Your First 90 Days: Strategy to Execution

Building Your HR AI Roadmap

Part 1: Why Most HR Teams Get AI Backwards

What they found surprised them. The biggest time drain wasn't recruitment, which everyone assumed, it was documentation. Policy updates, job description rewrites, onboarding guides, performance review templates. Work that required care and accuracy but not creative judgment. Work that took a skilled HR professional four hours and could, with the right AI prompt, take forty minutes. They piloted Microsoft Copilot across those documentation workflows first. Within eight weeks, the team reclaimed an estimated 22 hours per person per month. Not by replacing anyone. By stopping the slow bleed of low-complexity, high-volume writing tasks.

Here's the principle buried in that story: Unilever didn't ask "what can AI do for HR?" They asked "where is our time going, and which of those places doesn't actually need a human brain?" That sequencing, audit first, tool second, is exactly what separates HR teams that get real results from AI and teams that buy subscriptions, run one enthusiastic pilot, and then quietly go back to their old spreadsheets. This lesson is about building a roadmap that works the way Unilever's did: grounded in your actual workflows, your actual team, and your actual constraints.

The Mistake That Costs HR Teams Six Months

Before Klarna made headlines for its AI customer service results, its HR team went through a quieter and less reported struggle. In 2022, they rolled out an AI writing assistant to help recruiters draft job postings faster. The tool worked. Recruiters liked it. But adoption stalled at about 30% of the team after the initial excitement faded. The reason wasn't the tool, it was the sequence. They'd given people a solution before those people felt the problem sharply enough to change their habits. Most recruiters had their own copy-paste systems for job postings. The AI was faster, but not fast enough to overcome the friction of learning something new when your old method was "good enough."

Klarna's HR team regrouped. They shifted their approach: instead of offering AI as an upgrade to existing workflows, they identified the workflows people actively complained about. Interview scheduling coordination. Offer letter personalization. First-draft responses to candidate FAQs. These were the tasks where people said "I hate doing this" or "this takes forever." When they introduced AI tools into those pain-point workflows, specifically ChatGPT Plus for drafting and Calendly's AI scheduling features, adoption hit above 80% within six weeks. Same tools, different entry point. The principle: lead with pain, not with features.

This is the mistake that costs HR teams six months of slow adoption and mild frustration. They introduce AI where it's technically useful rather than where people are emotionally ready to change. A roadmap fixes this by forcing you to distinguish between "tasks AI could help with" and "tasks people actually want help with." Those two lists overlap, but they're not identical. The overlap is your starting point. Everything else comes later, once your team has built confidence and habit with the tools.

What a Retailer Taught Us About Phasing

Marks & Spencer's HR team faced a specific problem that many retail HR operations know well: extreme seasonal volume. Twice a year, ahead of Christmas and summer, their recruitment team processed thousands of applications for temporary roles. Screening CVs, sending acknowledgment emails, scheduling phone screens, drafting rejection letters. The volume was brutal and the work was largely mechanical. In 2023, they piloted an AI-assisted screening and communications workflow using a combination of their ATS's built-in AI features and templated prompts in ChatGPT Plus for correspondence drafts. The result was a team that handled 40% more applications with the same headcount during peak season.

What makes the M&S example instructive isn't the volume win, it's the phasing. They didn't try to automate everything at once. Phase 1 was communications only: acknowledgment emails, screening questions, rejection letters. No AI touching selection decisions. Phase 2, rolled out three months later, added AI-assisted job posting drafts and internal mobility announcements. Phase 3, still in development, is exploring AI summarization of interview notes. Each phase built team confidence and surfaced process gaps before the next one started. By the time they reached Phase 3, their recruiters were experienced AI users, not anxious beginners. The phasing wasn't just practical, it was psychological.

Click to enlarge

Comparing AI Adoption Approaches: What Actually Works

Click to enlarge

The L&D Coordinator Who Built a Roadmap in One Afternoon

Not every AI roadmap story comes from a large enterprise. Sarah Okonkwo is an L&D coordinator at a 120-person financial advisory firm in Toronto. She's a team of one. In late 2023, she spent a Friday afternoon doing something deceptively simple: she opened a blank document and wrote down every recurring task she did in a typical month, then timed herself doing three of them to get honest numbers. What she found: she was spending roughly 11 hours a month writing and formatting training materials, course outlines, facilitator guides, participant handouts, for content that was largely standard compliance and onboarding training. Not creative work. Formatting and rewriting work.

Sarah's roadmap had exactly four items on it. First: use Claude Pro to draft first-pass training outlines from her bullet-point notes. Second: use Canva AI to format participant handouts from those outlines. Third: use Notion AI to maintain and update her training content library. Fourth: revisit in 90 days and add the next layer. She didn't need an enterprise software purchase or an IT project. She needed three tool subscriptions costing under $100 a month combined, and one clear workflow per tool. Eight weeks later, she'd reclaimed about 8 of those 11 hours. She used them to finally build the manager development program her firm had been requesting for two years. One person. One afternoon of planning. Real results.

Turning Your Audit Into a Prioritized Plan

Once you've mapped your workflows and scored them for AI-readiness, using the framework from the prompt example earlier, you have raw material. The next step is sequencing. Not everything that scores high on AI-readiness should go first. You also need to weigh two other factors: risk and visibility. Low-risk, low-visibility tasks make the best starting points. Job description drafts, onboarding email templates, policy document summaries, these are internal documents that get human review before they matter. If the AI produces something clunky, a quick edit fixes it. No candidate is harmed. No employee is affected. You build skill in a forgiving environment.

Higher-visibility tasks, like candidate-facing communications, performance review language, or anything that affects compensation, should come in Phase 2 or later, once your team is comfortable with AI outputs and has developed good editing instincts. This isn't about being timid. It's about building the kind of internal confidence that makes AI adoption stick. Teams that start with high-stakes tasks and get one awkward AI output often develop a generalized distrust of the tools that takes months to undo. Teams that start with forgiving tasks develop an accurate mental model of what AI does well and where it needs a firmer editorial hand.

The final sequencing consideration is team buy-in. Even if a task scores high on AI-readiness and low on risk, if the person who owns that task is deeply skeptical of AI, starting there creates friction. Look at your team honestly. Who is curious and willing to experiment? Give them the first workflow. Their success, and their willingness to talk about it in team meetings, is your best adoption tool. HR professionals trust other HR professionals more than they trust vendor demos or management mandates. One enthusiastic peer sharing real results is worth more than any training session you could run.

Click to enlarge

Key Principles From Part 1

1. Audit before you adopt. Map your actual workflows and score them for AI-readiness before selecting any tool. Buying a tool without a workflow map is the single most common reason AI adoption stalls in HR teams.
2. Lead with pain, not features. Introduce AI into the workflows your team actively complains about, not the ones that are theoretically improvable. Emotional readiness drives adoption more reliably than technical suitability.
3. Phase your rollout across 6–8 week windows. One workflow category at a time lets your team build genuine competence and surfaces process problems before you layer on complexity.
4. Start low-risk and low-visibility. Internal documents and draft materials that receive human review before use are the safest first workflows. They let your team develop AI editing instincts without consequences for errors.
5. Score workflows on time cost, frequency, and judgment required. Tasks that score highest on all three dimensions, high time cost, high frequency, low judgment required, are your strongest Phase 1 candidates.
6. Peer success stories beat management mandates. Identify your most curious team member and give them the first workflow. Their real results, shared in team settings, create more adoption momentum than any top-down initiative.
7. A one-page roadmap beats a fifty-slide strategy. Clarity and specificity, tool named, workflow defined, success metric written down, matter more than comprehensiveness. A roadmap people can actually use is worth infinitely more than one that looks impressive in a presentation.

Where Most HR Roadmaps Break Down

In 2023, Unilever's HR leadership team made headlines for rolling out AI-assisted candidate screening across 50+ markets. What didn't make the headlines was the internal chaos that preceded it. The team had invested in a sophisticated AI hiring tool, trained managers on the software, and set a go-live date, but had never mapped which existing HR processes the tool would actually replace, complement, or disrupt. Recruiters were still manually uploading candidate data into a separate ATS. Hiring managers were getting AI-generated shortlists they didn't trust because no one had explained how the rankings worked. The tool was live. The adoption was not.

The Unilever experience isn't a failure story, the company course-corrected and now uses AI hiring tools effectively across its global operations. But the early stumbles reveal something important: most HR AI roadmaps are built around tools, not workflows. Teams ask 'which AI should we buy?' before they've answered 'which part of our current process is actually broken?' That sequencing error wastes budget, frustrates staff, and produces the worst possible outcome, an expensive tool that sits underused while the old manual process quietly continues in parallel.

The principle Unilever eventually applied, and what separates functional HR AI roadmaps from shelfware, is workflow-first thinking. Before any tool enters the picture, you map the human process: where time is lost, where errors cluster, where good decisions get delayed because someone is waiting on information they could have in seconds. The AI roadmap becomes a workflow improvement plan that happens to use AI. That reframe changes everything about how you prioritize, sequence, and measure success.

Mapping Your Processes Before Picking Your Tools

Marriott International runs one of the largest hospitality HR operations in the world, over 380,000 employees across 8,000+ properties. When their HR leadership team began building an AI roadmap in 2022, they started with a process audit that had nothing to do with AI. Department heads in HR were asked to log every recurring task their teams performed over a two-week period, along with rough time estimates. The output was a priority matrix: high-volume, low-complexity tasks on one end; low-volume, high-judgment tasks on the other. That matrix became their roadmap skeleton.

What Marriott found will feel familiar to most HR teams. The highest-volume tasks, answering benefits questions, scheduling interviews, sending onboarding reminders, formatting job descriptions, were consuming enormous collective hours but required almost no specialized judgment. These became their first AI targets. The lower-volume tasks, compensation conversations, performance improvement plans, terminations, conflict mediation, required significant human judgment, legal sensitivity, and relationship trust. These stayed human, at least for now. The roadmap wasn't a list of tools. It was a prioritized workflow map with AI entry points marked in.

You don't need a team of consultants to do what Marriott did. A shared spreadsheet and two weeks of honest task-logging will surface the same patterns. The goal is visibility. Most HR professionals are so embedded in their daily routines that they've never stepped back to see which tasks are genuinely high-judgment versus which ones just feel important because they've always been done manually. AI roadmap planning is partly a process design exercise, and that's work any experienced HR professional can lead without any technical background.

A Different Industry, The Same Pattern

Shift from hospitality to healthcare. Northwell Health, New York's largest healthcare network, employs over 80,000 people and faces one of the most complex HR environments imaginable, clinical licensing requirements, union agreements, shift-based scheduling, and some of the highest turnover rates of any industry. Their HR AI roadmap, developed between 2022 and 2024, didn't start with recruiting or onboarding. It started with a specific, painful problem: nurse managers were spending an average of 11 hours per week on scheduling-related administrative tasks. That number, surfaced through a simple internal survey, became the anchor for their entire first phase.

Northwell's approach illustrates a principle that applies equally to a 12-person startup and an 80,000-person health system: anchor your roadmap to a specific pain point with a measurable cost. Eleven hours per manager per week, multiplied across hundreds of nurse managers, is a number that justifies investment and creates urgency. It also gives you a clear success metric, if AI-assisted scheduling gets that number to six hours, the roadmap is working. Vague goals like 'improve HR efficiency' produce vague results. Specific pain points produce specific, defensible wins that build internal support for the next phase of your roadmap.

Comparing HR AI Approaches Across Organization Types

The Solo HR Professional's Roadmap

Not every HR team has the infrastructure of Marriott or Northwell. Meet Priya, an HR generalist at a 90-person SaaS company in Austin. She is the entire HR department. She handles recruiting, onboarding, benefits, performance reviews, compliance, culture programming, and anything else that touches people, alone. When Priya decided to build her personal AI roadmap in early 2024, she didn't have a project team or a consultant. She had a $20/month ChatGPT Plus subscription, a list of her own recurring headaches, and three hours on a Saturday morning.

Priya's roadmap had three phases, each built around a single recurring task she found disproportionately time-consuming. Phase one: job descriptions. She was writing four to six per month from scratch, spending 90 minutes each. With a well-structured prompt template, she got that to 20 minutes, including her own edits. Phase two: the weekly 'HR inbox,' which averaged 35–40 employee questions about benefits, PTO policies, and payroll timelines. She used Claude Pro to draft a comprehensive internal FAQ document, then updated it monthly. Direct email volume dropped by roughly 60%. Phase three: performance review prep, helping managers write balanced, specific feedback instead of vague ratings. She created a reusable prompt template and shared it with the 12 people managers at the company. What makes Priya's roadmap instructive isn't its sophistication. It's its honesty about constraints and its insistence on measurable impact at every step.

Sequencing Your Roadmap: The Three-Phase Framework

Across the examples in this lesson. Unilever, Marriott, Northwell, and Priya's solo operation, a consistent sequencing pattern emerges. The most successful HR AI roadmaps move through three distinct phases, and the organizations that skip or rush any phase tend to end up back at square one. Phase one is always about quick wins: high-volume, low-risk tasks where AI saves measurable time with minimal downside. These wins build credibility with leadership, confidence in your own team, and the practical experience you need before tackling anything more complex. Think job descriptions, scheduling emails, FAQ documents, and template-based communications.

Phase two introduces AI into processes that require some human judgment but benefit from AI assistance in preparation, summarization, or drafting. Performance review support fits here. AI doesn't make the assessment, but it helps managers write clearer, more specific feedback. Exit interview analyzis belongs here too. AI identifies patterns across dozens of interviews faster than any human could, but an HR professional still interprets and acts on those patterns. Candidate communication sequences, onboarding content personalization, and policy document updates all fall into this middle tier. The key discipline in phase two is defining exactly where AI hands off to human judgment, and making that boundary explicit to everyone involved.

Phase three is where organizations with strong foundations begin using AI for more strategic work, workforce planning analyzis, predictive attrition modeling using HRIS data, DEI reporting and trend identification, and compensation benchmarking. This phase typically requires some integration between AI tools and existing HR systems, and it benefits from the trust your team has built through phases one and two. Attempting phase three work without that foundation is exactly the mistake that produces the Unilever scenario, technically capable tools with no organizational readiness to use them. Sequence matters as much as selection.

Click to enlarge

What These Stories Actually Teach Us

1. Workflow mapping comes before tool selection, always. Unilever's early struggles came from reversing this order. Know your process pain points before you evaluate any software.
2. Anchor your roadmap to a specific, measurable problem. Northwell's 11-hours-per-manager-per-week metric gave their roadmap direction and gave their leadership a reason to invest.
3. The size of your team doesn't determine the quality of your roadmap. Priya's solo operation produced more measurable impact than many enterprise HR teams with full project budgets.
4. Phase sequencing is a discipline, not a timeline. Move to phase two when phase one is producing consistent results, not when a vendor says you're ready for an upgrade.
5. Your prompt library is infrastructure. The prompts your team refines and saves are assets that compound over time, training future staff and ensuring consistent AI output quality.
6. Define the human-AI handoff explicitly for every process. Where does AI stop and human judgment begin? If your team can't answer that for each task, your roadmap has a gap.
7. Measure before and after, even informally. Time estimates, email volume, draft quality, any honest before-and-after comparison gives you evidence that builds internal support for continued investment.

From Pilot to Policy: Making AI Stick in HR

In 2023, Unilever's HR team faced a problem that many large organizations know well: they had enthusiastic pockets of AI adoption scattered across recruiting, L&D, and compensation, and absolutely no coherent strategy connecting them. One team was using ChatGPT to draft job descriptions. Another had quietly subscribed to an AI interview-scheduling tool. A third was running AI-generated engagement surveys without telling employees. When a data privacy audit surfaced the inconsistency, the CHRO had to pause everything. The tools weren't the problem. The absence of a roadmap was.

What Unilever discovered, and what their HR leadership later shared publicly at an HR Tech conference, is that AI adoption without governance creates organizational debt. Every ad hoc decision made without a framework becomes a policy gap you'll fix under pressure later. The company ultimately built a cross-functional AI steering group that included HR, Legal, IT, and employee representatives. They mapped every AI touchpoint in the employee lifecycle. Then they assigned ownership, set review cycles, and created a single intake process for any new AI tool request. Within 18 months, they had a functioning HR AI roadmap that actually held.

The principle Unilever extracted from their painful pivot is worth internalizing: AI adoption in HR is not a technology project. It is a change management project that happens to involve technology. The tools are the easy part. The hard part is aligning your team on what AI is for, who decides when to use it, and how you protect the humans in the system. That framing changes everything about how you build your roadmap.

Click to enlarge

How a Mid-Size Retailer Built Governance Before Tools

Primark's HR team in the UK took a counterintuitive approach when they began exploring AI for people operations. Instead of starting with tools, they started with a single question: what decisions in HR should AI never make alone? They gathered their HR business partners, their legal team, and three employee representatives for a half-day workshop. The output was a simple two-column list: AI-assisted decisions (screening resumes for minimum qualifications, drafting offer letter language, summarizing exit interview themes) and human-only decisions (final hiring choices, performance improvement plans, terminations, promotions). That list became the backbone of their governance policy.

Once governance existed, tool selection became much easier. Every vendor pitch got filtered through one question: does this tool respect our human-only list? Tools that automated final decisions were eliminated immediately. Tools that surfaced insights for human review moved forward. Primark's HR director noted that this sequence, governance first, tools second, saved them from at least two expensive vendor contracts that would have created compliance problems under UK employment law. The roadmap protected them before they even knew they needed protection.

A School District That Got the Sequence Right

Jefferson County Public Schools in Colorado runs HR operations for over 14,000 employees, teachers, administrators, support staff. When their HR director began exploring AI in 2023, she faced a challenge unique to public sector HR: every policy decision is subject to union agreements, state employment law, and public records requests. She couldn't just try things and see what happened. So she built her roadmap in phases, starting with zero-risk applications: using Microsoft Copilot to draft internal HR communications, summarize meeting notes, and generate template letters. No employee data. No decision-making. Just writing assistance.

After six months, her team had built real AI fluency in a safe environment. They understood what the tools could and couldn't do. They'd caught errors, learned to verify outputs, and developed judgment about when AI drafts needed heavy editing. Only then did they expand into higher-stakes territory, using AI to analyze aggregated, anonymized turnover data and surface retention themes. The phased approach meant that when they presented AI-assisted insights to union leadership, they could demonstrate the safeguards in place. Trust had been built before it was needed.

Turning Your Roadmap Into a Living Document

The biggest mistake HR teams make with AI roadmaps is treating them as one-time strategy documents. They get created, presented to leadership, filed somewhere, and never touched again. AI tools change fast, a capability that didn't exist in ChatGPT six months ago might be standard today. Your roadmap needs a quarterly review cycle built in from day one. At each review, ask three questions: What did we try? What worked? What do we need to retire or upgrade? Keep it short. A 60-minute quarterly review is enough if you're asking the right questions.

Employee trust is the metric most HR AI roadmaps forget to measure. You can track time saved, cost reduced, and process speed easily. Tracking whether your employees trust that AI is being used fairly in decisions that affect them is harder, but more important. Build in an annual employee pulse question specifically about AI use in HR. Something as simple as: 'I trust that AI tools used in HR decisions are applied fairly and with human oversight.' That single data point will tell you more about roadmap health than any efficiency metric.

The HR professionals who build the most effective AI roadmaps share one habit: they stay curious without chasing every new tool. They subscribe to one or two reliable sources, they test new capabilities before recommending them, and they maintain a short list of 'tools we're watching.' This discipline, wide awareness, narrow adoption, keeps your roadmap focused. You don't need to use every AI tool available. You need to use the right ones well, in the right sequence, with the right guardrails. That's what a roadmap is for.

1. AI adoption without governance creates organizational debt, policy gaps appear at the worst possible moment.
2. The most effective HR teams establish a human-only decision list before selecting any AI tools.
3. Start with zero-risk, writing-assistance applications to build team fluency before moving into higher-stakes workflows.
4. A roadmap has four components: use case inventory, governance framework, capability plan, and review cadence.
5. Phase your adoption, foundation tasks first, analyzis and decision support only after trust and skills are established.
6. Employee trust in AI fairness is a metric worth measuring explicitly, not just assuming.
7. Treat your roadmap as a living document with a quarterly review cycle, not a one-time strategy deliverable.
8. Wide awareness of AI tools combined with narrow, disciplined adoption is more effective than chasing every new release.

Click to enlarge