Notes for owners · CRM, automation and RevOps
Practical AI applications for sales and marketing teams
The useful AI applications for a sales team are unglamorous: call summaries, follow-up drafts, enquiry sorting, data cleaning and prospect research.
The GullySales team · Updated 15 Sept 2026 · 8 min read
On this page
The applications that pay for themselves in a small sales and marketing team are unglamorous. Summarising calls and writing them into the CRM. Drafting follow-up messages and quotation covering notes. Sorting incoming enquiries by what they say. Cleaning and deduplicating customer data. Researching a prospect before a call. Translating between English and the languages your customers actually use. In every one of those, a person still approves whatever the customer sees.
Where it genuinely helps, and what a person must still do
| Task | What the tool does well | What a person must still do |
|---|---|---|
| Call notes | Transcribe, summarise, extract the next action and the objection | Correct the facts, decide what happens next |
| Follow-up messages | Produce a usable first draft in seconds, in the right language | Add what was actually discussed, approve before sending |
| Enquiry sorting | Read the enquiry text and route it by product, city and urgency | Judge whether the enquiry is a buyer |
| Lead scoring | Rank enquiries on patterns from your own won and lost history | Set the rules, and check them every quarter |
| Data cleaning | Find duplicates, fix formats, standardise company names | Decide which record survives |
| Prospect research | Summarise a company's website, products and recent news | Verify anything you will say on the call |
| Content drafting | Outlines, product descriptions, variations of an advertisement | Facts, claims, prices, and the voice |
| Report summaries | Turn a month of CRM data into a readable page | Explain why, and decide what to do |
The pattern is consistent. It is good at the writing and the sorting. It is unreliable on facts about your business and useless at judgement.
Call summaries are the highest-return starting point
If your team makes thirty or more calls a day, this is where to begin. A recorded call becomes a transcript, and the transcript becomes a five-line summary: the customer's requirement, the objection raised and the agreed next step. That summary is written into the CRM against the right record.
Two things change immediately. The CRM stops being empty, because nobody has to type notes after every call. And a sales manager can read what customers actually said instead of asking the team.
For example, an inside sales team of four making 120 calls a day between them might spend an illustrative two minutes per call writing notes. That is four hours of the team's day. Most of that comes back. It also works across languages. A call conducted in Kannada or Hindi can be summarised in English for the manager without anybody translating by hand, which matters in any Indian business selling across states.
Sorting and scoring enquiries, carefully
An enquiry that arrives as free text, on a form or on WhatsApp, contains more than most CRMs capture. A tool can read it and set the product, the city, the quantity and the urgency, then route it to the right person. That alone removes the delay where an enquiry sits unread because nobody knew whose it was.
Scoring is more delicate. A score built from your own won and lost history is useful. A score bought as a black box is not, because your team will not trust a number they cannot question, and a distrusted score is worse than none. Keep the rules visible, review them each quarter against what actually converted, and never let a low score stop a human from calling.
Cleaning the data nobody wants to clean
Most small business CRMs and customer lists contain the same company written four ways, mobile numbers with and without country codes, and three records for one buyer. This is genuinely tedious work and genuinely well suited to a tool.
Run it once as a project: deduplicate on phone number, standardise company names and cities, fill missing designations from public sources, and flag records with no contact for two years. Then put a rule in place at the point of entry so the mess does not rebuild.
Drafting, with the facts supplied
Use it for the shape, not the substance. A follow-up sequence, a quotation covering note, forty product descriptions for a catalogue. Or six versions of an advertisement to test, and a first draft of a case study from your own notes.
Two rules keep this safe. Give it your own material as the source, the price list, the specification sheet, the warranty terms, rather than asking it what your product does. And have a person read anything that leaves the building. The failure mode is not bad English. It is a confident sentence about a tolerance or a delivery time that nobody checked.
What not to do
Do not let a tool hold an unsupervised conversation with a customer about price, delivery or clinical matters. Do not paste patient records, financial details or anything under a customer confidentiality clause into a free consumer tool. Do not replace a qualification call with a score. Do not publish content nobody read. And do not buy a platform before you have run one of these tasks by hand for a month, because you will not know what to ask for.
How to start, and what to do next
Pick one task that is costing your team hours: notes, follow-up drafts or list research. Run it for four weeks on a small tool and measure one thing, hours saved or reply time. Write down what may and may not be given to the tool, and who signs off anything a customer sees. Only then connect it to your CRM.
Ask your sales team where their day actually goes, in half-hour blocks, for one week. The answer is usually notes, chasing information and typing the same message again. Those three are where to apply this first.
If you want a view on which of your processes is ready for this and which needs fixing by hand first, tell us what you need.
