90 minutes. That's a recoverable, testable, concrete win.

---

Marketing

Primary combination: Workflow Copilot + Personalization Engine + Generative Research

Why these together: Marketing sits at the intersection of content production and distribution. Workflow Copilot accelerates content creation: headlines, body copy, ad variants, email drafts, social posts. Personalization Engine makes the distribution relevant: different content for different segments, behaviors, and moments. Generative Research compresses market intelligence, trend analysis, and competitive monitoring from analyst hours to hours.

Data prerequisites:

• Workflow Copilot: brand style guide in accessible format; example outputs the model can reference for tone calibration
• Personalization Engine: minimum 30 days of user behavior data; personalization surface that supports dynamic rendering (email platform, CMS, or ad delivery with variant support); feedback loop connecting personalization decisions to conversion outcomes
• Generative Research: web access; internal content corpus for brand-voice alignment; citation requirements defined in advance

Sequencing recommendation: Start with Workflow Copilot for content production. Marketing teams see immediate throughput improvement: more drafts in less time, less blank-page time, easier iteration. No behavior data required, no model training, no infrastructure beyond the copilot integration. Add Personalization Engine once you have sufficient behavior data (at least 30 days of consistent signal) and once your delivery infrastructure supports variant rendering. Generative Research can run in parallel from day one for research-intensive teams.

First-win milestone: Content production throughput. How many assets (emails, ads, landing pages, social posts) does the team produce per week with vs. without the Workflow Copilot? Track it over the first 60 days. Volume alone isn't the right metric, but volume at maintained quality, measured by A/B test performance or engagement rate, makes the case clearly.

---

Legal and Compliance

Primary combination: Document Review + RAG Assistant + Anomaly Agent

Why these together: Legal is the most document-intensive function in most organizations. Document Review replaces the hours of manual contract reading that blocks commercial velocity. RAG Assistant gives legal and compliance teams instant access to policy, regulation, and precedent without requiring a lawyer to be on call. Anomaly Agent provides continuous compliance monitoring across transaction and communication streams.

Data prerequisites:

• Document Review: sample contracts or documents representing the full range of what's reviewed; known exception types; the standard or template to compare against
• RAG Assistant: current legal policies, standard contract templates, regulatory guidance documents; refresh process to update when regulations change
• Anomaly Agent: compliance-relevant data streams (transactions, communications, expense reports); baseline period; defined escalation path when anomalies are flagged

Sequencing recommendation: Start with Document Review for the contract type that generates the most review volume: NDAs, vendor agreements, or employment contracts, whichever creates the largest queue. The ROI case is cycle time: days to review a contract vs. hours. This is the easiest pattern in the legal stack to get leadership approval for because the time savings are immediately auditable. RAG Assistant is a parallel deployment for teams where policy Q&A is a significant load. Anomaly Agent is appropriate once you've defined what compliance monitoring looks like for your function and have the baseline data to support it.

First-win milestone: Review cycle time reduction.

"Legal teams that deploy Document Review for standard NDAs reduce review cycle time from 3 business days to same-day in the majority of cases. The commercial team notices first. That visible speed improvement makes legal AI adoption self-reinforcing." (Rework Legal AI Analysis, 2026)

If a standard NDA takes 3 days to clear legal review and Document Review brings it to same-day, the sales team notices, the procurement team notices, and the legal team's reputation for enabling rather than blocking the business improves.

---

The Departmental Pattern Set

The Departmental Pattern Set is a decision framework that defines, for each business function, the three-pattern combination with the highest first-year ROI probability: a starting pattern with no upstream dependencies, a second pattern that compounds value from the first, and a deferred pattern that requires governance maturity the first year rarely achieves. The pattern set for each function is determined by two axes: where the function's highest-volume, most-repetitive decisions occur, and where structured historical outcome data already exists to train or calibrate the model.

Rework Analysis: Based on McKinsey's finding that sales and marketing leads all business functions in AI adoption, and PwC's data showing 71% of organizations using AI in finance operations, the department-level sequencing pattern is consistent: functions with high data volume and repetitive decision cycles deploy faster and achieve clearer ROI than functions with lower volume and more judgment-intensive workflows. Rework's implementation data shows that teams following the Departmental Pattern Set sequencing (start with no-dependency pattern, add dependent pattern after 30 days of upstream data) achieve production deployment in 60-90 days. Teams that attempt all three patterns simultaneously average 5-7 months to production.

Summary reference table

The pattern that belongs in your year 1 plan is the one with the highest ROI and the lowest prerequisite debt for your specific function. The one you defer is the one with the highest governance investment and the least mature data foundation. Sequence for reality, not ambition. McKinsey's finding that only about one-third of organizations have moved from experimenting to scaling confirms this: the teams that scale are the ones that picked one function, proved value there, and expanded. For the underlying data that makes or breaks each of these combinations, data readiness check by AI pattern gives you the per-pattern audit checklist.

---

Learn more

• What Is an AI Sales Operator? 4 Patterns Working Together
• AI Sales Ops Implementation Roadmap
• Sequencing AI Patterns in a Multi-Year Roadmap
• Stacking Patterns to Build AI Agents
• Buy vs. Build Decision for Each AI Pattern
• Pattern Dependencies and Prerequisites