Customer Relationship Management Software

Kate Leggett has covered CRM at Forrester for more than a decade. Two of her post titles, read together, describe the category better than any evaluation summary.

The first: Forrester data shows high CRM adoption but low satisfaction.

The second: CRM software has become overengineered, and complexity is killing its value.

That is the analyst who runs the coverage saying the category's core problem is its own product. Not gaps in capability, not immature technology, but too much of everything, badly enough integrated that the value gets lost on the way to the user.

Anyone who has watched a sales team quietly maintain a parallel spreadsheet will recognise the diagnosis.

How CRM got overengineered

The mechanism is worth laying out because it explains why this is structural rather than a vendor failing.

CRM started as a contact database with a pipeline attached. Then it absorbed marketing automation, service case management, field service, partner management, quoting, forecasting, territory planning, commission calculation, and a platform layer for building whatever else you needed.

Each addition was individually justified. A company already holding the customer record is the obvious place to also hold the service history, and the vendor selling the record has every incentive to sell the rest.

The result is a system that can do almost anything and requires configuration to do the specific thing you need. That configuration accumulates: custom objects, validation rules, required fields, approval processes, and integrations built by three generations of administrators who no longer work there.

The cost lands on the person entering the data. A seller who must complete eleven fields to log a call will complete them badly or not at all, and every downstream capability, the forecast, the analytics, the AI, rests on what that seller typed.

Which produces the pattern Forrester's data captures. Adoption is nearly universal because no large organisation can operate without a system of record. Satisfaction is low because the system of record extracts more effort from its users than it returns.

Inside The Forrester Wave: Customer Relationship Management Software, Q1 2025

Published on 24 March 2025 and authored by Kate Leggett with two contributors, the evaluation covers twelve providers.

Microsoft placed as a Leader, positioning around what it describes as an integrated and autonomous CRM.

Pegasystems placed as a Leader, with Forrester assessing it as uniquely suited to actualise its vision, across a portfolio spanning customer service, sales, automation, and its decisioning engine.

Salesforce and Oracle are reported among the Leaders alongside them, though I could confirm placement directly only for the first two.

Forrester separately evaluates CRM for financial services, covering seven providers against thirty eight criteria in a Q1 2025 edition. That distinction matters more than it appears: industry-specific CRM has different data models, different regulatory requirements, and a partly different vendor set, and the general evaluation does not substitute for it.

Leggett's framing across the launch commentary is consistent: a market on the cusp of change, where some vendors are better positioned than others to incorporate AI in a way that transforms it.

Can AI simplify what AI-era vendors built?

The interesting claim in Forrester's commentary is not that AI will make CRM more powerful. It is that AI has the potential to simplify it.

That is a different and more useful proposition. The problem is not that CRM lacks capability, it is that accessing the capability requires knowing where it lives and how it was configured.

Three specific simplifications are plausible and worth testing against any vendor's roadmap.

Removing the interface. If a seller can ask for what they need and act through conversation rather than navigation, the accumulated complexity of screens and objects stops being their problem. The complexity still exists, but it moves behind a layer that absorbs it.

Removing the data entry. This is the significant one. If the system captures the call, the email, and the meeting, extracts the outcome, and updates the record without the seller typing, the fundamental bargain of CRM changes. The seller stops paying a tax to feed a system that serves management.

Removing the configuration. If a system can adapt to how a team works rather than requiring an administrator to encode it, the implementation burden that produces much of the complexity falls away.

The sceptical view is that AI becomes another layer on top of the existing complexity rather than a replacement for it, which is what happened with every previous simplification wave in this category. Mobile CRM was going to fix adoption. So was social. So was the platform approach.

The test that distinguishes them is whether anything gets removed. A vendor demonstrating an AI assistant alongside the same object model, the same required fields, and the same administration burden has added rather than simplified.

The data quality problem, from both directions

Everything in CRM depends on the record being accurate, and the record has always been the weak point.

Sellers enter what advances their interests: opportunities that justify their pipeline, close dates that survive the current quarter, activity that demonstrates effort. Managers know this and apply discounts. Forecasting becomes a negotiation about someone's optimism rather than a calculation.

Automated capture genuinely improves this. A meeting that happened, an email that was sent, a conversation whose content was analysed are facts rather than assertions, and systems that populate the record from observed activity remove a category of distortion that has existed since CRM began.

But it introduces a different problem that revenue operations functions are already encountering. When activity is captured automatically and enriched by AI, the distinction between what happened and what was inferred blurs. An opportunity stage advanced because a model read a transcript and concluded the buyer was interested is a judgement recorded as a fact, and it will feed forecasts, scoring models, and eventually agent decisions.

The practical requirement is provenance: knowing for any field whether a human entered it, a system captured it, or a model inferred it. Most CRM implementations cannot answer that today, and the volume of inferred data is rising fast.

Autonomous CRM, and who is accountable

Microsoft's positioning around an autonomous CRM describes where every vendor in this evaluation is heading: agents that qualify leads, draft follow-ups, update records, and progress opportunities without a person initiating each action.

The productivity case is straightforward. The accountability case is not.

A CRM has always been a record of what people did. An autonomous CRM is a record of what people and software did, and the two need to remain distinguishable for reasons that go beyond tidiness.

Compensation depends on attribution. If an agent sourced and qualified an opportunity that a seller closed, the commission plan needs a position on that, and most commission plans do not have one.

Customer relationships depend on knowing who said what. A prospect who received four agent-generated messages and then speaks to a human expects that human to know what was said, and the human expects the record to be accurate about what was promised.

And accountability for errors requires knowing what acted. An agent that emails the wrong contact, promises unavailable functionality, or advances an opportunity that was never real produces consequences someone has to own.

None of these are reasons not to deploy. They are decisions to make before deploying, because retrofitting attribution across a CRM that has been running agents for a year is considerably harder than establishing it at the start.

The front office war

Leggett has separately written about ServiceNow and Salesforce crossing battle lines in an escalating CRM conflict, which points at a structural shift worth understanding.

CRM's boundaries were historically defined by function: marketing, sales, service. The systems that competed with it were other CRMs.

The current competition comes from adjacent categories that own workflow. Service management platforms argue that customer service is a workflow problem and they already run the enterprise's workflows. ERP vendors argue the customer record belongs with the transaction record. Contact centre platforms argue they own the interactions. Data platforms argue they own the customer profile.

Each is claiming a piece of what CRM assembled, and each has a credible argument for the piece it wants.

For a buyer this makes the category boundary less useful than it was. The relevant question is not which CRM is best but which system will be the authoritative record for which entity, and how the others stay consistent with it. That is an architecture decision that a Wave evaluation is not designed to answer.

What this means operationally

For anyone running revenue operations, the current state produces a specific and awkward position.

The category's own analyst says complexity is killing value, and the remedy on offer is more technology from the vendors that built the complexity. That is not a reason to disbelieve the remedy, but it is a reason to measure it.

The measurable question is whether user effort falls. Not whether capability increases, which it always does, but whether the time a seller spends feeding the system goes down and whether the resulting data is better. Those two moving together is the only evidence that simplification is real.

The second question is what gets retired. Every CRM implementation carries configuration that made sense once and now exists because nobody is sure what depends on it. An AI layer over that inheritance is a faster interface to a system nobody understands. Genuine simplification involves deletion, and deletion requires someone willing to break something to find out what it was for.

High adoption and low satisfaction has been the stable state of this category for a decade. Whatever resolves it will be visible in satisfaction rather than in capability, and it will be measurable inside your own organisation long before an analyst reports it.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.