Kesey · Showing your work
Your pipeline numbers are probably right. Can you prove it?
Everything written about pipeline accuracy is about whether the number is correct. That is rarely the problem. The problem arrives when somebody outside your team asks where the number came from, and the honest answer is that a person typed it and no one can say which person, or when, or from what.
A number nobody can trace is a number nobody has to believe. Being right and being able to show your work are two different things. The first is whether the figure matches reality. The second is whether you can walk somebody back through it: what the number is, who set it, when, and what it came from. You can have the first without the second, and in a board meeting or a sale it is the second one you get asked for.
In a quiet year, answering those questions eats about 32 hours. In a year where somebody is buying you or auditing you, closer to 70, plus one line with no number on it at all, because you cannot buy back a history you never kept. The arithmetic for both is below.
Two questions
You are measuring one of these and being asked the other
Is the number right?
Is the number right. Is the close date realistic, is the amount current, has anybody touched this deal in six weeks. Every CRM hygiene project, every data quality dashboard and almost every article about pipeline accuracy is about this, and it is a real problem worth solving.
It is also the problem your team is best placed to solve on its own, because everyone involved already knows what the right answer looks like.
Can you show where it came from?
Who set it, when, and what it came from, and can you show that to somebody who has no reason to take your word for it. A completely correct figure with nothing behind it fails this, and the person asking is right to treat it as your opinion.
Nobody outside your company can check whether a number is right. They can only check whether you can show your work, so that is what they ask for.
These two get mixed up constantly, and it costs you a quarter of data cleanup that leaves you no better off on the day somebody asks. A tidied up field with no history is still a number a person typed. You made it more likely to be right and did nothing about whether you can show it.
The wider argument, which this page takes as read: build the tools, never the record. Tools are cheap to build and safe to throw away. The record has to hold up for people who did not build it, and being able to show your work is most of what holding up means.
Should you build your own CRM makes that case in full and costs a homemade record at about 256 hours a year. One row of that model is answering where a number came from. This page is that row, opened up.
The chain
Four links, and it is only as good as the weakest
What each kind of record can answer about one figure on one deal.
-
What the number is
The current value, as of a stated moment. Everyone passes this link, which is why it is the one that gets demonstrated.
Bought recordThe field, with an as-of timestamp that came from the system rather than from the person presenting.
Homemade recordThe field. The as-of is usually whenever the query was run, which is fine until two people run it on different days.
-
Who set it
A person, an automation or an import. This is where most homegrown systems stop, because nobody builds an audit trail for a database they already trust.
Bought recordNamed on the field, person or automation, because the platform records it whether or not anybody asked it to.
Homemade recordUsually nothing. Sometimes a last_modified_by that only survives until the next bulk update overwrites it.
-
When, and what it was before
The value's history rather than its current state. This is the link that decides whether you can answer a question about last quarter at all.
Bought recordField level history, kept by default and exportable, so a figure from March is retrievable in September.
Homemade recordAlmost never. History is a thing you build deliberately, and nobody builds it for their own system because they were there.
-
What it came from
The email, the call, the invoice, the signed document. The difference between a figure worked out from something and a figure somebody remembered.
Bought recordWhere the record is built from connected email, calendar and calls, the source is attached because the source is what created the row.
Homemade recordHeld by whoever was in the meeting. Which is a kind of paper trail, and it resigns.
Read down the right hand column: a homemade record answers the first one well and the other three out of somebody's memory. Nobody sets out to build it that way. You do not keep a paper trail for a system you trust, and you are right to trust your own system, until the day somebody who is not you has to.
Somebody's memory is fine until that somebody hands in their notice. What leaves with the person who built it is the same problem with a date on it.
Where it gets asked
Four rooms where this is the whole conversation
-
The board pack
Somebody asks why the number moved. The honest answer is often that the number did not move, the way it is counted moved, and the two are indistinguishable without a history. This is the cheap version: you lose an afternoon and some credibility, and nothing else happens.
"Last quarter this said 2.1. Is that the same 2.1?"
-
Diligence
The expensive version. A buyer or an investor does not audit your arithmetic, they test whether your account of your own business survives being checked, and they do it by sampling. Two or three figures they cannot trace is not a finding about those figures, it is a finding about the record, and it moves the price or the escrow rather than the slide.
"Show us how these twelve deals were categorised, and when."
-
Commission and quota
The one that happens weekly and costs the most in aggregate. A rep disputes a number, and without a chain the argument is resolved by seniority rather than by evidence. Everybody remembers who won. The next quarter's numbers get entered by people who learned something from that.
"That was closed in March. It is showing April."
-
An auditor, an insurer, or a customer's procurement
The one nobody plans for, because it arrives attached to something else entirely: a security review, a renewal with a large customer, a claim. The question is rarely about revenue. It is whether records are maintained in a way that makes them evidence, and a system whose answer is "ask Dave" is not.
"What controls are there on changes to this data?"
What it costs
A quiet year, and a year with one real request in it
A model with every input printed. Not a survey of anyone's numbers.
Nobody budgets for this, because it never arrives as an invoice. It arrives as somebody's afternoon. Here it is in hours, with the assumption under each line so you can argue with it.
- Traceability questions off the monthly pack 6 hours One question a month at half an hour to chase down. Most months there are none and some months there are four.
- Forecast and commission disagreements 10 hours A number disputed roughly every fortnight, half an hour each to settle. Settled by argument rather than evidence costs less time and more trust.
- One-off asks from finance, a customer or a partner 16 hours The remainder. Two full days a year of somebody reconstructing where a figure came from, which is the conservative reading.
- Everything above still happens 32 hours A diligence process does not pause the monthly pack, and in practice it makes the questions about it sharper.
- The request itself, in a six week window 38 hours Fifty questions at forty five minutes. Fifty is the low end of a real data request and forty five minutes assumes the answer exists somewhere.
- Reconstructing history that was never kept no figure Blank on purpose. You cannot go back and build a history you did not keep. All you can do is work out today's number again and ask them to take your word for it, which is the thing they were questioning.
At a loaded $125 an hour that is roughly $4,000 a year in the quiet case and $8,750 in the loud one, and the loud one arrives in a six week block on top of whatever else is happening that quarter. The 32 hour line is the one carried into the annual model on the build versus buy argument, deliberately at the quiet figure, so that total stays the conservative reading rather than the dramatic one.
The third line is the one that matters, and it has no number on purpose. Hours you can pay for. A history you never kept is not for sale at any price or any speed, and the only time you can fix it is before anybody asks.
What changed
Two things you can check yourself, and one you cannot
The first two are the vendor's own claim, read September 2026. The third is structural and needs no citation.
-
Field level history is kept by default, and exports
HubSpot keeps a history on every property and will export it, so a figure carries the person or automation that set it and the date it was set. That single capability answers two of the four links in the chain above, and it is on by default rather than something anybody had to decide to build. A homegrown record almost never has it, because you do not build a history for a database you already trust.
-
The record can be built from the work rather than typed after it
Day.ai creates companies, contacts and opportunities out of connected email and calendar, including history from before signup. That matters here for a reason that is not about convenience: a value produced by the work carries the artefact it came from, where a value somebody typed carries only the typing. It is the fourth link in the chain, and it is the one no amount of data hygiene can retrofit.
-
A typed number and a derived number are not the same evidence
No vendor claims this, so there is nothing to link. Two figures can be identical and carry completely different weight: one is somebody's opinion recorded at an unknown moment, the other fell out of something that demonstrably happened. People outside your company discount the first automatically, and they are right to. No amount of tidying changes which kind you have.
Vendors move and their own pages sometimes disagree with each other, so treat both cited lines as dated rather than permanent and check them before deciding anything on them. The four CRMs compared carries the fuller version, each claim sourced the same way.
If you are keeping the build
Three things that are cheap, real, and worth doing this month
Plenty of people reading this are not going to move, and should not. A homegrown record that fits the business and is actually looked after is a reasonable thing to own. If that is you, these three close most of the gap, and none of them is a project.
-
An append only change log on the six fields anybody ever asks about
Not every field. Amount, stage, close date, owner, and whichever two are peculiar to your business. Every change written to a table that nothing is allowed to edit or delete, with who and when. That is an afternoon of work, and it moves the second and third questions above out of somebody's memory. Six fields rather than all of them because the full version is a project and gets put off, and six fields ships this week.
-
A written rule for which system wins
When the CRM and the billing system disagree, which one is right, and in which situations. Somebody knows. It is almost never written down, because the person who holds it has never had to state it out loud, and it is the first thing that becomes unanswerable when they move on. One paragraph per disagreement, written before the argument rather than during it.
-
An as-of date on every number that leaves the building
Every figure in a board pack, a report or an email says what day it was true. Costs nothing, and it kills a whole category of argument: two people with different numbers usually have the same number from different days, and without the date they will spend an hour finding that out. It also makes it obvious when a number is stale, which is uncomfortable at exactly the right moments.
Where Kesey comes in
We find out which links you have before somebody else does
We are a revenue operations practice and a Day.ai Solutions Partner. The useful version of this work is unglamorous: take a dozen real figures off your last board pack, try to trace each one back to where it came from, and write down where it breaks. About a week. It is the same exercise somebody's analyst will eventually run on you, except the findings come to you first and cost you nothing but the week. Then either we close the gaps in what you have, or we move the record onto something that keeps the trail without anybody having to remember to. The tools you built stay yours either way.
What this page does not claim. No case study and no percentage. The hours are a model with its assumptions printed rather than a survey of anybody, and the two vendor claims are dated and checkable on their own pages. If your own numbers come out lower, they come out lower.
And if you run it and everything traces, that is a real answer and a good one. Knowing that is worth the week on its own, because the other way to find out is in a room where the cost is not measured in hours.
Questions
Frequently asked
How do you prove where a number came from?
You answer four questions about it: what the number is, who set it, when and what it was before, and what it came from. Platforms that keep a history on every field answer the middle two without anybody deciding to build that. The fourth is answered by records built out of your email, calendar and calls, where the number falls out of something that actually happened rather than something somebody typed. Some people call this data provenance or an audit trail. It is simpler than either name makes it sound.
Our numbers are accurate. Is that not enough?
It is enough for you and not enough for anybody else, because nobody outside your company can check whether a number is right. All they can check is whether you can show your work, so that is what they ask for. A quarter of data cleanup makes your numbers more likely to be right and does nothing about this, because a tidied up field with no history is still a number a person typed.
What does it cost us to not be able to show our work?
On the model set out on this page, about 32 hours a year in a quiet year, and around 70 in a year where somebody is buying you or auditing you. At a loaded $125 an hour that is roughly $4,000 and $8,750. There is a third line with no figure on it on purpose: a history you never kept cannot be built afterwards at any price. All you can do is work out today's number again and ask them to take your word for it.
What do buyers actually ask for when they are buying you?
They sample. A handful of deals, traced from the figure in the pack back to whatever created it, with dates. They are not checking your arithmetic; they are checking whether your account of your own business holds up when somebody pulls on it. Two or three figures you cannot trace reads as a problem with the whole record rather than with those deals, which is why it moves the price rather than a slide.
Can we fix this without replacing our CRM?
Mostly, and it is worth doing. A log that only ever adds rows, covering the six fields anybody ever asks about. A written rule for which system to believe when two of them disagree. And a date on every number that leaves the building. That is an afternoon, a paragraph and a habit, and it closes most of the gap from here on. What it cannot do is recover the history from before you started keeping it.
Which CRMs keep a history of every change?
HubSpot keeps a history on every field and will export it, which is the most directly checkable example. Day.ai, Clarify and Attio each build the record out of connected email, calendar and calls to different degrees, which answers a different question: what the number came from in the first place. The four CRMs compared carries the detail with each claim sourced.
Is this a reason to move off a homegrown CRM?
It is one thing to weigh rather than the decision. The underlying argument is to build the tools and never the record: tools are cheap to build and safe to throw away, and the record has to hold up for people who did not build it. Should you build your own CRM sets it out in full.
A number nobody can trace is a number nobody has to believe.
Take a dozen figures off your last board pack and try to trace each one. A week of that tells you more than a quarter of data cleanup, and it is the same exercise somebody else will eventually run on you.
Book a callThe hour and dollar figures on this page are a model with its inputs printed rather than a survey of any particular company, and the 32 hour quiet year figure is the same line carried into the annual cost model on Should you build your own CRM. Substitute your own numbers; if yours come out differently, the conclusion may too.
Product capabilities described here reflect Day.ai and HubSpot as their own published pages read in September 2026. Vendors change them, so check their documentation for anything your decision turns on.
Kesey is an independent revenue operations practice and a Day.ai Solutions Partner. It is not Day AI, Inc., and it is not affiliated with Clarify, Attio or HubSpot.