Automatic feedback collection the survey fires on the day of the event
A task closed in ClickUp, an event finished in the calendar, a ticket closed in the helpdesk: each of those events sends a short survey by itself. One question about how likely the client is to recommend you, plus 2 or 3 questions matched to the stage. Answers land in one place, and a score below your threshold reaches the account owner in Slack the same minute.
- 2 to 4 weeksfrom the workshop to a working survey, a Slack alert and a dashboard with the results
- 1 questionthat is all it takes to calculate how likely a client is to recommend you. The rest of the survey is 2 or 3 open questions, written separately for each stage
- 5 to 10 secondshow long the AI agent takes to answer a question about scores, with nobody opening the dashboard
The problem
Why do you find out a client was unhappy from their notice of termination?
The survey goes out when somebody remembers it. Usually two weeks after the project was handed over, sometimes never, because the account owner had three other urgent things that week. The score gets worked out as an occasion, once a quarter or once a year, from a spreadsheet that two people fill in using two different formats.
An answer collected late tells you about something other than you think. A client asked three weeks after a rollout is no longer rating the rollout. They are rating what happened in those three weeks: the invoice, an outage, a change of account owner. The answer closest to the truth comes in the first hours after the event, while the client still remembers a specific thing rather than a general impression of the whole relationship.
The measurement itself can rest on a single question. Frederick Reichheld described the NPS metric in The One Number You Need to Grow, Harvard Business Review, December 2003: asking how likely somebody is to recommend the company to a friend tracked growth better than multi-page satisfaction studies did. The score is simple to work out, as the share of 9s and 10s minus the share of scores from 0 to 6. A short question has one more advantage: you can ask it after every event, rather than once a year.
- The survey goes out two weeks after the event, or never goes out at all
- You hear about quiet dissatisfaction during the renewal conversation
- Results sit in several spreadsheets kept by different people
- One number a quarter does not say which stage of the work is dragging it down
Why now
What changed, so that a survey can go out on the day of the event?
The systems your team closes its work in have webhooks now. A task moved to the done column in ClickUp or Asana, an event closed in Google Calendar, a ticket closed in the helpdesk, a course finished on an e-learning platform: each of those can be picked up a second after it happened. Nobody has to remember it and nobody has to click anything through.
The second change is on the data side. Answers from Typeform, Google Forms or an HTML form in your brand colours land in BigQuery or Google Sheets, and Looker Studio draws the score trend and the spread of ratings by event type out of that. You do not need a data warehouse to see that training sessions score higher than rollout handovers.
The third change is the AI agent. A dashboard answers the question somebody anticipated when they built the filter. BOND, our company AI agent, answers a question asked in a plain sentence: which clients rated us below 7 in the last 30 days, what the average score from rollouts was this quarter. The answer comes back in 5 to 10 seconds, out of data you are collecting anyway.
What changes
What does the client service team actually get?
The survey fires on the day of the event
The workflow listens for events in your task system, your calendar and your helpdesk, and sends the form as soon as the work is closed. Nobody has to remember that a project finished.
Seconds from the event to a sent survey, instead of two weeks
Separate questions for a separate moment
After a rollout handover we ask about different things than after a training session or a closed service ticket. One question about likelihood to recommend, plus 2 or 3 open questions matched to the stage.
Six moments in the client lifecycle, six sets of questions (table below)
An alert before the client writes their notice
A score below the threshold agreed in the workshop reaches the account owner in Slack the same minute, with a link to the full answer and the history of earlier scores from that company. The response fits inside the same day.
A notification within a minute of the form being sent, with that client's score history
Results in one place, not in four spreadsheets
Answers land in BigQuery or Google Sheets, and a Looker Studio dashboard shows the trend and the spread of scores by event type. No more merging two spreadsheets into one before the quarterly review.
One dashboard instead of spreadsheets kept by different people
You see a recurring theme, not one bad day
Open comments are grouped by event type, so a phrase that keeps coming back is easy to tell apart from a single bad opinion. Three similar comments about the same stage are a signal about the process, not about a person.
The spread of scores and comments by event type, instead of one number a quarter
Questions about scores in a plain sentence
Who rated us below 7 in the last 30 days, what the average score from rollouts was this quarter, what keeps coming back in comments after training. The AI agent answers from the same data, with no dashboard open.
An answer in 5 to 10 seconds, with context from that client's earlier surveys
And in your company
What if our clients do not answer surveys?
Four sentences we hear most often, and what we say back to them.
Our clients do not answer surveys.
Usually because the survey turns up three weeks later and has twelve questions. We send one question on the day of the event, from the account owner by name, answerable in one click. If nobody is answering after 30 days, we say it plainly: the problem sits in the timing and the wording of the question, not in the tool.
We measure NPS once a year and that is enough for us.
If you need one number for a presentation, it genuinely is enough, and we will tell you so. Automation pays off when you want to know which stage of the work is spoiling the score: the rollout handover, the training or the service tickets. A yearly measurement cannot be split into stages after the fact, because the client is rating everything at once.
We have no IT department and no data analyst.
You do not need either. The webhook, the connection to the spreadsheet and the dashboard are on us. What we need from you is one decision: at which moments of the work you want an honest opinion, and who reacts to a low score. Everything from the event in the system to the alert in Slack is ours to build.
We are worried the automation will send a survey at the wrong moment.
That is why the workshop sets the exclusions and the frequency cap. A contact with an open complaint or an invoice dispute does not get asked how likely they are to recommend you. The same client does not get a survey twice within 30 days, and cases outside the rule go into a queue where the account owner decides.
- Event sources
- ClickUp, Asana, Jira, Google Calendar, helpdesk systems, e-learning platforms
- Forms and surveys
- Typeform, Google Forms, an HTML form in your brand colours
- CRM
- HubSpot, Pipedrive, Salesforce, any CRM with an API
- Data and dashboard
- BigQuery, Google Sheets, Looker Studio
- Notifications and teamwork
- Slack, Google Workspace, Microsoft 365, Notion
- The automation layer
- Make, n8n, Zapier, custom API integrations
Moments where a survey makes sense, and the events that trigger it
Six moments that go into a first rollout most often. The order is set in the workshop, because they do not all carry the same amount of information about retention.
| Moment in the client lifecycle | Event that triggers the survey | What we ask about |
|---|---|---|
| Rollout handover | task moved to done in ClickUp or Asana | likelihood to recommend, and the one thing that went worst |
| End of a training session | event closed in the calendar or a course finished on the platform | the trainer, and whether the knowledge is usable at work |
| Closed service ticket | ticket moved to the closed status in the helpdesk | response time, and whether the problem came back after the fix |
| First month of a retainer | 30 days from the contract start date, counted in the CRM | whether the scope is clear, and what is missing from the reporting |
| Before a renewal | 60 days before the contract end date in the CRM | likelihood to recommend, and what would have to change |
| Lost client | opportunity marked as lost in the CRM | the reason for the decision, and the moment it was made |
The rows are a starting point. A first rollout usually covers two or three moments, and the rest go in once it is clear the data from the first ones is genuinely being used.
How we work
Five steps from a survey sent from memory to a survey that sends itself
Process first, tools second. Every step ends with something you can judge before you decide on the next one.
Workshop
We walk your client lifecycle step by step and mark the events where feedback actually changes something. Separately, the ones where an opinion is already collected today, only by hand and irregularly. We set the alert threshold and the person the escalation goes to, so it does not drown in a shared department inbox.
Designing the solution
We pick the event source, the survey tool and the place the results live: BigQuery at larger data volumes, Google Sheets where a spreadsheet with filters is enough. We write the questions separately for each moment and agree the exclusions, so nobody gets a survey in the week they are filing a complaint.
Build and testing
We build the workflow in Make or n8n and test it against events from your last few weeks before it goes to production. We check that the trigger fires exactly once, that the question makes sense without the context of a previous conversation, and that the alert reaches the right Slack channel.
Team training
We show account owners how to read the dashboard and how to react to a low score alert before it turns into a quiet cancellation. The person responsible for client service learns to change the threshold, the wording of a question and the exclusion list on their own, without raising a ticket with us.
Measurement at 30 and 90 days
At 30 days we look at the response rate and at whether the alert threshold is shouting too often. At 90 days there is a trend, and you can see whether the score from a specific stage is moving. We tune the moments and the questions to what actually comes back from clients, rather than to what we assumed in the workshop.
Before and after
How the month changes for the person running client service
Before
- The survey goes out when somebody remembers it
- Client dissatisfaction surfaces during the renewal conversation
- Before the quarterly review somebody merges two spreadsheets into one
- There is one score for the whole company and it does not say what pulled it down
After
- The survey fires by itself, on the day of the event, with no person involved
- A score below the threshold reaches Slack the same minute
- The trend and the spread of scores sit in a dashboard, kept current
- The score is split by stage: rollout, training, service tickets
You see the first answers in the week you switch over. We judge the response rate and the alert threshold at 30 days, and the score trend at 90.
For the person signing this off
What the person responsible for client service gets out of it
The question of why that client left stops being a conversation about impressions. You have a date, a score, a comment and the moment the signal showed up, and alongside it a record of who reacted to it and when. That turns the conversation with the board from explaining yourself into showing your work.
The second point is about you personally. Losing a large client is rarely a surprise to everybody, usually somebody on the team knew and had nowhere to write it down. The automation records the signal regardless of who is on duty that day, so responsibility for reacting is attached to a name rather than spread thin.
- A dissatisfaction signal with a date, a score and a comment, not from the account owner's memory
- An alert assigned to a person, not to a shared department inbox
- The score split across the stages of the work, ready for the quarterly review
- You change the alert threshold and the wording of the questions yourself, with no ticket to us
What it costs
The workshop has a published price, the rollout is quoted after it
The process workshop costs EUR 3,500 net per week and runs from 1 to 7 weeks. Feedback collection usually needs a single week, because the process is narrow: from the event to a saved answer. We quote the rollout after the workshop, once it is clear how many moments are in play and how many systems the events have to come from.
You pay for the tools directly to the provider: Typeform, Make, the cloud under BigQuery. On your own infrastructure, n8n costs nothing beyond the server. Same with the language model once an AI agent goes in, where a single query costs the provider roughly EUR 0.005 to 0.07. No margin from us on any of it.
Process workshop
EUR 3,500 net
per week, and feedback collection usually needs one
A self-contained piece of work. It ends with a map of the moments, the wording of the questions and the escalation path, which you could put in place without us.
- A map of the client lifecycle with the survey moments marked on it
- The wording of the questions for each moment, together with the exclusion list
- The alert threshold and an escalation path with a named owner
- A tool recommendation: Typeform, Google Forms or a form of your own
- Most popular
Workshop and survey rollout
Quoted after the workshop
scope follows from the number of moments and source systems
The option most companies choose. A working workflow 2 to 4 weeks after the workshop, tested on your own events from the last few weeks.
- Everything in the workshop
- A workflow in Make or n8n, wired into the task system, the calendar and the helpdesk
- Forms in your brand colours and answers written to BigQuery or Google Sheets
- Slack alerts below the threshold, with the client's score history
- A Looker Studio dashboard and training so the team can read it
Feedback and an AI agent
Quoted after the workshop
for teams that ask about scores several times a week
Everything in the rollout plus BOND, which answers questions about scores in a plain sentence, with no dashboard open.
- Everything in the survey rollout
- An AI agent wired into the answer store and the CRM: who scored low, what the score is per stage
- A summary of the open comments from the last 30 days
- Access for account owners and for the board, in English and in Polish
The workshop ends with a document that stands up without us: the map of moments, the wording of the questions, the alert threshold and the escalation path. If it turns out that one question in an email after project handover and a template in Google Forms would do the job, we say so and show you how to set it up. The matter then ends at the workshop, rather than at a rollout you do not need.
Frequently asked questions
What is automatic feedback collection?
It is a workflow that listens for events in your systems and sends a short survey after each one, with no person involved. A task closed in ClickUp, an event finished in Google Calendar, a ticket closed in the helpdesk: each of those triggers a question matched to the stage. Answers land in one place, and a score below the threshold goes to the account owner as an alert.
How long does a rollout of this automation take?
A week of workshop and 2 to 4 weeks of rollout. The first working survey usually goes live 14 days after the workshop, because it covers one moment and one event source. Further moments go in later, once it is clear the data from the first one is genuinely being used in conversations with clients.
How do you calculate NPS?
From a single question about likelihood to recommend, on a scale from 0 to 10. Scores of 9 and 10 are promoters, scores from 0 to 6 are detractors, and 7 and 8 do not count towards the result. NPS is the share of promoters minus the share of detractors. We calculate the same metric separately for each event type, so you can see which stage of the work is dragging it down.
What if the client does not answer the survey?
We send one reminder after a few days and stop there, because a third message damages the relationship more than it helps the measurement. A non-answer is recorded too, so you can see the response rate for each moment separately. When it is very low after 30 days, we go back to the wording and the send time rather than piling on more email.
Can a survey reach a client at the wrong moment?
That is why the workflow has an exclusion list and a frequency cap, both set in the workshop. A contact with an open complaint or an invoice dispute does not get asked how likely they are to recommend you. The same client does not get a survey twice within 30 days, and cases outside the rule go into a queue where the account owner decides.
Where do the answers land and who has access to them?
In BigQuery at larger volumes, or in Google Sheets when a spreadsheet with filters is enough. A Looker Studio dashboard shows the trend and the spread of scores, and you grant access the same way you would for any other report in Google Workspace or Microsoft 365. The data stays in your systems, not in ours, and processing follows <a href="https://eur-lex.europa.eu/eli/reg/2016/679/oj" rel="noopener">GDPR 2016/679</a>, in particular Art. 5(1)(c) on data minimisation.
How is this different from a survey sent by hand once a quarter?
The timing, and the split by stage. A manual survey asks about the whole relationship after the fact, so the answer blends the rollout, the service and the last invoice. The automation asks on the day of the event and attaches the answer to that event, which is what lets you break the score down by stage. The second difference is the reaction: the alert goes out within a minute, rather than after a quarterly review.
Why add an AI agent when there is already a dashboard?
A dashboard answers the question somebody anticipated when they built the filter. An AI agent answers a question asked in a plain sentence, for example which clients rated us below 7 after a rollout in the last 30 days, and it does that in 5 to 10 seconds. BOND takes around 2 months to put in place and it is a separate decision, not a condition for starting the surveys.
Let us find the moment where you lose the signal from your client
The free consultation takes 30 minutes. You tell us what your client lifecycle looks like and when you ask for an opinion today, even if the answer is once a year and irregularly. We tell you which moments a survey can cover straight away, and which ones need the CRM data tidied up first. You do not have to share any systems before the call. You are writing to Pawel, the founder, not to a form handled by a bot.
We run at most two rollouts at a time, so that someone is genuinely sitting on each of them.
Book a free consultationPS. If you found out this year that a client was unhappy from their email giving notice, that is exactly this problem. A survey sent on the day of the event gives you a chance to react while the client is still asking, rather than once they have decided.