Start here
Pick the right problem before the shiniest tool
Professional-services leaders are not short of AI stories. Every conference, every vendor call, every LinkedIn post promises a revolution. What they rarely provide is a reliable method for choosing where to actually start.
The difference between firms making AI work and those still stuck in pilot purgatory comes down to one discipline: picking the right problems first. Not every task can be automated. Not every task that can be automated should be. That distinction matters more than any technology choice you will make. What follows is a repeatable method for finding those problems, evaluating them honestly, and knowing where to act with confidence.
Define it first
What “value” actually means
Before hunting for opportunities, you need a shared definition of what you are solving for. Generic answers — “save time”, “be more efficient” — will not survive a board conversation or a budget conversation. Value from AI lands in one of three categories:
- Efficiency gains. Automating routine work to save time and cost. This is your fastest return and the win that funds everything else. It is also where you win over the sceptics, because few people mourn the drudgery tasks.
- Enhanced decision-making. Better analysis, clearer options, fewer errors. Quality rises, variability drops, and your managers have the right information to hand when they need it.
- Innovation enablement. New products, services, or business models that were not viable before. An AI research assistant monitoring market signals, competition, and improvement opportunities around the clock is qualitatively different from a slightly faster spreadsheet.
The mistake most firms make is stopping at efficiency. The firms pulling ahead are aiming for all three.
The villain
AI hype is the noise, not AI itself
There is a reason so many AI pilots fail to make it to production. Generic AI tools are built to impress in demos, not to survive contact with a real business brief — one with messy data, legacy systems, compliance requirements, and people already stretched thin.
The villain is not AI. It is the noise: vendor promises that do not map to your operating model, advice from people who have never run a professional-services firm, and the pressure to adopt something visible before you have thought through what valuable looks like.
The antidote is method — specifically, a method for evaluating opportunities before you commit resource to them.
Where to look
The opportunity spectrum, from low-hanging fruit to strategic wins
Start with high-volume, rule-based work. These tasks are frequent, predictable, and expensive in aggregate, so the return from automating them is clear and fast.
- Repetitive data entry. Moving numbers from email or PDF into a system — invoices, delivery notes, CRM updates — is exactly what AI handles well. Extraction, validation against rules, and form completion cut manual effort dramatically and remove the error rate that slows payment cycles and frustrates suppliers.
- Document summarisation. Long contracts, research reports, and meeting transcripts become short briefs with headings and action points. Your senior people read once and understand fast, rather than spending forty minutes catching up before a client call.
- Email triage and categorisation. Automatically sorting, tagging, and prioritising inboxes — with suggested replies for standard requests — cuts response times without extra headcount. A human reviews, adjusts, and sends; the right issues simply surface first.
- Scheduling. A smaller but consistent win. Proposing times, managing time zones, and sending invites and confirmations with no back-and-forth needs zero senior attention once set up.
These are not headline-grabbing use cases. They are the ones that quietly return hundreds of hours a year and build the internal credibility you need to go further.
Once your team has proven it can land a basic automation and trust has been earned, the strategic opportunities open up — where AI improves outcomes, not just speed.
- Customer insight at scale. Clustering feedback, support tickets, and NPS comments to find themes and root causes turns reactive complaint handling into proactive improvement. Churn signals that once took a quarterly review to surface can trigger a save action within days.
- Decision support. Assembling options, costs, risks, and recommendations into a one-page brief does not remove judgement from your senior people. It removes the four hours of prep before they can apply that judgement. The human stays in charge of the final call.
- Knowledge management. The one firms consistently underestimate. Searching across SharePoint, Confluence, Slack, and Drive by meaning rather than exact keyword — with citations back to source — changes how quickly new hires become effective and how much time your senior team spends re-answering questions they have already answered in writing somewhere.
Build in order
The KWA framework prevents the confident wrong answer
The mistake most firms make is trying to deploy AI agents before their knowledge and workflows are in shape. The result is an agent that confidently produces the wrong answer, or one that cannot find the information it needs. The KWA framework — Knowledge, Workflow, Agents — exists precisely to prevent this.
- Knowledge first. Your institutional knowledge needs to be centralised, structured, and searchable before AI can use it reliably. If it lives in individual inboxes and personal drives, no agent can access it consistently.
- Workflow second. Your processes need to be documented and standardised before they can be automated. If the work happens differently depending on who is doing it, automation will inherit the inconsistency.
- Agents third. Once Knowledge and Workflow are solid, agents can do real work in production — with appropriate human oversight and escalation paths in place.
Skipping the first two layers is why most AI deployments take longer and cost more than expected. The technology is rarely the problem. The readiness is.
Score the process
Four characteristics of a strong candidate
The best processes to automate share four traits. If you see three of the four, you have a strong candidate.
- High volume. The task happens frequently. Saving ten minutes once is irrelevant; saving ten minutes a hundred times a week is material.
- Rules-based. The decisions have clear criteria. If you can explain the logic to a new starter in a short conversation, AI can probably handle it.
- Digital inputs and outputs. The data already lives in systems and can be made available. No scanning physical documents, no mystery spreadsheets only one person understands.
- Low complexity. Few variables, few edge cases, few things that can go wrong. Start here, build confidence, then expand.
Take invoice processing as a worked example. A mid-size firm’s accounts team receives invoices in a shared inbox, enters data from PDFs by hand — roughly fifteen minutes per invoice — and manages an error rate between five and eight per cent. Payments slow, suppliers chase, the team is frustrated. After automation with OCR extraction and validation checks: around thirty seconds per invoice, spent mainly on reviewing the result, and an error rate below one per cent. For a team at that volume, the payback period is typically around two months. Your numbers will vary, but the pattern holds.
Just because you can
The “could versus should” test
Not every automatable task should be automated. Before committing, apply five lenses.
- Does AI accelerate learning? Used well, it coaches, gives feedback, and offers worked examples that make people more capable faster. Used poorly, it removes the practice that builds genuine skill. Be deliberate about which tasks you automate and which you want people to learn by doing.
- Does it tighten feedback loops? Faster iteration is valuable when the direction is right. More cycles that do not move the needle only generate more work. Anchor acceleration to outcomes, not activity.
- Is it genuinely always-on work? Autonomous AI running checks and drafts around the clock is powerful for tasks that require it. For everything else, set guardrails, build monitoring dashboards, and invest in your team’s understanding so the capability stays in-house.
- Is it work people would gladly hand over? Ask the team what they would offload tomorrow if they could. The answers will surprise you, and the buy-in from that conversation is worth more than any implementation plan.
- Does it preserve human connection? Do not automate work people love, that builds relationships, or that keeps your team connected to clients and to each other. That is how you lose engagement and lose people.
Structure the brief
The Agent Opportunity Canvas
Before any automation project goes to a provider for scoping, structure the opportunity in four parts.
- Current process. Document the steps, time taken, and pain points. If it is not written down, it is guesswork — and most providers cannot quote accurately, let alone deliver fixed-price, without this foundation.
- Desired outcome. Set clear goals in business terms: time saved, error rate reduced, revenue impact. Define what “done” looks like before you start.
- Data and resources. Identify what data exists, what inputs are required, and which systems need integrating. Surface visibility and access issues before they become surprises mid-project.
- Implementation approach. Choose the right AI capabilities for the task, define where humans stay in the loop, and plan a phased rollout. Pilot first, prove value, then scale.
With a well-documented canvas, many automation providers will quote reliably and deliver on fixed price. Without it, expect scope creep and disappointment.
By function
Entry points by department
For leaders mapping their first moves by function:
- Marketing. Content generation, audience segmentation, and campaign analysis. Expect faster output and higher relevance — but ensure your team retains critical thinking and original voice, because AI-generated marketing increasingly sounds the same.
- Sales. Lead scoring, meeting summaries, proposal and quote drafts. Sales people should spend more time selling and less on admin. Hold them accountable for outcomes, not the AI’s performance.
- Customer service. FAQ handling, ticket categorisation, and sentiment analysis. Lower response times improve satisfaction, but monitor with checks and escalate to a human whenever the situation warrants it.
- Operations. Document processing, inventory forecasting, and quality control. Executed well, this can reduce cost by 20–30% while improving both customer and employee experience.
The discipline
Three principles for choosing your first move
Focus on business value, not the most interesting technology. Value earns funding and builds internal trust; technology for its own sake does neither. Start with well-defined, contained processes — a small blast radius if something goes wrong, and faster learning when it does. And measure your current state before you change it, because you cannot prove return on investment without a baseline.
The 5 Steps for AI Leadership framework — Align, Activate, Amplify, Accelerate, Govern — provides the sequence for moving from your first working automation to firm-wide adoption. The order is non-negotiable. Skipping steps is the reason most rollouts stall after the first pilot.
AI does not create value by being impressive. It creates value by doing useful work reliably, at scale, with the right humans in the loop.
The firms that will look back in three years and say AI changed their business are not the ones that adopted it fastest. They are the ones that adopted it most deliberately — starting with the right problems, building the knowledge and workflow foundations first, and proving value in contained pilots before committing at scale. That is what making AI work so your team can deliver actually means in practice.