CRM and AI: what Dreamforce and HubSpot Unbound really change for an SME

Within two days, the two CRMs most widely used by SMEs announced that they were changing nature. Salesforce presented AIforce at Dreamforce on 15 September 2026; HubSpot unveiled a CRM rebuilt around what it calls Growth Context at its Unbound conference on 16 September.

Behind the slogans we see one idea: the CRM stops being a screen you fill in and becomes a layer of context that agents act on. Here is what that changes, and does not change, for an SME in French-speaking Switzerland and beyond.

What was announced, without the marketing gloss

At Salesforce, AIforce is an interface layer that carries the platform's business logic, semantics, security and governance to wherever people use an AI agent. It has three building blocks: Claudeforce, born of the Anthropic partnership, ships 37 prebuilt sales skills, from prospecting to pipeline hygiene;

Slackforce brings CRM data into Slack, where you can open an account or log a call without leaving the chat; Agentforce Coworker puts an AI "teammate" inside the Salesforce interface that calls the agents the company has already deployed, within its rules and permissions.

Marc Benioff summed up the ambition as an "interface revolution", according to SiliconANGLE's report of 15 September 2026.

At HubSpot the shift is even more radical for an SME: the Smart CRM now updates itself from calls, emails and meetings; a "Context Home" scores data completeness and flags the gaps that limit AI performance; the Breeze assistant takes an objective and hands the work to specialised agents (proposal, report, campaign plan).

HubSpot quotes flattering numbers, 2.2 times more leads for Breeze users and 81% more campaigns for early Marketing Studio users, but, as SiliconANGLE noted on 16 September, the company did not explain how those comparisons were calculated. We read them as orders of magnitude, not as promises.

The macro picture pushes the same way. In a press release dated 16 September 2026, Gartner forecast a 49.5% increase in worldwide AI spending this year. The same week, half a dozen security vendors launched "kill switches" and rollback features for AI agents in production.

The market is accelerating and growing wary at the same time; that is exactly the posture we recommend.

The real change: the CRM becomes a context layer

For twenty years the value of a CRM depended on how disciplined salespeople were about filling it in. Every SME owner knows the result: half-empty records, opportunities never closed, a pipeline that reflects only what someone had time to type.

What Salesforce and HubSpot are announcing, each in its own way, is a reversal of the burden: the system captures, structures and links, and the human validates.

That reversal has a consequence few press releases highlight. If the CRM becomes the context that agents read, then the quality of that context becomes the limiting factor for all commercial AI. HubSpot understood this well enough to turn it into a feature: a data-completeness score on the home page.

We already made the point in our article on AI-ready data: leaders' confidence in their data far exceeds its actual state.

A prospecting agent that monitors "more than 40 buying signals", as HubSpot promises, is worthless if accounts have no industry filled in, if contacts are duplicated, or if pipeline stages mean different things from one team to the next.

The second change concerns the interface itself. When your salespeople update an opportunity from Slack or from a voice assistant in the car, the question of who is allowed to do what leaves the perimeter of the CRM screens.

Salesforce stresses a "governed" architecture with zero data retention; HubSpot promises the assistant stays within permissions. These are good intentions from vendors. They do not replace a rights model you have defined yourself.

An example: a 45-person industrial SME in the canton of Vaud

Take a manufacturer of laboratory equipment based in the canton of Vaud: 45 employees, six salespeople, around 1,200 active accounts in HubSpot and a sales cycle of four to nine months. The owner asked us the question many are asking this week: should we switch the agents on as soon as the autumn release lands?

We started by measuring rather than activating. The findings: 38% of accounts with no industry, 22% of contacts with no decision-making role, three different definitions of the "proposal sent" stage depending on the salesperson, and a share of customer exchanges happening on WhatsApp, hence invisible to the CRM.

On that basis the prospecting agent would have produced elegant, wrong recommendations, and the sales team would have stopped trusting it within three weeks.

The plan we settled on takes six weeks. Two weeks to clean and normalise: deduplication, mandatory industry field, a stage dictionary signed off by all six salespeople.

Two weeks to connect automatic capture on a single channel, the business email system, and to review every evening what the Smart CRM created or changed. Two weeks to activate a single agent, meeting preparation, with a human reading every brief before the visit.

At the end of the period, preparing a meeting takes under 15 minutes instead of around 40, and above all the team has a precise list of what it lets the AI do alone and what it keeps. Only then do the next agents make sense.

Three questions to ask your vendor, and yourself

Dreamforce and Unbound will trigger a wave of sales calls. Before signing an amendment or activating an option, we suggest three questions.

• Where does my data go and who reads it? Ask in writing whether your customers' prompts and documents are used to train models, where they are hosted and how long they are kept. Salesforce highlights zero retention; demand the same level of precision from every supplier, especially if you process data of European customers under the GDPR or Swiss data under the revised FADP.

• What happens when the agent gets it wrong? An agent that sends a wrong quote or misqualifies a prospect costs more than an empty form. Check that there is an action log, a rollback mechanism and a named human for each type of action. Back in 2025 Gartner estimated that more than 40% of agentic AI projects would be cancelled by the end of 2027 for lack of clear value or risk controls; this week's announcements change nothing about that.

• Which business metric will move in 90 days? Not "adopt AI", but a response time to inbound requests, a stage-to-stage conversion rate, a preparation time. If nobody can name that metric, activation can wait.

To these three questions add the regulatory dimension. An agent that scores or classifies prospects and customers falls under the transparency obligations of the AI Act, applicable since 2 August 2026, including for a Swiss SME selling into the Union. We covered this in our practical guide to the AI Act.

What we take away

The Salesforce and HubSpot announcements are serious and convergent: the CRM becomes a context layer that agents act on, and the interface dissolves into everyday tools. For an SME the good news is real: manual data entry, the first cause of failed CRM projects, is going to recede.

The flip side is just as real: data quality and the permissions model become the assets that decide whether these agents produce value or noise.

Our recommendation is simple. Do not chase the autumn release. Measure the state of your CRM, clean what needs cleaning, activate one agent at a time with a business metric and an accountable human, then widen. Vendors are selling an interface revolution; what you are buying is data discipline.

If you want to locate your SME before deciding, the 39 Advisory AI Barometer gives you a ten-minute reading of your maturity on data, processes and governance. And if you would rather talk it through, get in touch.