Is an AI executive assistant worth it for a small business?

Small businesses feel administrative cost differently from larger organisations. The owner often performs the work after hours, so it may not appear as a payroll line even though it consumes delivery, sales and recovery time. At the same time, every new tool creates setup, subscription, privacy and supervision cost. An assistant is valuable only when it closes recurring loops with less total effort and acceptable risk. Faster writing is useful, but it does not automatically remove the task, find the attachment, maintain the record or obtain the final response. This guide provides a two-week ROI method that works before a vendor supplies a case study. It does not invent average hours saved or assume every returned hour becomes revenue. The decision uses the business's actual workflow volume, contribution value, recovered work, review burden and failure cost. It also protects work that should remain human.

Which small-business problems create the strongest value?

The strongest cases are frequent digital workflows where information already arrives through connected systems and the correct result can be verified. Repetition and visible completion matter more than novelty.

Examples include turning invoices into reviewable bill records, preparing meetings from real history, drafting replies with source facts, retrieving client documents, tracking personal commitments and coordinating calendar changes. Each begins with an authoritative source and ends in a state the owner can inspect. The one-person professional practice guide maps these loops across a solo operation.

Weak starting points include rare strategic decisions, relationship repair, ambiguous negotiations and work that happens mainly by phone or in inaccessible portals. The model may help prepare, but the administrative endpoint remains human. List one week of interruptions and rank them by frequency, minutes, error cost and digital availability before selecting a trial workflow.

How should a business calculate AI assistant ROI?

Calculate net value from work removed and outcomes recovered, then subtract every direct, human and risk cost. Do not multiply a vendor's generic time-saving claim by an executive billing rate.

Use this monthly model:

net value = (net hours returned × realistic contribution value per hour) + recovered gross profit + avoided measurable loss - total assistant cost

ROI % = net value ÷ total assistant cost × 100

Total cost includes subscription, usage, connected software, setup amortisation, review, corrections, security work and failure recovery. Net hours returned are manual baseline time minus assisted operating time, including approvals. Contribution value should reflect what the business can actually do with the time. If demand is constrained, an hour returned may improve rest or service quality without creating immediate revenue; that is real but should not be mislabelled as sales.

What should be measured during a two-week trial?

Record each chosen workflow from source to correct endpoint, including manual baseline, assistant time, review, corrections, failure and whether the result stayed useful.

Use ordinary days, not a curated inbox. Include at least one ambiguity and provider failure. Measure desktop and mobile if both matter. The result should tell the owner which workflow works, not merely whether the assistant seems intelligent.

How should recovered work be valued?

Value a recovered commitment, invoice or opportunity only when evidence shows the assistant changed the outcome and the economic contribution is known. Avoid counting pipeline face value as benefit.

If the assistant surfaces a forgotten R10,000 invoice and the client pays, the value is not automatically R10,000; much of that may already have been expected, and gross profit differs from revenue. The attributable benefit may be reduced delay, lower collection risk or avoided write-off. If meeting preparation prevents ten minutes of searching, count the net minutes. If a follow-up recovers a qualified project, use expected gross contribution adjusted for probability and document the assumption.

This conservative method protects the decision from one dramatic anecdote. It also makes improvements repeatable: the business can see whether commitment tracking or document retrieval, rather than general chat, creates the real return.

Which work should remain human even if ROI looks positive?

Professional judgement, sensitive relationships, novel negotiation and consequential authority should remain with a qualified person unless a narrowly governed process proves otherwise. Financial value does not erase duty.

The assistant can assemble evidence, identify missing information, prepare a draft and maintain records. A lawyer decides legal advice; an adviser decides suitability; an owner handles a distressed employee or strategic client. External email and shared-calendar changes should use meaningful approval. Payment release, binding commitments and destructive changes need stronger controls.

Keeping these decisions human does not weaken ROI. Good preparation can shorten the expensive judgement step without pretending to replace it. The current cost comparison helps price software, human and hybrid arrangements against the same completed-work unit.

When is the assistant not worth the subscription?

It is not worth it when eligible workflow volume is low, data cannot be connected safely, review consumes the saved time or failures create disproportionate risk.

Suppose a tool saves three hours of drafting but requires two hours of correction, one hour of rule maintenance and a monthly security workaround. Net capacity is zero before subscription. A small inbox may be handled more cheaply with filters and calendar defaults. A deterministic automation may beat AI when inputs never vary. Occasional human support may suit work dominated by calls and exceptions.

Stop or narrow the trial when the assistant repeatedly loses sources, hides uncertainty, creates duplicates or needs the owner to restate business context each time. Sunk setup effort is not a reason to continue.

What would make the result trustworthy enough to buy?

Require repeatable positive net value across normal and awkward cases, with acceptable access, approval, correction and exit controls. ROI and trust are one decision.

Before buying, identify the eligible plan and total price, documented in the AI assistant cost guide. Confirm that the connection scopes match the workflow, confidential data handling fits obligations, outward actions wait at the right boundary, sources remain inspectable and disconnect works. Run the same scenario several times and compare against the manual baseline.

Set a review point. Usage, prices and volume change. Keep workflows with demonstrated return; remove unused connectors and rules. A broad assistant earns breadth one proven loop at a time.

Who it is not for

An AI executive assistant is not a remedy for unclear processes, absent demand or work that the business has not decided how to delegate. It can accelerate confusion as easily as administration.

Businesses without a stable digital source of truth should first improve basic records. Those requiring specialised enterprise controls may find eligible plans too expensive. Owners seeking emotional judgement, trusted representation or physical coordination should evaluate human support. A service is also unsuitable when its provider cannot satisfy privacy, security, revocation or contractual requirements for the information involved.

The decision can be “not yet” or “only for meeting preparation.” Partial adoption is more rational than forcing every task through one tool to justify the subscription.

Conclusion: is the investment worthwhile?

An AI executive assistant is worth it when it produces repeatable net capacity: correct workflows completed, valuable work recovered and avoidable losses reduced after subscription, setup, review, correction and risk are counted. Run a two-week trial against a manual baseline, define completion before starting and include awkward cases. Value time at a realistic contribution rate and recovered work conservatively. Keep judgement, relationships and consequential authority human. The Hank proposition is designed around source-linked administration rather than summary volume, but it still needs real small-business trial evidence before publishing an ROI claim. That evidence—not a generic productivity percentage—should decide the purchase. Buy the smallest eligible plan only when the business can point to the exact recurring loops it no longer has to carry.