Ask ten sales managers what a CRM system is and you get ten answers, most of them the name of a product somebody once used. The plain version is duller and more useful. It is a shared client base in which every customer has a card, and the card carries the history: who called, what was agreed, what was promised, and when the next step falls due.
That definition is worth holding on to, because arguments about CRM software often turn out to be arguments about something else. This is not a buyer's guide and we neither sell nor implement CRM systems. It is a plain description of what the category does, where it stops, and which part of the job is still waiting for somebody to pick up a handset.
Where this comes from
Ten years of running call centres, up to fifteen hundred live operators at peak, across France, Mexico and beyond. That history is where the operating figures below come from — how little of a call archive anybody ever listens to, what a floor does with a queue it cannot clear. Everything attributed to a named publisher is cited in the sentence that uses it. None of it is research and none of it is an industry standard.
What a CRM system actually is
The acronym stands for customer relationship management, which explains nothing to anyone who has not already worked with one. The difference it makes is easier to see from either side of a rollout. Before: what is known about a customer lives in a manager's head, in their personal phone and in a notebook. After: it lives somewhere a colleague covering a shift, a director pulling a report and a new hire on their second day can all open.
There is a single question that settles whether a team needs one. What happens to the customer base if the best salesperson resigns tomorrow? If the honest answer is that half the contacts walk out with them, the problem is real, and a shared base is what closes it. If the honest answer is that nothing much changes, the team is probably small enough that a well-kept spreadsheet is still doing the job.
Everything else that gets sold under the same name — reporting suites, document builders, scoring models, forecast dashboards — sits on top of that base. It inherits whatever quality the base has, which is the reason so many handsome dashboards describe a business nobody recognises.
The four working parts
Interfaces vary wildly and the set of working parts does not. From the free tier to a heavy corporate build, the same four things are underneath.
- The customer card. Company, contact, phone number, where the enquiry came from, who owns it, and whatever billing details the business needs to raise an invoice.
- The touch history. Call recordings, correspondence, quotes and notes, in order, so anyone can see who spoke to this customer last and how it ended.
- The deal pipeline. Named stages from first enquiry to payment. Every deal stands on one stage, and a glance across the stages shows where work is piling up.
- Reminders and automation. A task to call back on a date, a stage that advances itself, a confirmation that goes out after payment, an alert when a task runs past due.
Those four decide whether the rest is worth anything. When the card is half empty, the history has gaps and the stages mean different things to different people, the reports on top are not wrong so much as meaningless — they describe the data-entry habits of the sales team rather than the state of the business.
Reports built on a carelessly kept base do not lie to you. They faithfully describe how people fill in fields.
Where a CRM earns its keep
There are four places where a CRM pays for itself, and each one can be counted on a team's own numbers rather than taken on faith.
The first is enquiries that never reach anybody. A form on the site, an email and a message in a chat app land in one queue with an owner and a deadline instead of three inboxes with none. This matters more than it sounds. Harvard Business Review, drawing on MIT lead-response research, puts a company that calls a new lead within five minutes at 21× more likely to qualify it than one that waits thirty; separately, Harvard Business Review found that only 1% of the B2B companies it studied responded inside that five-minute window. A queue with an owner is the cheapest thing anyone has ever done about that number.
The second is touches nobody remembers to make. A B2B deal rarely closes on the first conversation. A manager holds the hottest few customers in their head and the rest cool off quietly. Salesforce puts the share of marketing leads that never convert at 79%, most of it poor follow-up; an industry compilation reports that 48% of salespeople never make a single follow-up attempt, and the same kind of compilation puts 80% of closed sales at the fifth touch or later. Both of those last two are compiled rather than independently published, which is worth saying out loud, but the direction they point in matches every call archive we have ever opened.
The third is simply being able to see. A director who can read deal counts by stage, average value and stated reasons for loss is having a different conversation with the team than one working from impressions. The pipeline view is not a management fashion; it is the difference between asking why last month was soft and pointing at the stage where the deals are piling up.
The fourth arrives later, once the base has some age on it: repeat business. Customers who bought a year ago, who said no on price, who went quiet mid-conversation — they accumulate, and one filter pulls them out as a list. Then the list sits there. Not because nobody sees the value, but because the same people who would work it are busy answering today's enquiries.
What a CRM does not do
Here is where the biggest misunderstanding lives. The system stores data, shows the pipeline and reminds somebody of a task. Then it stops. It does not dial the number. It does not carry the conversation. It does not ask what the customer's timeline is, or what they were comparing you against, or who else signs off.
Which produces a picture familiar to anyone three months past a rollout. The base is populated, the pipeline is drawn, reports go out every Monday, and conversion has not moved. It has not moved because a reminder that says call Miller back turns into a call only when a human being has both the time and the appetite to make it.
One filter will tell you whether that is your situation. Open the task list and show the callbacks past their due date. Whatever number comes back is the size of the gap, and the system sent every one of those reminders on time. Nothing there is a software fault. The software did its whole job.
Telephony integration, and where it stops
Telephony is normally the first integration anyone connects after the basic setup, and it earns its place. The number dials from the card in one click. An incoming call pops the right card onto the screen before the manager says hello. The recording and the duration drop into the history by themselves. A missed call becomes a task with an owner and a deadline instead of a line in a log.
Then the manual zone starts. The recording is in the card; what was said in it stays unknown until a person listens to the whole file. In our own operations, under 5% of recorded calls are ever reviewed by hand — that is our operating figure from running live floors, not a published study — and a team leader with ten people reporting to them has no realistic path to a higher number.
So decisions get made on the summary line the manager typed afterwards. Too expensive. Thinking about it. Call back in March. Each of those is a compression of several minutes of conversation into a few words chosen by the person with the most reason to compress them charitably, and it is what the pipeline stage, the loss reason and next quarter's forecast are all built from.
There are two honest ways to close that gap. Transcribe every conversation and score it automatically, so the content of the call reaches the card rather than a note about the call. Or hand the first contact to an agent that asks the questions in a fixed order and fills the fields itself. They are not alternatives so much as two halves of the same fix.
Small, mid-sized and large
What a company should actually buy depends on two numbers it already knows: how many deals pass through in a month, and how many people touch each one. Everything else in the vendor comparison is downstream of those.
For a small business the job is to stop losing things. A handful of people, one pipeline, telephony and the website wired in, and as few fields on the card as the work genuinely needs. The characteristic mistake here is not buying the wrong system — it is configuring a good one into uselessness with mandatory fields that everyone then learns to fill with a full stop.
A mid-sized company adds pipelines rather than fields: new business, repeat business, service. With them come access rights, end-to-end reporting from ad spend to payment, and a real requirement that people fill things in — which is a management problem wearing a software costume.
A large company is doing integration work: the accounting system, stock, billing, sometimes a data warehouse behind all three. At that point the CRM is one component of an IT landscape and the project is measured in months. The four working parts have not changed; the number of systems that have to agree about them has.
Why rollouts fail
The failure patterns repeat from company to company with almost no variation.
- Bought and trained, never governed. The licences are paid, the training happened, nobody wrote the one page that says what must be filled in and when. People keep calling from their personal mobiles and the cards stay half empty.
- A pipeline drawn for the presentation. Stages with names like interest confirmed, which nobody can distinguish from the stage before it, so deals get moved on instinct and the pipeline view stops describing anything.
- Reports read, recordings never opened. The director lives in the dashboard and never listens to a call, so everything about what customers actually ask and where conversations actually break stays locked in audio nobody has time for.
- Marketing outruns the sales floor. Enquiries arrive faster than the team can work them, the system dutifully stacks them in a queue, and by Friday the queue has gone cold. This one is not a rollout failure at all. The constraint is the number of conversations that fit in a working day.
The cure is boring, which is why it is usually skipped: one page of rules instead of a manual, a short list of mandatory fields instead of a long one, and a few deals reviewed by hand every week — recordings included, not just cards. And somebody has to own the unworked queue by name, before the next campaign doubles it.
Four numbers that say whether it is working
Four measures come out of any CRM in a couple of minutes and, between them, say whether the rollout took.
- The share of enquiries that have an owner assigned within the first hour.
- Time to first response on an inbound enquiry.
- The share of closed-lost deals that have a reason filled in.
- Overdue tasks per manager.
Take them on the same day every week. A monthly reading hides the collapses that show up on a weekly one. When the first three rise and the fourth falls, the rollout worked. When the picture is the other way round, the cause is rarely the software: enquiries are arriving faster than the team can answer them, and no amount of configuration adds hours to a day.
Who dials the numbers in the CRM
Our product starts exactly where the system has finished its part. The task exists, the number is on the card, somebody has to dial. That somebody is an AI voice agent on the phone. We do not sell CRM systems and we do not implement them; we take the calling.
In practice it runs like this. An enquiry lands in the CRM and the agent calls it in the first seconds rather than the first afternoon, establishes what the customer wants, what their budget and timeline look like, answers the questions that come up every day, and hands a qualified contact back to the card with a transcript attached. A cold list is worked the same way: the agent goes through it end to end, and the sales team only ever sees the people who agreed to keep talking.
The other half of the pairing is what happens to the recordings. Locator, our speech analytics, scores 100% of conversations against 30+ parameters rather than the handful a supervisor can listen to, and returns the content to the card: stated reasons for loss, the objections that keep coming back, the places where the script was abandoned. The card stops saying a call happened and starts saying what was in it.
Rates are published per minute and the calculator prices a volume before anyone talks to us. Training the agent on your product and your script takes about three days, and a business check-up comes before that to gather the scenarios. The agent is configured against your own recordings and your own pipeline stages rather than built from scratch, and what it plugs into is the base you already have.
A CRM records the intention to call. Everything after that is somebody picking up a handset — or something that does it in seconds, every time, on every card.