CRM automation cleanup first, automations second
A CRM is supposed to work for the sales rep, not be an expensive database someone retypes emails into. We clean up fields, pipeline stages, duplicates, lead sources and permissions, then wire in the automations: routing, scoring, follow-up, tasks and a nightly report. In the morning the rep sees a list of what to do today, not 80 deals with no order to them.
- 47%of newly created records carry at least one error that matters for the work. Study of 75 managers, Harvard Business Review, 2017
- 14 daysfrom the workshop to the first automation running in your CRM on real data
- 2 to 6 weeksfrom the workshop to a clean CRM with automations wired in, depending on the state of the database
The problem
Why did the CRM turn into an expensive database?
A form submission lands at 5:40 pm. The rep sees it the next morning, creates the deal by hand, copies the details out of the email and sets a reminder for tomorrow. A week of silence later someone should call, except the pipeline view holds 80 deals and nothing in it says which one is urgent. On Friday the manager asks for a report, so an hour goes into stitching numbers together in a spreadsheet. Nobody here is lazy. The system just does not work for anyone.
Underneath sits something no automation will fix for you. The industry field is free text, so the same industry is written twelve different ways. The stage called "negotiation" means a sent quote to one rep and a discount conversation to another, so the forecast from the pipeline is fiction. The same company sits in the database three times: under its registered name, under a branch name, and under whatever the client typed into the form. A rule built on data like that repeats the mistake faster than a person would.
The scale of this has been measured. In a study published in Harvard Business Review in 2017, 75 managers each checked the last 100 records created in their own company. 47% of those records carried at least one error that mattered for the work, and only 3% of the datasets cleared a basic quality bar. A CRM record typed in by hand from an email is exactly that kind of newly created record.
- Reps spend hours clicking through the CRM instead of talking to customers
- The forecast drifts away from reality, because a stage means something different to each person
- The same company sits in the database several times, so the contact history is split
- Nobody knows which campaign the client who signed actually came from
- The board report is assembled by hand and is out of date on the day it is sent
Why now
What changed, so that cleaning up a CRM is no longer a quarter-long project?
HubSpot, Pipedrive, Salesforce and Livespace all expose an API that reads the whole record set without anyone opening records one at a time. A field inventory, duplicate detection and a count of which fields are actually filled in take a few days on a copy of your data. That used to be a quarter of an intern's time, which is why nobody ever did it.
The second change: CRM automation stopped being one large project. It breaks into pieces that go in one at a time and can each be judged on their own. Lead routing assigns an enquiry in seconds, lead scoring sets the calling order, follow-up after a meeting keeps what was agreed out of someone's notebook, and CRM and ERP synchronisation ends double entry. This page sits one level above those. It is about giving them something solid to stand on.
The third change is the AI agent. A rule does what it was told, but it will not answer a sales director who asks how many deals have sat still for longer than 14 days. BOND, our company AI agent, answers that in a plain sentence in 5 to 10 seconds, with no CRM login and no report request.
What changes
What does the team get once the CRM is clean?
A stage means the same thing to everyone
Every stage gets an entry condition and an exit condition, written in one sentence. Only then does a weighted forecast mean anything, because a deal at the quote stage marks the same event for every rep.
Stage definitions on one page, signed off by the sales team, not by us
Fewer fields, more of them filled in
We count which fields are actually filled in and which have sat empty for two years, and we cut the second group. The fields that matter become required when the stage changes, not when the record is created.
Closed lists instead of free text, so industry and source can be filtered
One company, one record
We find duplicates by VAT number, email domain and fuzzy name match, so that "Kowalski Logistics Ltd" and "Kowalski Logistics" end up as one company. You approve the merge rules and we run them on a copy of the database first.
Three matching criteria instead of the name alone, which registers spell several ways
You can see where a signed client came from
The source is written when the record is created, not from the rep's memory: the channel of the last touch and the first contact with the brand, rebuilt from utm parameters. The conversation about ad budget stops being a conversation about impressions.
First touch and last touch in HubSpot, Pipedrive, Salesforce or Livespace
A report without the Friday spreadsheet
Pipeline by stage, weighted forecast and stalled deals are calculated from the CRM every night and land in the board inbox. Nobody copies numbers and nobody explains why two reports show a different total.
One source for the numbers: the CRM, not a sheet glued together from three exports
Pipeline questions in a plain sentence
"How many open deals above 50 thousand do we have", "which ones have sat still for longer than 14 days", "how many leads from the last campaign reached a quote". The AI agent answers from CRM data in 5 to 10 seconds.
An answer in plain English, with no CRM login and no report request
Proof
Do you run your own CRM this way?
Yes. Since March 2026 the CRM in our founders' agency has been run by an AI agent, the same one behind the numbers on our AI automation agency page. Below is the scale of that work, together with what the number does NOT prove.
Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof.
JustBottheir agentJustIdea Agency, a performance marketing agency of around 30 people
- 23
- people from the company work with the agent
- 38
- systems wired in permanently
- 112
- processes run without a human
- 475
- thousand lines of the agent’s code
The agent is wired into HubSpot, the website forms, Slack and the data warehouse. Records are created from form submissions, from ads and from referrals.
8 of 38 systems it works in
Google Ads
Meta Ads- HubSpot
- ActiveCollab
- Ahrefs
GA4
Search Console
BigQuery
- Before
- A person created the contact and the company in the CRM: read the enquiry, retyped the details from the email and checked by hand whether the company was already in the database. How complete the data was depended on who happened to have time.
- After
- The agent creates the contact and the company in HubSpot, checks for a duplicate by domain, fills in the company details and notifies the team on Slack. A person steps in at the conversation, not before.
3,432 operations on leads and 1,301 on form submissions in 30 days, with no human involved
And in your company
What if our CRM looks different?
Four sentences come up most often, and this is how we answer them.
If you are after a narrower piece, there are separate pages: CRM implementation and integration when there is no CRM yet, and HubSpot integration or Pipedrive integration when the question is about one system.
We run HubSpot and we already have automations in it.
Then we start by checking what those automations stand on. A workflow that splits leads on a free-text industry field works exactly as well as that field is filled in. Usually most of the rules can stay, and the ones worth rewriting are those reading fields with no closed list behind them.
Our CRM is a Google Sheet.
Then this is not CRM automation, it is an implementation, and we say so in the call. A spreadsheet can be automated for a while, but it has no permissions, no change history and no stages, so every automation stops where the cleanup should begin. It makes more sense to stand up a CRM once and come back to automations after that.
Our reps will not fill in even more fields.
We usually come out of the workshop with fewer fields, not more, because half of the existing ones have been empty for months. Company details are filled in automatically from the VAT number and the email domain, and the only required fields are the ones you cannot move to the next stage without. The rep types less than today.
We have 6 years of history in the CRM, that cannot be cleaned up.
It can, just not on production from day one. The inventory and the duplicate merge run on a copy of the database, following rules you approve, and we show you the result before anything is switched over. We do not delete records on impulse: we set a retention period for contacts that never entered the sales process, because <a href="https://eur-lex.europa.eu/eli/reg/2016/679/oj" rel="noopener">GDPR</a> requires a defined period rather than "forever".
- CRM
- HubSpot, Pipedrive, Salesforce, Livespace
- Lead sources
- Website form, Meta Lead Ads, LinkedIn Lead Gen Forms, live chat, Google Ads landing pages
- Team tools
- Google Workspace, Microsoft 365, Slack, Notion, ClickUp, Jira, Asana
- Sales and finance
- Shopify, WooCommerce, BaseLinker, Xentral, Xero, sevDesk, Comarch Optima
- Data and reporting
- BigQuery, Google Sheets, Google Analytics, SQL databases
- Automation layer
- Make, n8n, Zapier, custom API integrations
What we clean up in a CRM, and how you know it is missing
Six areas add up to CRM automation, plus reporting on top. The order comes out of the workshop, because a different area is the bottleneck in every company.
| Area | How you know it is broken | What changes after the cleanup |
|---|---|---|
| Fields and value lists | half the fields sit empty, industry typed by hand a dozen ways | a field list with an owner, closed value lists, required only on a stage change |
| Pipeline stages | two people hold a deal at a different stage with the case in the same state | a stage has an entry and an exit condition, so a weighted forecast can be calculated |
| Duplicates | the same company sits in the database 2 or 3 times under different names | matching by VAT number, domain and fuzzy name, merge rules approved by you |
| Lead sources | nobody knows which campaign the client who signed came from | the source is written when the record is created, first touch and last touch kept apart |
| Permissions and views | everyone sees the whole database and exporting every contact is one click | roles and views narrowed to what each person handles, export restricted |
| Tasks and deadlines | reminders live in the reps' calendars, not in the CRM | a task is created by a stage change, and an overdue one reports itself |
| Reporting | the board report is glued together in a spreadsheet on Friday afternoon | pipeline, forecast and stalled deals calculated from the CRM every night |
The rows are a starting point. In most companies the first phase covers three to four areas, and the rest goes in when it turns out to be needed.
How we work
Five steps from a database to a CRM that works
Each step ends with something you can judge before you decide on the next one.
Workshop
We break the sales process into stages: who does what, at which point they type something into the CRM and what they look for outside it, in email or in a spreadsheet. We check which systems you already run, including the sheet that quietly acts as a second CRM.
Inventory on a copy
We read the whole record set through the API and count the facts: how many fields are actually filled in, how many records are duplicates, how many deals have stalled, which stages have no definition. You get a report with numbers before anything changes on production.
Cleanup and rules
We cut the fields nobody uses, close the value lists, write down the stage definitions and the duplicate merge rules. We agree permissions, views and a retention period for records that never entered the sales process.
Automations one at a time
We wire the automations in one by one, each running alongside the current process first: routing, scoring, follow-up, tasks from a stage change, ERP synchronisation and the nightly report. The order comes from the workshop, not from what is easiest to build.
Measurement at 30 and 90 days
After 30 days we look at data completeness and whether the stages hold their definition. After 90 days we look at response time, the number of stalled deals and whether the forecast matches what actually closed.
Before and after
How the sales team's week changes
Before
- The morning goes on retyping the enquiry into the CRM and checking by hand whether the company is already there
- The pipeline view holds 80 deals and says nothing about which one to call first
- The forecast for the board is put together from memory, because stages do not mean the same thing
- The sales report is glued together in a spreadsheet on Friday afternoon
After
- The record is created from the enquiry, with company details filled in and the duplicate already checked
- The list of what to do today follows from the score and the dates held in the CRM
- The forecast is calculated from stages that have an entry and an exit condition
- The CRM report lands in the board inbox on Monday morning, with nobody working on it
The first automation runs 14 days after the workshop. We judge data completeness and the forecast at 30 days, and the effect on sales at 90.
For the person signing this off
What the sales director gets out of it
You stop running sales on whatever the team says at the weekly review. The forecast follows from stages that have a definition, not from optimism. When the quarter closes you know how many deals have stalled and since when, because that is a number in the system rather than a topic for discussion.
The second thing carries more weight in a board conversation. A clean CRM shows which campaign the client who signed came from. That is an argument in the budget conversation you do not have today, and it takes the risk off you when someone asks about the return on marketing spend. We start with a process workshop, not with building.
- A forecast calculated from stage definitions, not from what the reps declare
- Stalled deals and overdue tasks visible the same day
- The source of every signed client written in the CRM, ready for the budget conversation
- Automations go in one at a time, so each can be judged and switched off on its own
What it costs
The workshop has a fixed price, everything after it is quoted
The process workshop costs EUR 3,500 net per week and runs from one to seven weeks depending on scope. For a CRM it is usually one to two weeks, because the sales process touches marketing, customer service and finance. The cleanup and the automations are quoted after the workshop, once we know the state of the database and how many processes are in scope.
CRM licences and the cost of running the automations go straight to the vendor (HubSpot, Pipedrive, Make, n8n), or you pay nothing at all when the workflow runs on your own infrastructure. Where an AI agent is in scope, the model provider charges EUR 0.005 to 0.07 per query. No commission on top from us.
Process workshop
EUR 3,500 net
per week, usually one to two weeks for a CRM engagement
A self-contained piece of work. It ends with a map of fields, stages and duplicates, plus the order of automations you can execute without us.
- Inventory of fields and value lists: what is filled in, what sits empty
- Pipeline stage definitions with an entry and an exit condition
- Duplicate report and a proposal for merge rules
- Order of automations with metrics to measure at 30 days
- Most popular
Workshop and CRM cleanup
Quoted after the workshop
scope follows from the state of the database and the number of processes
The option most companies choose. A clean CRM with automations wired in, 2 to 6 weeks after the workshop, tested on a copy before production.
- Everything in the workshop
- Cleanup on production: fields, value lists, stages, permissions and views
- Duplicate merge following rules you approve
- Automations one at a time: routing, scoring, follow-up, tasks, nightly report
- Training for the team and for the person who will change rules without us
CRM and an AI agent for sales
Quoted after the workshop
for teams that ask about the pipeline a dozen times a day
Everything in the cleanup plus BOND, which answers questions about the pipeline in a plain sentence, from CRM data.
- Everything in the CRM cleanup
- AI agent wired into the CRM: pipeline, forecast, stalled deals, lead sources
- Alerts on stalled deals and overdue tasks
- Access for reps and for the director, an answer in 5 to 10 seconds
The workshop ends with a document that stands on its own without us: a map of fields and stages, a duplicate report and the order of automations. If the inventory shows there is nothing worth automating yet, because the sales process is still taking shape, we say so and the engagement ends at the workshop.
Frequently asked questions
What is CRM automation?
It is moving repetitive work off the sales rep and onto the system: creating a record from an enquiry, filling in company details, assigning an owner, scheduling the next contact, creating tasks from a stage change and generating reports. The condition is a clean CRM, because an automation runs a rule against the data that is actually in there. No rule works on a free-text field filled in twelve different ways.
Where does CRM automation start?
With an inventory: which fields are actually filled in, what the pipeline stages mean, how many duplicates sit in the database and where data leaks out of the CRM into a spreadsheet or an email thread. We read the record set through the API and hand you a report with numbers before we change anything on production. Only then do we put automations on top, because otherwise they repeat the existing mistake faster than a person would.
Which CRM systems do you work with?
HubSpot, Pipedrive, Salesforce and Livespace, because they have an API you can read from and write to, and hook an external workflow into. If you run a different CRM we check its API in the free consultation and tell you plainly what is possible and what is not. If there is no CRM yet, this is not the right service and you start with an implementation.
How long does a CRM cleanup and automation take?
The workshop takes one to two weeks, then 2 to 6 weeks of work depending on the state of the database and the number of processes. The first automation usually runs 14 days after the workshop, so the team sees an effect before the project ends. Merging duplicates in a database with several years of history is often the longest item, because the ambiguous cases have to be reviewed record by record.
Do we have to change CRM for this to work?
Almost never. In most companies the problem is not the system, it is how the system is configured and what has been put into it. We propose a change of CRM only when the current one has no API, or when the fields and permissions you need cannot be set up in it. We say that at the workshop, not after an implementation contract has been signed.
What will you do with old records and duplicates?
A report first: how many records are doubled and how we recognise them. The merge follows rules you approve, and it runs on a copy of the database first so you can see the result before production. We do not delete records on impulse: we set a retention period for contacts that never entered the sales process, because GDPR requires a defined period rather than an open-ended one.
How is this different from a CRM implementation?
An implementation answers the question of which CRM to stand up and how to connect it to your other systems. CRM automation starts where the system is already in place and in use, but demands too much clicking from people and gives back too few numbers you can trust. In practice we often do one after the other, and the free consultation settles which, not a price list.
Who changes the rules in the CRM after the rollout?
You do. Part of the work is training the person who will maintain it: adding a rep, changing a scoring threshold, adding a pipeline stage, making a different field required. We also leave a description of the rules in one place, so the knowledge does not sit in one person's head. We come back when a new process appears, not every time somebody in the team changes role.
Let us check what the CRM does for you today, and what you do for it
The free consultation takes 30 minutes. You tell us which CRM you work in, how many people use it and what hurts most: retyping data, the forecast, the report or the duplicates. We show you what can be cleaned up straight away and what needs a workshop first. You do not have to share access to any system at this stage.
We run at most two rollouts at a time, so that someone is actually sitting on each of them.
Book a free consultationPS. If somebody at your company glued a sales report together in a spreadsheet this week because the CRM would not give it up, that is exactly this problem. A clean CRM hands that report over every night, and gives the rep back the hours that go into clicking today. We start with the workshop, because with no stage definitions there is nothing to automate.