Key points
Choose a recurring task with a measurable problem.
A calculation, transfer, deadline or alert does not necessarily require AI.
Measure time saved after review and correction, together with quality.
Start with the task, not the tool
Meeting preparation, notes, contract clauses and complaints can all be candidates. Record their frequency, the work required and the consequences of an error. Our sales performance guide helps place each task in the wider process. Compare cases of similar complexity during the pilot.
Where can AI assist?
AI can help interpret varied text. Stable rules and structured data may be better served by conventional automation.
Activity | Assistance to test | Required check |
|---|---|---|
Meeting preparation | Summarise authorised records with source references | Facts, freshness and missing information |
Lead qualification | Extract the customer’s need | Apply agreed qualification criteria |
Proposal preparation | Draft from approved content | Verify scope, prices and commitments |
CRM updates | Suggest notes and field values | Have material changes approved by the salesperson |
Complaint handling | Classify the issue and retrieve relevant documents | Validate routing and the customer response |
An overdue invoice alert or a calculation from a structured price list does not need to become an AI project. For incoming enquiries, establish qualification criteria first.
Check data and responsibilities
Information must be available, current and accessible to authorised users. Specify which document versions are valid and what to do when information is missing. The assistant must not invent it.
Name who uses the output, who checks it, who authorises customer commitments and who takes over when the system fails. The NIST AI Risk Management Framework supports this contextual approach to testing and responsibilities; it does not guarantee a sales result. AI agents that take several actions require additional controls.
Organisation and assessments
Make organisational decisions with a clear assessment
Our assessments identify practical changes to help your teams and processes work together.
When should the process be simplified first?
Clarify ambiguous approval rules and missing responsibilities before embedding them in a tool. An assistant might prepare a quote faster while sending it remains blocked because nobody knows who can approve a discount exception.
Identify what limits completion of the whole case. A local time saving has little value if another step absorbs it. Check features already available in the CRM or invoicing system before adding another assistant.
An example: from customer request to proposal review
This is an illustrative example, not a claimed engagement result. A company receives requests containing emails, attachments and references to earlier offers. The salesperson has to assemble the information, find approved material and prepare a proposal.
First, define the information required and the procedure for gaps. AI can then suggest a summary and draft proposal, identifying its sources. Pricing calculations and discount rules stay in the pricing system. Sales or presales approves the scope and commitments before sending.
A pilot compares similar cases with and without assistance. It measures total preparation time, text requiring revision, omissions and errors caught before sending. The company may keep document summaries but drop automatic drafting if it requires too much correction.
After acceptance, billing deadlines can be automated through rules linked to contractual events. This does not make them an AI application.
Decide whether to extend, adapt or stop
Before testing, define the baseline, scope, quality criteria and review date. Track:
Total time per case, including review, correction and exceptions.
The share of eligible cases actually handled with the tool.
The frequency and severity of errors, especially customer commitments.
Running costs and internal support workload.
Compare cases of similar complexity and record volume changes. A result may also reflect clearer rules or a more experienced team. A serious error can justify suspension even if average handling time improves. Low adoption may call for better integration into daily work before abandoning the use case.
Our lead-to-cash action plan example illustrates a sequence that improves data and working rules before automating tasks. It combines experience from several engagements; the gains shown are estimates.
Frequently asked questions
Does using AI require a new CRM?
Does using AI require a new CRM?
Not necessarily. Examine the functions, data and integrations already available. A change of CRM should address a demonstrated need, rather than serve as a prerequisite for every AI use case.
What is a useful first pilot?
What is a useful first pilot?
Choose a bounded task with observable volume, time and quality, and people available to review the output. Define a baseline and the decision the pilot must inform.
Can AI send a proposal directly to a customer?
Can AI send a proposal directly to a customer?
For proposal drafting, retain human approval of references, prices, terms and commitments. Any move towards autonomous sending requires its own assessment and controls.
How should time saved be measured?
How should time saved be measured?
Include review, correction and any work transferred to other teams. Compare similar cases. Report usable time released separately from actual spending reductions or additional margin.
Sources
NIST, AI Risk Management Framework 1.0 — Core, 2023.
The use cases, checks and illustrative example are Spentia methodological proposals, not claimed client outcomes.
