Industries / Technology / Software and SaaS / ERP and CRM software
Sales
Sales software keeps the record of a company's customers, their contacts and every deal in progress, so that sellers and their managers can see who to call next and what is likely to close.
5
Milestones achieved, from 1987 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1987
Contacts on a desk computer
Act! dates its product to 1987, when Pat Sullivan and Mike Muhney created Activity Control Technology, a program that kept one salesperson's contacts, calendar and history of calls on a personal computer. Act!'s own timeline calls it the first CRM, short for customer relationship management. Symantec, which bought its maker, Contact Software International, on 2 June 1993, described ACT! in 1995 as "an easy-to-use contact database with a graphical activity schedule, a full featured word processor and a report generator".
Credit
Pat Sullivan and Mike Muhney
1995
One sales system per company
Siebel Systems, which began operating in July 1993, shipped version 1.0 of Siebel Sales Enterprise in April 1995. It held a whole company's customer, product and competitor information in one shared database for field sellers, telesales and marketing staff, in many languages and currencies. Its 1996 stock-market prospectus said earlier office automation had aimed to cut the cost of finance and manufacturing, while sales systems "focus primarily on increasing revenues". By 31 May 1996 only 21 organizations had licensed it, all on a trial or limited basis; by 2001 Siebel's revenue was $2.08 billion.
Credit
Thomas M. Siebel and Siebel Systems
2002
License sales drop 34%
Siebel's revenue from software licenses fell 34% in 2002, to $700.3 million, and its staff fell from 7,403 to 5,909. Its annual report cited industry figures in which corporate IT spending fell about 3% in dollars that year. In September 2005 Oracle, the database and business-software company, agreed to buy Siebel for about $5.85 billion.
LessonSales of large up-front licenses fell more than ten times as fast as the IT budgets they were paid from.
Credit
a reversal for Siebel Systems; the buyer was Oracle
2016
Predictions inside the record
Salesforce announced Einstein in September 2016, putting machine-learning predictions into its CRM. Predictive Lead Scoring ranks new leads by how likely they are to buy, Opportunity Insights warns a seller when a deal is trending up or down, and Automated Activity Capture files a seller's emails and meetings against the right customer record without any typing. The Irish trade site TechCentral.ie reported that the lead scores learn from signals such as lead source, industry, job title, web clicks and emails.
Credit
Salesforce
2025
Phone calls empty the CRM
Google Threat Intelligence Group reported in June 2025 that a group it tracks as UNC6040 telephoned employees while posing as their IT support staff. The callers talked them into approving a modified copy of Salesforce's Data Loader, a tool for bulk export, as a "connected app" on their company's Salesforce account. The app then copied out customer data, which a related group later used to demand payment under the ShinyHunters name. Google said no flaw in Salesforce was used, and that one of its own Salesforce instances, holding contact details of small and medium businesses, was hit the same way that June.
LessonThe customer list was taken through one staff member's approval of an app, not through a software flaw.
Credit
reported by Google Threat Intelligence Group
Watch for
An agent for every lead
Salesforce says that in July 2025 it began using its own AI agent to follow up every sales lead, where its roughly 400 sales development reps, the staff who make first contact, had followed up about one lead in four. By 1 October 2025 it counted 68,000 leads re-engaged, 156,000 emails sent, 2,400 replies, 800 meetings booked by the agent and "dozens of deals" closed.
Achieved whenSalesforce or an outside party publishes the revenue won from agent-handled leads against a comparable group of leads handled by people, with unsubscribe and complaint rates.
Claimed by
Salesforce (a company blog post by Lisa Lee), October 2025
A qualifying agent beats ChatGPT
Microsoft says its Sales Qualification Agent in Dynamics 365 Sales, which researches each lead, writes to it and asks qualifying questions, beat ChatGPT on Microsoft's own Sales Qualification Bench: by 6% on company research, 20% on personalized outreach and 16% on replies to leads. The test used more than 300 synthetic leads, scored from 0 to 10 by language models acting as judges, with people checking a sample; Microsoft says it plans to publish the test so customers can repeat it.
Achieved whensomeone outside Microsoft runs the published test on this agent and on rival agents and finds similar margins.
Claimed by
Microsoft, December 2025
Open
01
Call notes against a key
A public test of the AI note-takers that listen to sales calls and fill in CRM fields. Record 100 scripted discovery calls, the first real conversation between a seller and a possible buyer, each with a written answer key for eight fields: budget, decision-maker, need, timeline, competitor named, next step, its date and deal size. Play every call through each note-taker and score each field it writes as right, wrong, missing or invented.
The briefCredit · unclaimed02
Sales agents, tested from outside
A head-to-head test of the AI sales agents that CRM vendors now ship, run inside their own products on identical leads. Make 200 fictional inbound leads, each from a made-up company with web pages the tester hosts and an inbox the tester controls, so that no real person is contacted. Load the same leads into each vendor's trial account, let each agent research, write and reply, and have sellers who do not know which agent wrote what grade every message, then check every factual claim against the hosted pages.
The briefCredit · unclaimed
Open 01 · the brief
Call notes against a key
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of the AI note-takers that listen to sales calls and fill in CRM fields. Record 100 scripted discovery calls, the first real conversation between a seller and a possible buyer, each with a written answer key for eight fields: budget, decision-maker, need, timeline, competitor named, next step, its date and deal size. Play every call through each note-taker and score each field it writes as right, wrong, missing or invented.
Why it is still open
CRMArena-Pro, Salesforce AI Research's benchmark, tests agents on written call transcripts, not recordings. The largest open synthetic set, DeepMost's SaaS sales conversations, is text with conversion outcomes and has no field answer key. Microsoft's Sales Qualification Bench is Microsoft's test of its own agent. A search found no published field-by-field accuracy for any note-taker.
Built with what exists
two volunteers reading scripts, or synthetic voices; trial or free accounts of CRMs that write call notes into records; an open pipeline for comparison, made of a speech-to-text model such as Whisper or ElevenLabs Scribe (see the Note "Fewer word errors") and a language model; a spreadsheet for the key and scores.
First step
Write 20 call scripts and their answer keys from scenarios in the DeepMost set, record them with two volunteers, and run one note-taker on all 20.
Done looks like
100 calls, 8 fields and at least 4 tools, 3,200 scored fields in all, published with the recordings, scripts and keys under an open license. A second person re-scores 10% of fields and the agreement rate is published.
Open 02 · the brief
Sales agents, tested from outside
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A head-to-head test of the AI sales agents that CRM vendors now ship, run inside their own products on identical leads. Make 200 fictional inbound leads, each from a made-up company with web pages the tester hosts and an inbox the tester controls, so that no real person is contacted. Load the same leads into each vendor's trial account, let each agent research, write and reply, and have sellers who do not know which agent wrote what grade every message, then check every factual claim against the hosted pages.
Why it is still open
Microsoft compared its agent only with ChatGPT, scored by language-model judges, and has not yet published its test. Salesforce's figures come from its own use. Open sales benchmarks, such as SalesBench-100 and the Sales Agent Benchmark on GitHub, test agents through simulated tools, not the vendors' shipped agents.
Built with what exists
trial or developer accounts for Salesforce with Agentforce, Microsoft Dynamics 365 Sales with the Sales Qualification Agent, and HubSpot with its Prospecting Agent; a static web host for the fictional companies; a mail domain the tester owns; the scoring dimensions Microsoft published, so its results can be compared.
First step
Build 20 fictional company pages and leads, open two vendor trials, and read each vendor's terms on publishing test results.
Done looks like
200 leads through at least 3 agents, with a published score per agent on research, outreach and replies, the count of false statements per 100 emails, and the share of leads correctly handed to a person. A second grader repeats 20% and the agreement is published, and each vendor sees its results 14 days before publication.