ALYVO
← All posts

What is a CRM, and where did it come from?

CRM was never really about software. It’s a question: “How do I remember my customers, and do the right thing for them at the right time?” Software is just what the answer looked like in different decades.

The history: each generation solved one problem, and left behind (or created) a new one.

The paper era (before the 1980s)

Card files, Rolodexes, salespeople’s notebooks — and in a different industry altogether, the mailing lists in the hands of mail-order merchants (America’s Montgomery Ward mailed its first catalog in 1872; Sears followed in the late 1880s). What this solved was the most primitive problem of all — forgetting. At least the names got written down.

What’s worth noticing: these were already two unrelated worlds. The salesman’s card file managed “my customer relationships”; the merchant’s mailing list managed “people worth sending a catalog to.” Each world had its own pain — the former, “the relationships live in one person’s head and desk drawer, and when that person leaves, the customers leave with them”; the latter, “a paper list can’t be queried or segmented — there’s no way to answer ‘which customers haven’t bought in six months?’” Everything that follows is these two worlds chasing their own pains toward their own tools.

Database marketing (the 1980s)

Kestnbaum and others pioneered database marketing: for the first time, “customers” became data that could be queried and segmented. It solved paper’s inability to segment — the labor of recording and remembering was absorbed by machines. The decade left two legacies that still shape everything today.

One is RFM — for catalog mailers (the Sears and L.L.Beans of the era), the printing and postage on a single catalog were expensive, and “is this batch of people worth mailing?” was a life-or-death question. Out of that practice grew a method for picking people with three numbers — recency, frequency, and amount spent — later systematized into a textbook framework by Arthur Hughes and others. The RFM that every e-commerce marketing tool now ships built-in, and every e-commerce course teaches, was fundamentally invented forty years ago to save catalog postage (whether merchants actually use it is another matter — built into the tool doesn’t mean anyone presses the button). And the problem it solves — every touch has a cost, so spraying is burning money — is exactly the problem of LINE’s per-message pricing today.

The other is the loyalty program — in 1981, American Airlines launched the AAdvantage mileage program, proving that “using data to identify and keep high-value customers” is a business in itself. This line is the direct ancestor of today’s B2C marketing branch — and the problem it left behind stayed on its own line: segmentation could pick the people, but “action” meant mailing catalogs and direct mail in weekly batches, with a print shop and a post office standing between the data and the touch.

Meanwhile, in the other world of the same period — the sales frontline of phone calls, visits, and deal progress — there were no tools at all yet; the SFA line wouldn’t arrive for a few more years. In other words, the two worlds got their first computer tools one after the other, then kept walking their separate roads, until the 1990s, when analyst firms put the same “CRM” label on both and they started to look like one family. So strictly speaking, B2B and B2C CRM never split up later — they have been two different species since the paper era; they just ended up sharing a name.

Contact management / SFA (late 1980s–1990s)

Both lines kept moving through this era; they just moved differently. The B2C line advanced steadily along the track of the previous section: RFM was codified into a textbook framework by Arthur Hughes (1994), loyalty programs spread from airlines to retail and credit cards, and campaign management software (the Unica cohort, early 1990s) appeared, becoming the precursor of marketing automation. Gradual, quiet, no dramatic events.

The drama was on the B2B side: this line finally got its first batch of tools — ACT! (1987) and GoldMine reached the sales desktop, and in 1993 Tom Siebel founded Siebel Systems and made sales force automation an enterprise-grade product. The term “CRM” also settled in the mid-1990s — who coined it is undecided; what made the name win was analyst promotion. For the first time, frontline interactions had a system: contact records and deal progress went into a database, management could see a pipeline for the first time, and the labor of “tracking the sales process” was absorbed.

But the moment B2B got its tools, it also got two new problems. The surface one was the barrier: enterprise systems were expensive, required your own server room, took years to deploy — small businesses couldn’t afford them. The deeper one was misaligned incentives: salespeople were required to type what they already knew into a system, and got little in return — the beneficiaries were the boss reading the reports and the colleague who would inherit the account. Worse, “the customer relationships are in my head” had been the salesperson’s negotiating leverage; filling in the CRM meant surrendering it, and the activity log conveniently became surveillance material. The cost of input landed on the individual; the value of output landed on the organization. The vicious cycle kicked off: sales fill it in half-heartedly → the data rots → the reports can’t be trusted → the boss pushes harder → sales get even more half-hearted.

The evidence for this incentive structure only began to surface after the era ended — and it has run all the way to the present. For more than twenty years starting in 2001, the major studies using the definition “implementation missed its original goals” have measured the rate at around fifty percent (looser or stricter definitions land at seventy or thirty — sources at the end of this post). In other words: a company that pays to implement a CRM has roughly coin-flip odds of hitting its original goals. And as the tools moved from the server room to the cloud, the number barely moved — the problem was born in the 1990s, the technology has turned over several generations since, and the failure rate hasn’t changed. The disease is not in the technology; it’s in the incentive structure.

Only in recent years has a solution aimed at the actual disease appeared: the new generation of B2B CRMs (the Attio and Clarify cohort) plug into email and calendars and capture automatically, deleting the labor of “filling things in” entirely — with no input cost, there’s no half-hearted filling, and mechanically that strikes the source of the misaligned incentive. To be clear: these products are young, and whether they have actually improved implementation outcomes has no comparable research data yet — what can be said is that the direction matches the disease, not that the results are proven. And even if the direction is right, what automatic capture reclaims is records, not relationships — the dinners, the private LINE threads (in Asia, deals are often negotiated in a salesperson’s personal LINE), the tacit knowledge in a salesperson’s head, the customer’s trust in this particular person — systems still can’t capture any of it. “When they leave, the customers leave with them” has been weakened, not eliminated. Still, the direction itself confirms the through-line: the fix is not making the form easier to fill in — it’s absorbing that labor altogether.

The SaaS era (1999–)

Salesforce moved CRM onto the cloud under the banner of “No Software.” This generation doesn’t belong to either line — it was a revolution in delivery form, and it swept across both: B2B’s Salesforce and B2C’s Mailchimp (2001) are both products of this wave. It solved “can’t afford it, can’t install it” — open an account, pay monthly, and you’re running; the labor of “IT setup and maintenance” was absorbed, and for the first time both lines’ tools reached small and mid-sized businesses.

There’s a good piece of theater here, with a subtle symmetry: Tom Siebel had struck out on his own after his internal proposal at Oracle was passed over by Larry Ellison; six years later, another senior Oracle executive, Marc Benioff, also left to found a company — and marketed software itself as the enemy. The logo was the word “software” with a prohibition sign through it; in 2000 he went as far as hiring actors to picket Siebel’s user conference proclaiming “the end of software” (Siebel called the police, which mostly helped the stunt make the news). The reigning champion Siebel didn’t take the little web-page toy seriously — then in 2005 announced its acquisition by Oracle (closed in early 2006), the brand gradually disappeared, and Salesforce took over as the new champion. The first time in CRM history that an incumbent was overthrown by a lower-barrier form factor — and right now, the AI generation versus the SaaS generation may be replaying the same script.

The problems left behind split by line once again: the B2B line kept the data-entry resentment from the previous section (SaaS made the tools cheap; it didn’t make the forms any more worth filling); the B2C line’s was “the data is in the system, but every email and every campaign still has to be sent by hand” — impossible to keep up with once customer counts grow. That’s what the next section, marketing automation, set out to solve.

The marketing automation era (mid-2000s–2010s)

The B2B line did not stand still in this era — once SaaS CRM stabilized, it platformized: Salesforce opened AppExchange and grew from a single product into an ecosystem, HubSpot (2006) brought inbound marketing into B2B acquisition, and the mid-to-late 2010s sprouted sales engagement and call analytics as a generation of satellite tools. Busy — but in character it was stacking tools on top of an existing system: the shape of the record didn’t change, and the data-entry incentive problem carried over untouched.

What actually changed generational protagonists was B2C: marketing automation made “triggered sends” and “journeys” standard equipment — a customer abandons a cart and mail goes out automatically, a birthday triggers a coupon (the same category exists on the B2B side too; Marketo’s lead nurturing is exactly that — but what redrew the industry’s map was the e-commerce branch running from Mailchimp to Klaviyo). It solved the problem of manual execution never keeping up — the labor of “repetitive execution” was absorbed. The new problems it created came in two layers.

The first layer was a side effect: automation pushed the marginal cost of “sending one email” toward zero, so mass-blasting exploded — every brand on automatic bombardment, consumers’ inboxes turned into disaster zones — forcing a platform counterstrike. In 2013, Gmail shipped the Promotions tab and rerouted marketing mail out of the primary inbox wholesale. The marketing world wailed in the year of launch, but the later data showed open rates didn’t actually collapse (about half of users with tabs enabled still browse them) — the real loss wasn’t in that year’s numbers, but in “marketing messages systematically separated from personal mail” becoming the default reality from then on. When execution costs zero, attention becomes the new scarce resource — a lesson Taiwanese merchants are currently retaking on LINE, where per-message pricing makes sending expensive again: price forcing you to do the segmentation you could lazily skip in the email era.

The second layer is more fundamental: automation only faithfully executes the rules people set, and the judgments — “who to target, what to say, when to send” — still had to be thought up by a person. Judgment became the new scarce resource — large companies hire a CRM specialist to do exactly this full-time; SMBs don’t have that person. Hence a perverse situation: the stronger the tools became, the wider the gap grew between those who can use them and those who can’t.

Now (the AI/agent era)

The labor that every previous generation walked around — the one that never got absorbed — is judgment, and this generation is attempting to absorb it. “Attempting,” because unlike the five settled chapters above, this one is still in progress, and nobody knows how it ends. And if you look closely at what is actually happening, what’s being absorbed is the work of judgment, not the authority of judgment (whether to go ahead — a human still presses the button). The “work” is not a single flash of insight but a chain: one campaign, end to end, is reading the data, judging who it’s for, then judging what to say, and finally judging when to send — every step requires reading the situation before the next step is possible, and only when the whole chain has been walked does an answer take shape. This distinction is the key to understanding this era. That work is judgment labor — the labor of producing answers; the authority is the power to make the call. The chain has dozens of links; the call is a single press. They were never the same thing.

Today’s CRM splits into roughly three lines: B2B sales CRM (managing deals), B2C CRM / customer engagement (managing consumer relationships and retention marketing), and the service line (managing the problems customers bring to you — order lookups, returns and exchanges, complaints). The story above only covered the first two, but service has its own lineage: call centers from the 1960s, ticketing systems in the 1990s (the Remedy cohort), SaaS-ified in the 2000s by Zendesk (2007) and Intercom (2011) — it too grew up as its own tributary, never split off from anything.

That it gets counted as “CRM” is, once again, annexation by name: when analyst firms defined CRM in the mid-1990s, the scope already covered sales, marketing, and service — the three customer touchpoints. Siebel’s suite shipped a built-in service module, and Salesforce later launched Service Cloud (2009) — three species that grew up separately, packed into the same box by the same label. The only foundation the three genuinely share is “a customer file plus interaction records”; the difference is what each line does with those records. The story didn’t follow this line because what it manages is “solving the problems customers bring to you,” not “proactively running the relationship” — it intersects least with the retention-marketing main line.

But it deserves one note: in the AI generation, service is the line being absorbed fastest — because service judgments are single-case, most questions have standard answers, and the cost of getting one wrong is low, AI can step onto the floor and answer directly. Marketing judgment, by contrast, affects thousands of people at once — which is exactly why it needs “human approval” and service mostly doesn’t. Textbooks also carry an operational / analytical / collaborative academic taxonomy; nobody talks that way in practice, and it gets no further discussion here.

Closing: a ledger of labor

Forty years of CRM history is really a ledger of labor: recording and remembering, process tracking, setup and maintenance, repetitive execution — each generation of tools crossed one item off, and the opening question — “how do I remember my customers, and do the right thing for them at the right time?” — advanced one small step each time. The last item on the ledger, and the hardest, is judgment — this generation’s tools have started working on it, and the outcome is not yet written. Only one thing is certain: every time a kind of labor has been absorbed, the industry’s map has been redrawn. This time will be no exception.

References

CRM implementation failure rates

A note on definitions: “failure” usually means “the implementation missed its original goals,” not “the system doesn’t run”; studies vary in how strictly they define it. The original Gartner (2001, ~50%), Butler Group (2002, ~70%), and CSO Forum (2002, 69.3%) reports are not publicly available; the numbers are cited via industry roundups.

Dates and events