Gartner predicts AI agents will outnumber human sellers 10 to 1 by 2028. It also expects fewer than 40% of sellers to say those agents improved productivity. That is not an adoption problem. It is a systems problem.
Sales teams already have too many tabs. Adding an agent for prospecting, another for call notes, and a third for CRM updates can create more digital activity without giving a salesperson one extra hour with a qualified buyer.
We disagree with the idea that the first goal should be an autonomous AI salesperson. For most businesses, the better first project is much less dramatic: remove the administrative work surrounding a real sales conversation.
What an AI sales agent actually does
An AI sales agent is software that can complete a sales task using company data and connected tools. Unlike a basic chatbot, it can take an action such as researching an account, preparing a meeting brief, drafting a follow-up, updating the CRM, or routing a lead.
The word agent does not mean the software should operate without supervision. Autonomy is a design choice. An agent can prepare a CRM update and wait for approval, or it can write directly to the record. Those two workflows carry very different risks.
The useful distinction is between internal work and customer-facing work.
Internal agents prepare information for the sales team. They summarize account history, identify missing CRM fields, or turn a call transcript into proposed next steps. Customer-facing agents send messages, answer questions, or contact leads.
We would start internally. A bad meeting summary can be corrected before it leaves the business. A bad email sent to 5,000 prospects becomes a reputation problem.
Tool sprawl is already eating the productivity gain
Gartner's July 28 warning is blunt. More agents will not automatically create more productivity. The firm says sales leaders risk agent sprawl unless they redesign their data, automation, and seller experience.
Agent sprawl happens when every sales tool adds its own assistant, memory, and workflow. One agent reads the CRM. Another sees email. A third lives inside the call-recording platform. None has the full context, so salespeople spend time correcting contradictions and copying information between systems.
The failure usually looks ordinary:
-
A prospecting agent contacts an account that already has an open opportunity
-
A follow-up draft ignores the buyer's objection from yesterday's call
-
A CRM agent overwrites a field that the account executive changed manually
-
Two agents create duplicate tasks for the same person
These are not model-intelligence problems. They are ownership and integration problems.
Before adding another tool, decide which system owns each piece of information. The CRM should usually remain the source of truth for accounts, contacts, opportunities, and activity. Agents can read from it and propose updates, but they need explicit rules for conflicts and missing data.
Start with the work after a sales call
Many teams start with automated outbound because the demo is easy to understand. Upload a lead list, generate messages, send at scale.
We think that is the wrong starting point for most companies.
Generic outbound is already crowded. Scaling weak personalization can damage deliverability and make a good market look unresponsive. It also puts the agent directly in front of potential customers before the team knows whether its data and controls work.
Post-call administration is a better first workflow. The agent can:
-
Match the call to the correct account and opportunity
-
Draft a concise summary with evidence from the transcript
-
Extract objections, commitments, and the next agreed action
-
Propose CRM field updates
-
Prepare a follow-up email for the salesperson to approve
This work is repetitive, but it depends on context. It is also easy to audit because the recording and transcript provide evidence.
The salesperson stays responsible for the relationship. The agent removes the typing.
The ROI needs a baseline, not a vendor promise
Salesforce's 2026 State of Sales research found that the average seller spends only 40% of working time selling. Sellers expect agents to reduce prospect-research time by 34% and email-drafting time by 36%.
Those figures describe expectations, not guaranteed savings. The business case should begin with the team's actual week.
Suppose 10 salespeople each spend 12 hours a week on research, notes, CRM updates, and follow-ups. At a fully loaded employment cost of ₹1,000 per hour, that administrative work represents ₹6.24 million a year.
If an agent removes 25% of it, the team gains 1,560 hours of annual capacity, worth ₹1.56 million at the same cost. That does not mean ₹1.56 million appears in the bank. The value arrives only if the time becomes more customer conversations, faster responses, better pipeline coverage, or lower hiring needs.
A ₹2 million first-year project would fail that narrow calculation. A ₹600,000 implementation with a reliable 25% reduction could justify a controlled rollout.
This is why time saved is not enough. Measure what the team does with the time.
Measure capacity and commercial outcomes together
An agent can make activity numbers look impressive. More emails. More tasks. More CRM updates. None of those guarantees more revenue.
We would track a small set of connected measures:
-
Administrative hours per seller
-
Time from customer meeting to approved follow-up
-
Percentage of CRM records that need manual correction
-
Qualified opportunities progressed per seller
-
Revenue or gross margin per seller
The first three show whether the workflow functions. The last two show whether the business benefits.
Keep a comparison group during the pilot. If five sellers use the agent and five continue with the current process, the team can separate the agent's effect from seasonality, a new pricing offer, or one unusually large deal.
Gartner predicts that sales leaders who overhaul data, automation, and user experience will be five times more likely to gain AI ROI than those choosing quick fixes. That does not mean every company needs a huge transformation program. It means the agent must fit one coherent sales system.
Clean data before increasing autonomy
Salesforce reports that 51% of sales leaders using AI say disconnected systems are slowing their initiatives. It also found that 74% of sales professionals are focusing on data cleansing.
That unglamorous work matters.
If the CRM contains duplicate contacts, stale opportunity stages, and incomplete ownership fields, an agent will act on those errors at higher speed. It may produce a polished meeting brief for the wrong subsidiary or send a follow-up from the wrong account owner.
Start with read access and proposed changes. Log the source used for every important fact. Require approval before the agent sends customer communication or changes an opportunity stage.
Autonomy can increase after the team measures error rates. Low-risk actions such as adding a call note may become automatic. Pricing commitments, contract language, and account reassignment should remain controlled.
Buy the feature, build the workflow
If a company uses a mainstream CRM and only needs call summaries, the built-in assistant may be enough. Building custom software for a standard transcription problem is usually wasteful.
Custom development becomes useful when the workflow crosses several systems or reflects how the business actually sells. A distributor may need inventory and regional pricing in every quote. A services company may need project capacity before promising a start date. A marketplace may need to coordinate buyers, sellers, and payment status.
In those cases, the valuable work is not inventing another general-purpose model. It is building a context layer that connects the CRM with the relevant business systems, then defining exactly what the agent may read, propose, and change.
The first release should solve one expensive handoff. If it does not reduce work or improve a commercial measure within a fair pilot, stop. Deploying more agents will not rescue a workflow that never had a clear outcome.
Axentia builds AI agents and full-stack systems around the real way a sales team operates, including integrations, approval controls, and ROI measurement. If your team is losing selling time to fragmented tools and manual updates, book a call with us and we can map the first workflow worth automating.