If you manage performance reviews in an owner-managed services firm, you know the pattern. The first three writeups get proper thought. By review ten, the paragraphs get shorter and the same phrases start appearing under different names. By the end of the cycle, what you’ve written reflects the last month rather than the full year, because there was no time to reconstruct the rest.
AI-assisted review drafting addresses that specific failure mode. The tools available in 2026 can pull together a year’s worth of task data, 1:1 notes, and prior feedback, then generate a first-pass summary for each person. The manager still makes every judgement call. They just start with something real to work from.
What does AI actually do in a performance review?
AI-assisted review tools do two distinct things. They collect and summarise year-round data, pulling in goals, task logs, 1:1 notes, and prior feedback to produce a factual record of what each person achieved. Then they draft a narrative paragraph for the manager to edit. The output is a starting point, not a finished document. Platforms such as 15Five, Lattice, and Leapsome have these capabilities built in.
Charter Works’ employer research identifies two patterns in common use: tracking what people completed from task management systems over the period, and coaching managers through the writing process with AI-suggested phrases and structures. Asana, for example, has built-in AI that can compile a list of completed tasks per person and generate an output summary ready to feed into a review.
A second capability, available to firms with regular feedback processes, involves short recurring pulse surveys processed by NLP. The tool extracts themes from free-text responses and flags sudden shifts in team sentiment, giving managers sight of emerging morale problems before they surface as resignations or silences.
In all three cases, the AI produces material for a person to use. Rating performance, recommending pay changes, and employment decisions stay with the manager who knows the context.
Why does it matter for an owner-managed services firm?
In a business with five to fifty staff, performance reviews typically fall on one or two people, twice a year. The time available is always less than the list requires. Reviews get compressed. People who asked for feedback recently get better writeups; people who did their job quietly get generic ones. That inconsistency compounds over years into retention problems that are much harder to unpick.
The direct case is time. Operator playbooks report that a basic AI-assisted review process can be implemented in 60 to 90 days when scope is kept tight. The time saving, comparable to the 20 to 50 percent reductions AI delivers in document-intensive work like contract review, comes from eliminating the blank-page problem at the start of each writeup, not from removing the manager’s judgement.
The second reason to care is legal. Under UK GDPR, performance data, including 1:1 notes, ratings, and manager comments, is personal data. Using it in public AI tools without a data protection impact assessment or staff notification is not a hypothetical risk. The ICO is explicit: employees must be told what data is being collected for their evaluation, how it is processed, including whether AI is involved, and who sees the outputs.
Where does AI help most in the review cycle?
AI earns its time-saving in three places: the data collection phase, turning notes and task logs into a structured record before the season starts; the drafting phase, producing a first-pass narrative the manager then edits; and the year-round feedback layer, summarising 360-degree inputs and flagging sentiment shifts from pulse surveys. The review conversation itself stays fully human.
For data collection, the practical entry point is connecting your project management tool to an AI summarisation layer. Asana, Jira, and similar tools can compile completed tasks per person and generate a summary of output over a period. That summary becomes the factual spine of the review: what each person shipped, over the full year, not just the last six weeks.
For drafting, purpose-built HR platforms take the next step, processing the factual record alongside any 1:1 notes and generating a draft narrative per competency or performance dimension. The manager’s job is then to review the draft for accuracy and add the context that task data cannot capture: how someone handled a difficult client conversation, whether they stepped up when a colleague left suddenly, what changed in their confidence over the twelve months.
For year-round feedback, short recurring pulse surveys processed by NLP can track team sentiment between formal review cycles. The ICO’s guidance on workplace monitoring requires telling staff this is happening, explaining the purpose, and documenting the lawful basis for processing.
When to use AI and when to keep it out of the process
The line the ICO draws is clear: AI can assist with drafting and summarising, but any decision that significantly affects someone’s employment, pay, promotion, or continued role must involve meaningful human oversight. Under Article 22 of UK GDPR, if AI produces outputs that influence those decisions, employees have the right to know it was involved, to understand the logic, and to request a human review.
For an owner-managed services business, that translates into three practical rules.
First, draw the line at ratings and outcomes. AI drafts words, suggests phrases, and summarises data. It cannot set the final rating, recommend a pay increase, or influence a promotion without a named person signing off and being able to explain their reasoning. The draft is an input to the manager’s decision, not a substitute for it.
Second, be careful what data you feed in. The ICO’s 2023-24 Workplace Technology guidance classes AI systems used in worker evaluation as high-risk processing, requiring a DPIA before deployment. That assessment should document what data goes in, what decisions it will influence, and how you will explain the process to staff.
Third, watch for bias. Research on algorithmic fairness in HR contexts shows that AI systems can reproduce patterns from training data in the language they produce, using different adjectives or phrases for different groups without any explicit instruction to do so. The ICO and the Centre for Data Ethics and Innovation both recommend testing AI-drafted HR material for differential language patterns across demographics. For a small firm, a periodic spot-check by a second reader is a proportionate control.
If you serve EU clients or use tools embedded in EU-marketed platforms, the EU AI Act classifies AI in employment evaluation as high-risk, with additional transparency and documentation obligations.
What do you need in place before you start?
A 90-day rollout of basic AI-assisted reviews is realistic for an owner-managed business that keeps scope tight, according to operator playbooks for comparable firms. The speed depends entirely on what you already have: documented goals, a habit of keeping 1:1 notes, and a project tool where completed work is logged. Without those inputs, the AI has nothing to summarise.
Tooling choices for a firm of this size include purpose-built HR platforms (15Five, Lattice, and Leapsome carry HR-specific controls, audit trails, and role-based access that generic AI tools do not), lighter stacks that connect Asana or Jira to an AI summarisation layer via Zapier or Make, and basic HRIS tools such as BambooHR or Gusto if you do not yet have structured employee records.
Before deploying any of them, three things need to happen. Carry out a DPIA: the ICO requires one before using AI tools that profile workers or may significantly affect them. Decide the lawful basis for processing, legitimate interests is the common choice for performance management, balanced against employees’ rights and documented clearly. Tell staff what is happening and why, including what data is used, what the AI does with it, and how their manager will use the output.
The practical Monday move, without buying anything new: open a shared document for each person on your team and begin logging notes after every 1:1. Two minutes per meeting. When review season comes, the AI has a real record to work from.
Performance reviews earn their place when the preparation is solid and the data spans the full year. AI handles that preparation problem well. For an owner-managed services business with five to fifty staff, the tools and process are within reach. If you want to talk through what a rollout would look like for your firm, Book a conversation.
