Why does HubSpot now put context first?
HubSpot's Fall 2026 release puts what it calls Growth Context at the centre: combining “business, team, and customer context in one place” to fuel its AI. Alongside it sits Context Home, which gives “a score that shows how complete your context foundation is and where the gaps are” (HubSpot).
HubSpot reports that businesses using AI with high-quality context create 3.6x more MQLs, win 3.2x more deals and close over 2x more tickets (same source). Those are vendor figures and show direction, not a forecast. The direction is still telling: the platform vendor is saying the quality of context matters as much as the agents.
Independent analysts add a caution. Info-Tech's SoftwareReviews advises buyers to “use Context Home as a starting point, but independently assess CRM accuracy, consent, permissions, ownership, and retention” (SoftwareReviews). A completeness score tells you what is there. It does not tell you whether it is right.
Where does context usually live today?
In most enterprises, context is scattered. Brand guidelines sit in a PDF from a past agency. Product data is in an ERP and three spreadsheets. Policies live in email threads. Approved answers live in the heads of the best service staff. Customer history is split across a CRM, a ticketing tool and messaging apps on personal phones.
For multi-market groups it is worse. Each market has its own price list, its own exceptions and its own language. An agent serving Kuala Lumpur, Singapore and Manila needs all of it, and needs to know which applies to whom.
How do you build context before agents go live?
1. Start from the use case
Do not try to document the whole business. Take the first use case, for example first-wave service questions on WhatsApp, and list exactly what an agent would need to answer them well. That list is your first context backlog.
2. Write it down where the agent can read it
Move approved answers into a knowledge base. Put product and pricing data in the CRM or connect it from the system of record. HubSpot supports knowledge base articles in multiple languages, organised into language groups, on Service Hub Professional or Enterprise (HubSpot Knowledge Base). One source, read by people and agents alike.
3. Unify the customer record
An agent cannot recognise a loyal customer whose history is split across systems. One record per customer, with conversations from every channel attached, is the part of context that takes longest and pays most. Our messaging-first service and legacy CRM migration pieces cover how.
4. Write the rules
List what the agent may and may not do, in plain language. Which topics hand over to a person. Which promises need approval. Which data it may read and write. Confirm how your AI vendors treat your data; SoftwareReviews also advises buyers to “confirm model-training, enrichment, connector, retention, and cross-customer data-use policies rather than relying on platform defaults” (SoftwareReviews).
5. Give every part an owner
Context decays. Prices change, policies change, products launch. Each part of the context needs a named owner and a review date. Without owners, the agent is accurate on launch day and wrong a quarter later.
How do you know the context is good enough?
Test before customers do. Run the agent against a set of real historical questions and have your best people grade the answers. Look at every wrong answer and ask why. Most trace back to missing, outdated or contradictory context, which is fixable. Re-test after each fix. Once live, review handoff reasons weekly; they are a map of where context is still thin.
Track a small set of measures: answer accuracy on the test set, handoffs caused by missing knowledge, and the age of the oldest unreviewed content. Context completeness scores help, but they do not replace grading real answers.
What does good context look like in practice?
A customer asks on WhatsApp whether a product is available in their city and how long delivery takes. A well-grounded agent knows the customer, sees their last order, answers from the current product and delivery data for that market, in the customer's language and the brand's voice, and offers a person if the answer is not clear. Every element of that answer is context. None of it comes from the model.
What should leadership do?
Treat context as a business asset, not an IT task. Fund it before agents. Assign owners at the level of the people who own the policies and prices, not only the CRM administrators. And ask, before any agent goes live: would our best employee give the same answer?
To find where your context is thin before the first agent, start with an AI readiness assessment.
