Home · Solutions · Other solutions
Solution · Other solutionsThemes by unit within days of a pulse, and recognition that leaves a record
Pulse surveys and kudos in Teams, summarised into themes
Pulse comments are filtered, classified and summarised per unit, validated by an HR analyst, and delivered to every team lead as a readout card in Microsoft Teams.
Executive summary
Thousands of survey comments unread, recognition lost in a thread, communities nobody can find.
This runs on tools the organisation already owns.
Leadership sees themes by unit within days of a pulse closing, because reading and coding no longer depend on how much time one analyst had.
the Power BI theme and score model; the Microsoft Lists recognition record; Microsoft Teams community membership
Business problem
Engagement
Engagement data is collected well and used badly. Scores are easy to chart. The reasons sit in the free text, and reading two thousand comments by hand takes weeks, so it gets done partially or not at all. By the time a summary exists, the next pulse is due and the summary describes a mood that has moved on.
Team readouts are where the value leaks. A manager who receives a company score and no reasons cannot act, and engagement is managed at team level or nowhere. Building 140 readouts by hand costs a week of somebody's life per pulse, so it happens once, badly, or never.
Recognition and communities fail for a different reason: nobody owns them. Recognition depends on a manager remembering to post, and nothing makes it visible past the thread or records it where a talent conversation could use it. Communities run on a founder's enthusiasm and an intranet page that ages. Each piece is a nice-to-have, none has an owner with time, and the tools were never joined up.
How it works today
The sequence below repeats every quarter in organisations that survey conscientiously and act on averages.
- SystemInternal communications sends a Microsoft Forms pulse link to one distribution list covering everyone
- WaitingResponses accumulate for two weeks and nobody looks at them until the window closes
- PersonThe analyst exports responses to a spreadsheet and starts tagging free-text comments by hand
- WaitingTagging stops when the next priority arrives, typically after the first couple of hundred comments
- PersonA leadership deck is assembled weeks later from the comments that were actually read
- Risk of errorTeam leads receive a company-level score with no reasons attached to their own unit
- PersonRecognition is posted when somebody remembers, in a channel, and recorded nowhere
- PersonCommunity sign-up means emailing a lead, who adds the person by hand or forgets
Why the current process costs more than it appears
Nobody planned this work; it accumulated.
- Analyst weeks are the visible cost, and they buy a partial reading. The themes presented to leadership are the themes of the comments that fitted the available time, not the themes of the workforce.
- Decisions get made on averages while the reasons sit unopened. A unit whose comments have said the same thing for three pulses running is invisible in a score that moves by a tenth of a point.
- Silence teaches. Employees who write a comment and see nothing change write shorter comments next time, or none, and response rates fall exactly where engagement is weakest.
- Recognition that lives in a thread motivates for a day and informs nothing. Without a record by value and unit, a promotion conversation cannot reference it at all.
- Communities stall at their founders because joining requires knowing whom to email. The cost is not the community; it is the network that never forms across seven countries.
Cost of inaction
An employee who writes a comment learns something from the silence that follows it, and what they learn is that writing the next one is optional. Response rates drift down, the sample thins, and the number the board looks at becomes least reliable in precisely the units where it matters most.
Recognition and communities decay more quietly. A kudos channel with eleven posts is not a programme, and a community whose sign-up form points at a leaver is a line on an intranet page. Neither appears in any budget, and both are visible to every new joiner in their first month.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
A software and services company with 1,600 employees across seven countries, working hybrid on Microsoft 365; two internal communications people and one HR analyst run the engagement programme.
Four pulses a year in Microsoft Forms, a 62% response rate giving 992 responses per pulse, of which 0.55 carry a free-text comment, so 546 comments each time; 140 team leads expect a readout; 180 recognitions a month; eleven communities listed on the intranet.
Comments are exported and tagged by hand until time runs out; a leadership deck follows weeks later; team leads get the company number; recognition sits in a channel; community sign-up is an email to a lead.
Four minutes to read and code each comment, 45 minutes to assemble each team readout, and six minutes to compile each recognition, all done by three people who also have a day job.
A UiPath Robot collects pulse responses per unit and enforces the agreed minimum group size; UiPath GenAI Activities filter personal data out of the comments and then classify and summarise them against a taxonomy HR owns; an analyst validates themes and quotes in UiPath Action Center before anything is published; Power BI trends themes by unit, team leads get a readout card in Microsoft Teams, recognition is sent from a card into a Microsoft List, and one tap joins a community.
In the modelled case the time from a pulse closing to a published, validated theme summary falls from weeks to days, every team lead gets their own reasons rather than the company average, and recognition acquires a record. Illustrative throughout, and not a client result.
Proposed solution
This runs on tools the organisation already owns. The pulse stays in Microsoft Forms, with a link per unit so that responses can be grouped without asking anyone to identify themselves. A UiPath Robot collects the responses on a schedule, strips identifiers, and applies the minimum group size agreed with employee representatives: a unit below the threshold is reported inside its parent, never separately.
The comments are then processed in a fixed order that matters. UiPath GenAI Activities first filter personal data out of the text, so no name reaches a model. What follows is sentiment, classification against a theme taxonomy that HR owns and can change, and a short summary per unit with supporting quotes. Nothing is scored at the individual level and no individual sentiment is stored.
Publication is gated by a person. The HR analyst opens a validation task in UiPath Action Center, reads the proposed themes and the quotes chosen to support them, corrects categories the model got wrong, and releases the summary. Power BI then holds the trend: themes and scores by unit across pulses, which shows whether an intervention worked. Each of the 140 team leads receives a readout card in Teams with their own unit's reasons.
Recognition and communities use the same surface. A recognition card in Teams tags the recognition to a company value, writes it to a Microsoft List and feeds a weekly digest, so it survives past the thread. A community directory card lists what exists and who leads it; one tap adds the person to that community's team, done by a robot with a least-privilege account rather than by a founder with a spreadsheet.
Microsoft Forms pulse collection; Microsoft Teams Adaptive Cards; Microsoft Lists; UiPath Action Center validation tasks and actionable notifications in Teams; UiPath AI Trust Layer model allow-list, masking and audit; Power BI semantic model
The collection robot and the anonymity rules, the taxonomy prompts and the classification chain, the validation flow and its release step, the readout, recognition and community cards, the weekly digest and the Power BI trend model
Microsoft Teams membership and card delivery through the UiPath Integration Service Teams connector; Forms responses and the recognition list through the OneDrive & SharePoint connector; organisational structure from the HR system and Microsoft Entra ID
How the automated process works
- AutomationThe pulse opens in Microsoft Forms with a link per unit, and a UiPath Robot collects responses on a schedule
- SystemThe robot strips identifiers and checks the minimum group size; units under the threshold are reported inside their parent
- AutomationGenAI Activities filter personal data from the comments, then score sentiment, classify against the HR taxonomy and draft a summary per unit
- PersonThe analyst reviews themes, quotes and reclassifications in UiPath Action Center and releases the summary
- AutomationPower BI updates the theme and score trend, and each team lead receives a readout card in Microsoft Teams
- PersonTeam leads read their own reasons, choose an action, and record it against the theme for the next pulse to test
- AutomationRecognition sent from a Teams card is tagged to a value, written to a Microsoft List and compiled into the weekly digest
- SystemThe community directory card adds a member to the relevant team in Microsoft Teams on one tap
Human-in-the-loop model
Automation handles
- Response collection per unit, identifier stripping and the minimum group size rule
- Personal-data filtering, sentiment, theme classification and the per-unit summary draft
- The Power BI trend, the 140 readout cards and the weekly recognition digest
- The community directory and the membership change that follows a sign-up
People decide
- Which themes and which quotes are published, reviewed before release rather than after
- The theme taxonomy, the survey design and the anonymity threshold
- What to do about a theme, and how the loop is closed with the unit that raised it
- Who is recognised and why, and who leads a community
Before and after
Systems and integrations
Every entry can be checked in vendor documentation. The evidence class is stated next to each one.
Inputs
- Microsoft Forms pulse responses per unit
- the HR theme taxonomy
- organisational structure from the HR system and Microsoft Entra ID
Automation layer
- UiPath Robots
- UiPath Orchestrator
- UiPath GenAI Activities
- UiPath Action Center
- UiPath AI Trust Layer
Target systems
- the Power BI theme and score model
- the Microsoft Lists recognition record
- Microsoft Teams community membership
Human touchpoints: the validation task in UiPath Action Center; the team-lead readout card; the recognition card; the community directory card
Technologies used
runs the pulse with a link per unit and delivers responses without a separate survey tool
Acollect responses, enforce the group-size rule, run the schedule and hold the audit trail
Afilter personal data, score sentiment, classify against the HR taxonomy and draft the per-unit summary
Athe analyst validates themes and quotes and releases the summary, from a task that can be completed in Teams
Areadout cards for team leads, the recognition card and the community directory
Athe recognition record by value, unit and date, retained under Microsoft Purview policy
Amodel allow-list, EU region routing and audit over every model call on employee text
Athemes and scores by unit across pulses, which is the trend leadership actually asks for
AIllustrative economic model
The arithmetic is open, so it can be argued with.
Reading is priced at 4 minutes a comment and assembling a team readout at 45, both ranges we see in HR teams rather than numbers from a client stopwatch. An engagement programme creates three blocks of desk work: 546 comments per pulse, 140 team readouts per pulse, and 180 recognitions a month compiled by hand at 6 minutes each, all at €39 an hour fully loaded. The calculator prices the largest block, the readouts, with the four-pulse cadence folded into the monthly volume so that 140 per pulse becomes 47 a month; the other two blocks carry their own arithmetic into the next section. Nothing here was measured at a client.
Run the numbers on your data
An illustrative estimate from your own inputs. It models released capacity; it is not a promise of savings.
Business benefits
- Leadership sees themes by unit within days of a pulse closing, because reading and coding no longer depend on how much time one analyst had
- Team leads get their own reasons instead of a company average, which is the level at which anything about engagement can actually be changed
- Employees see that comments are read and answered, so the next pulse gets a better response rate because writing one is visibly worth it
- Recognition becomes visible and recorded, so it carries past Friday and can be cited in a talent conversation months later
- Communities grow past their founders, because joining takes one tap rather than knowing whom to email
- Anonymity is enforced by a rule rather than by the care of whoever built the export, which is what makes the programme defensible
The management view
- Themes trend by unit across pulses, so the question moves from what the score is to whether the intervention worked
- Falling sentiment in a unit becomes an early signal rather than a finding in an exit interview
- Recognition by value and by unit is visible as a picture, which shows where the culture is being practised and where it is being described
- Time to published insight is short enough that the next pulse can ask what changed, which turns a survey into a loop
Board-level KPIs
Security and governance
Security is designed with the process, not after it.
- Responses are anonymous by construction: identifiers are stripped by the robot before any processing, and a unit below the agreed group size is never reported on its own
- Personal data is filtered out of the comment text before a model sees it, and no individual-level sentiment is stored anywhere in the chain
- Model calls run under the UiPath AI Trust Layer with a model allow-list, EU region routing and an audit record, inside the EU region of UiPath Automation Cloud
- Published summaries are versioned and attributed to the named analyst who released them, so a theme can always be traced back to what was validated
- Recognition entries are visible to the recipient, their manager and HR, retained under Microsoft Purview policy; the membership robot uses a least-privilege account with its secret in the Orchestrator credential store or Azure Key Vault
Why now
Boards have started asking for engagement evidence by unit and by trend, and an average across 1,600 people cannot answer that question in either direction
Language models classify and summarise free text reliably enough that a validated theme summary is routine work, provided a person still signs it off
Hybrid working across seven countries has made visible recognition and easy joining the difference between a network and an org chart, and the modelled €1,375 a month of readout assembly buys none of it today
Relevant executive roles
Engagement evidence arrives by unit and trend, early enough to act on, and every published theme has a named validator
The analyst moves from reading comments to acting on them, and 140 team leads get something they can use
Recognition and communities become recorded programmes with a picture attached, rather than a thread and an ageing intranet page
Common questions and objections
Identifiers are stripped and names are filtered out before any model call, no individual sentiment is stored, and an HR analyst validates every theme and quote before publication. The design goes to employee representatives before the first live pulse, not after.
For an annual instrument that may well be true. Quarterly pulses in Microsoft Forms have none, and the recognition and community pieces have no analytics at all; nor can a vendor trend them by unit alongside the pulse.
Some will, and the open rate by unit shows exactly which. That is a conversation a CHRO could not have when every manager received the same company slide.
When this is not the right solution
- Leadership is small enough to read every comment personally, in which case reading them is the better use of the time
- A dedicated engagement platform is already in place and its analytics are genuinely used
- Surveys run annually only, so the loop this builds has nothing to close between measurements
- Employee representatives object to automated processing of survey comments, and the classification step is precisely what would have to go
A question for the next management meeting
Employees wrote us several hundred reasons last quarter and this board discussed one average; which of the two should we be managing on?
Implementation approach
Delivery runs in stages, so it can be stopped at any point.
We deliver
- A taxonomy workshop that classifies two past surveys' anonymised comments and reviews the result with HR
- The collection robot, the identifier stripping and the minimum group size rule agreed with employee representatives
- The GenAI configuration: filtering order, classification prompts, summary format and quote selection
- The validation flow in UiPath Action Center, including what an analyst may change and what is versioned
- The Power BI trend model and the 140 readout cards, addressed by unit
- Recognition and community cards, the Microsoft List record and the weekly digest
- One pulse run end to end as a pilot, measured on corrections, validation time and readout opens
We need from you
- Anonymised comments from the last two surveys, with unit labels intact
- A taxonomy owner in HR and a named analyst who will validate each pulse
- Agreement with employee representatives on the design before the first live pulse
- The organisational structure that decides which unit a response belongs to
Stages
Discovery
Classify two past surveys, build the taxonomy with HR, agree the anonymity threshold
Design
Filtering order, prompts, summary format, validation rules and the readout card
Build
Collection robot, GenAI chain, Action Center flow, Power BI model and the Teams cards
Pilot
One pulse end to end, measuring corrections per hundred comments and readout opens
Rollout
Recognition and community cards, then the trend views across pulses
Optimisation
Taxonomy tuning, prompt adjustment, new questions as the programme matures
Quick win. Effort is driven by the state of the theme taxonomy, the number of languages the comments arrive in, and how long agreement with employee representatives takes.
If the next pulse changes nothing, the one after it will have fewer answers.
Send us one anonymised export of free-text comments from a past pulse, with the unit labels intact. We come back with the theme summary your leadership team would have received, and the taxonomy behind it.
Summarise one pulse of commentsThe neighbouring process usually has the same problem
Your CSAT score gives you a number. Your customers' emails give you the reason, and nobody reads them.
View solution Other solutionsGoals and ratings in Teams, not a hunt through the HR systemGoals live in email, ratings history is buried in the module, and HR chases completion three times a cycle.
View solution Other solutionsThe same right answer to every HR policy questionTwo advisors answer the same handbook questions all year, from different versions of the handbook.
View solution Other solutionsTraining and policy sign-offs tracked, reminded, evidencedYour evidence pack exists only in the week somebody builds it, and two people on the list left in March.
View solution Case studyThe HR agent on TeamsAn employee asks on Teams about leave, a certificate or a benefit.
View case study Case studyAutomated new-hire onboardingNew hires ready to work sooner — less paperwork for HR.
View case studyIndustries we deliver this in most oftenManufacturing & industryRetail & e‑commercePublic sectorServices & ITShared services