Why Unstructured Work Makes Good People Underperform

Good people underperform when work is unstructured, because attention is spent finding information, changing tools and repairing preventable errors. The remedy is not more pressure or indiscriminate AI, but a clear operating system that makes the next action, owner and source of truth obvious.

The problem is usually the work design

When a lead goes unanswered, managers often treat it as a motivation issue. Yet the person responsible may be monitoring email, chat, a personal task list, a CRM queue and several spreadsheets. The task was not refused. It disappeared between systems.

The same pattern appears in operations. A coordinator copies an address from a form into a CRM, then into an invoicing tool. A project manager asks for status in chat because delivery data is incomplete. A director rebuilds a forecast because stages mean different things to different salespeople.

These are structural taxes on human performance. They consume concentration without creating customer value. They also make capable employees look inconsistent because success depends on memory and local workarounds.

How can you tell if this applies to your company?

• Staff ask where information is stored before they can start work.
• The same customer details are entered into more than one system.
• Follow-up happens from personal inboxes and cannot be inspected.
• Managers need meetings to discover basic status.
• Work stalls when one experienced person is absent.
• Reports require manual reconciliation before anyone trusts them.
• Customers repeat information after a handover.
• Teams keep private spreadsheets because the central system does not fit the process.

If several signs are familiar, training alone will not fix the issue. The company needs to map the work and remove unnecessary decisions, transfers and re-entry.

Where does the cost accumulate?

Four mechanisms create most of the avoidable cost: context switching, missed follow-up, manual re-keying and management reconstruction.

Context switching is not simply opening another tab. A person must recall the customer, purpose, current state and next decision each time work resumes. Interruptions are especially expensive in proposal writing, analysis, account planning and complex delivery.

Missed follow-up is an ownership failure when no system creates and monitors the next action. Telling people to “be more proactive” leaves the mechanism unchanged. A reliable process records the commitment, due point, responsible person and escalation rule.

Manual re-keying introduces both labour and uncertainty. Even accurate copying needs checking. When values conflict, staff must decide which source is correct. Without a defined system of record, correction becomes negotiation.

Management reconstruction occurs when operational systems cannot answer simple questions. Leaders then request updates, build shadow reports and hold status meetings. This hides the underlying data defect while adding more work.

How do you estimate the cost?

Use a process-level estimate rather than trying to price “human inconsistency”. Choose one queue, such as new enquiries, proposal follow-up, order entry or monthly reporting. Observe a representative operating period and record only evidence you can defend.

For each activity, capture:

1. The people who touch the work.
2. The active handling time.
3. The waiting time between steps.
4. The number of repeated entries or checks.
5. The frequency and handling effort of corrections.
6. The opportunities or service commitments that expire.
7. The manager time spent requesting and reconciling status.

Then separate necessary work from avoidable work. Speaking with a prospect to understand risk is necessary. Copying contact details from a form into two systems is avoidable. Reviewing an unusual contract is necessary. Searching chat for the latest approved version is avoidable.

A practical calculation

Consider a 40-person agency assessing its proposal process. It does not begin with assumptions about AI savings. The operations lead samples completed proposals and records handling steps: enquiry capture, qualification, document preparation, internal approval, sending and follow-up.

The agency finds that coordinators repeatedly correct incomplete records, consultants search for current templates, and directors ask whether proposals have been followed up. It applies each employee's loaded internal cost to observed avoidable handling time. It separately records expired follow-ups and correction effort, without claiming that every delayed proposal would have closed.

This produces three defensible figures: administrative handling cost, correction cost and value at risk. The management team can test scenarios with the ROI calculator, but should keep assumptions visible. A credible case uses ranges and distinguishes cash savings, capacity released and possible revenue improvement.

Do not count released capacity as immediate cash unless spending will actually fall. If automation gives a salesperson more selling time, the financial result depends on whether that capacity is used and converts. This distinction prevents an attractive spreadsheet from becoming a false business case.

What should systems remove?

A good system removes avoidable coordination while preserving judgement. It captures information once, validates it near the source and passes it to the next authorised step. It makes ownership and due dates visible. It also produces an audit trail without asking employees to document the same event again.

For sales, this usually means a CRM designed around the real funnel rather than generic stages. Required fields should match decisions the business genuinely makes. Automations can acknowledge enquiries, create tasks, alert owners and stop sequences when a person replies.

For delivery and finance, an ERP or connected operating system can carry approved customer and project data forward. Access should follow roles. Changes to important fields should be traceable. The custom CRM comparison helps frame when configuration is enough and when a tailored implementation is justified.

AI belongs at points where language or variable documents make fixed rules insufficient. It can classify enquiries, extract fields, draft a summary or suggest a reply. Deterministic automation should still handle known routing, required fields, deadlines and permissions. Using a language model for every step adds variability where none is needed.

An agent should not silently fill missing facts. It should flag uncertainty and ask a person or customer for clarification. The objective is dependable flow, not the appearance of complete data.

How do you redesign the work?

Follow five named steps with clear owners.

Map the current path

The process owner walks a real item from entry to completion. Record systems, handovers, decisions, waiting points and exceptions. Front-line employees must participate because formal procedures rarely include all workarounds.

Choose the source of truth

The system owner names the authoritative location for customer, opportunity, project and invoice data. Other tools may display or enrich those records, but they should not create competing masters.

Define ownership and service rules

Each queue needs an owner, an expected response condition and an escalation path. Avoid vague shared responsibility. If a commercial lead owns a proposal, the CRM should show that ownership and create the next action.

Automate the stable steps

Use rules for deterministic work and AI for bounded interpretation. Add human approval where a decision affects price, contractual terms, sensitive data or a customer relationship. AI and automation implementation is appropriate when the work crosses several systems and needs auditability.

Operate with checkpoints

Review queue age, exception reasons, correction patterns and user workarounds. A drop in system usage is not automatically resistance. It may indicate that the design is adding work or failing to represent reality.

The operating owner should hold a short regular review with front-line users and the technical owner. Approve changes through a controlled backlog. Document definitions so that reports remain comparable after fields or workflows change.

What AI can and cannot fix

AI can reduce the clerical burden around unstructured inputs. It can turn an email into a draft CRM record, summarise a conversation and prepare a response from approved knowledge. It can monitor a queue continuously and bring exceptions to the right person.

It cannot decide what your sales stages mean, settle conflicting ownership or make poor source data trustworthy. It also cannot repair incentives that reward local optimisation over company outcomes.

This explains why personal AI tools often produce visible activity but limited operating improvement. McKinsey reports that 80% of surveyed organisations see personal productivity gains, while 37% report some EBIT impact. The gap is consistent with a broader lesson: individual assistance is easier than changing an end-to-end process.

The company should therefore prioritise workflows, not tool licences. Ask whether information moves from first contact to delivery and invoicing without re-entry, orphaned tasks or unclear ownership. If it does not, fix the architecture before adding more assistants.

Common failure modes

The first failure is automating a broken path. Faster duplication is still duplication. Standardise field definitions and ownership before increasing volume.

The second is measuring output rather than outcome. Messages sent and records created say little about correctness or commercial progress. Measure complete records, appropriate next actions, resolved exceptions and reduced correction work.

The third is excluding users from design. A workflow can be logically tidy and operationally impossible. Employees who handle exceptions know which cases violate the normal path.

The fourth is treating all variation as error. Skilled people adapt to context. The system should standardise clerical work while preserving authorised judgement. Overly rigid automation drives work into private channels.

The fifth is removing human access. Customers and staff need an obvious escalation route when the system misunderstands, data is sensitive or consequences are material.

For sales processes, repairing ownership, qualification and follow-up may require more than software. Sales optimisation and growth consulting focuses on the funnel and management practice around the tools.

The management decision

Do not frame the problem as humans versus AI. Ask why skilled people are performing clerical coordination that systems should handle. Then estimate the burden honestly and prioritise the process where a redesign will improve both capacity and control.

The best result is not a busier employee. It is a calmer operation in which the system remembers routine commitments, people see the same facts, and managers intervene where judgement is valuable. Our implementation method treats operating ownership and optimisation as part of delivery, not as work left behind after launch.

Key takeaways

• Apparent underperformance often starts with fragmented systems and unclear ownership.
• Estimate avoidable handling, corrections and management reconstruction at process level.
• Distinguish cash savings from released capacity and possible revenue improvement.
• Use deterministic automation for fixed rules and AI for bounded interpretation.
• Keep people responsible for judgement, exceptions and consequential decisions.

Frequently asked questions

Is context switching really a systems problem?

It becomes a systems problem when employees must monitor several channels, reconstruct status and copy information to complete one process. Some switching is inherent in varied work, but unnecessary switching can be designed out.

How accurate does the cost estimate need to be?

It should be accurate enough to compare priorities and test whether the investment remains worthwhile under conservative assumptions. Use observed samples, ranges and visible assumptions rather than false precision.

Should every manual task be automated?

No. Automate stable, frequent work where inputs and acceptable outcomes are clear. Retain manual handling when volume is low, exceptions dominate or the consequences require nuanced judgement.

Can a CRM alone solve missed follow-ups?

Only if the CRM reflects the real sales process, creates clear next actions and is used as the source of truth. Installing software without ownership, definitions and management practice simply relocates the problem.

How should employees be involved?

Ask them to map actual work, identify exceptions, test releases and review failures. Position the programme as removal of clerical friction, while being candid about how responsibilities and capacity will change.