The Bottleneck Isn't Your Tech Stack. It's Your Workflows.

You've added a scheduling app. Maybe a chat widget for your website. Your team uses ServiceTitan. You're subscribed to three software platforms you evaluated last year. By any reasonable measure, you're not behind on technology.

And yet your CSR team is still manually looking up job history on every call. Your dispatcher is still texting technicians to check availability. Your after-hours calls are still going to voicemail or to an answering service that takes a message and calls you back in the morning.

The technology isn't the constraint. The workflows are.

The "intermediate transition phase" most operators are stuck in

A 2025 paper published on arXiv by researchers at Old Dominion University in collaboration with practitioners from Deloitte and Accenture examined AI adoption patterns across industries. Their diagnosis: "Most organizations remain in an intermediate transition phase rather than operating in a fully agentic state." Rather than embedding AI as an operational actor within their workflows, most businesses use AI as "a productivity enhancer for individuals."

Map that directly to home services. The scheduling app helps one CSR book one appointment faster. It doesn't automate the booking workflow. The AI chat widget answers one customer's FAQ. It doesn't handle the downstream dispatch action. The technology is present. The workflow-level automation is not.

This is the intermediate transition phase: AI as tool, not as actor. Faster individuals, not transformed operations.

The real bottleneck is domain knowledge, not technology

The researchers were direct about where the actual constraint sits: "High-impact agentic AI workflows depend fundamentally on a deep understanding of business-domain processes. These processes often include informal rules, context-dependent decisions, exception handling, and human judgment developed through years of operational experience. Such knowledge is rarely captured in formal documentation."

In home services, this is the dispatcher who knows that a customer in a certain neighborhood always wants a specific technician. It's the CSR who knows that "my AC is making a noise" from a customer over 70 in July is an emergency, not a service call, even if the script says otherwise. It's the routing logic that accounts for traffic patterns, technician certifications, and truck stock in ways that no SOP document fully describes.

This knowledge isn't in your ServiceTitan configuration. It isn't in your training manual. It's in the heads of your experienced people — and it's exactly what an AI agent needs to be genuinely useful, rather than just fast.

What workflow-level automation actually looks like

Rather than thinking about AI as a feature you add to an existing process, workflow-level automation requires redesigning the process around what agents can own end-to-end.

Consider an inbound call. The current workflow: call arrives, CSR answers, searches CRM for customer history, checks the dispatch board, manually creates a job, calls or texts the technician, sends a confirmation. Seven steps, five of which are information retrieval and data entry.

The agentic workflow: specialized agents own each discrete step. Customer lookup agent identifies the caller and pulls history. Dispatch agent checks availability windows. Job creation agent writes the service request. Notification agent alerts the technician. Human coordinator reviews the output and intervenes only when the situation requires judgment — a complex complaint, an edge case the agent flags for escalation, a VIP customer with a non-standard arrangement.

The human is still in the loop. But they're reviewing outputs and handling exceptions — not doing data entry.

Why this is an organizational problem, not a procurement decision

The researchers were clear on this point: "The transition to agentic AI is not constrained by technological capability, but by organizational readiness, mindset, and operating models." You can buy the most capable AI platform on the market and still be stuck in the intermediate transition phase if you haven't done the work of identifying which workflows to delegate and in what sequence.

That work isn't glamorous. It's process documentation, knowledge capture, and honest assessment of where your operations are actually breaking. It's answering questions like: Which calls require judgment and which are purely transactional? Where does a CSR spend time searching rather than deciding? Which dispatch rules are written down and which exist only in a dispatcher's head?

The companies that pull ahead on AI aren't the ones with the biggest software budgets. They're the ones that did this organizational work first. Research on organizational AI readiness from Harvard Business Review consistently points to process clarity — not tooling — as the leading predictor of successful AI deployment.

A better filter for evaluating AI vendors

The researchers offer a principle that's directly applicable to vendor evaluation: "Agentic AI delivers the greatest value when applied to workflows that are manual, repetitive, decision-intensive, and span multiple systems or stakeholders."

Use that as a filter. Does the tool handle a workflow end-to-end — from inbound call to booked job in ServiceTitan — or does it handle one step and hand off to a human for the rest? Does it span the systems your operation actually uses, or does it live in its own silo? When it encounters an edge case, does it escalate intelligently or fail silently?

The operators who will pull ahead aren't the ones who buy the most software. They're the ones who look at their most manual, coordination-heavy workflows — and methodically hand them off to systems that can reason, act, and learn. That transition doesn't start with a vendor demo. It starts with understanding your own operations well enough to know what to delegate.