CRM with artificial intelligence: clean data, reliable forecasts

Team commerciale in riunione di revisione pipeline: SugarAI, il CRM con intelligenza artificiale

Monday meeting, four sales reps and a sales director. On the table is a forecast built the night before, with three deals that make the quarter. The director asks how long it has been since anyone spoke to the first customer: no precise answer comes, because those emails sit in one rep’s mailbox. On the second deal the last note dates back to March. The third has in fact already been won by a competitor, and nobody in the company has recorded it in the system. Meetings like this raise the right question about a CRM with artificial intelligence. What needs to be in the system for its indications to be reliable?

A model works on what it finds. If activities are not recorded, if lost deals carry no reason, if half the contacts are duplicates, the suggestion that appears on screen looks more like a bet than a forecast.

A CRM with artificial intelligence only reads what someone has written

Predictive features feed on three raw materials: the history of interactions, the outcomes of closed deals and clean master data. Without the first, the system cannot tell which customers are cooling off. Without recorded outcomes, lead scoring is built on nothing. With dirty master data, reports count the same customer twice and nobody trusts the numbers any more.

Here is the uncomfortable part. Before artificial intelligence you need minimum discipline in data entry, and the software does not bring that discipline: it comes from a management decision on what must be recorded and what can be left out.

Activities nobody closes and the pipeline that results

In sales analyses we often find pipelines full of opportunities that have not moved for months. Closing a lost deal costs an explanation. Leaving it open costs nothing. So the pipeline swells, the forecast adds up figures that cannot be compared, and management decides on a blurred picture.

A system that remembers deadlines, flags stalled opportunities and asks for the reason for closing solves much of the problem, on one condition: sales reps must get something back for the time spent filling in fields. A list of calls to make in the morning, a customer to follow up before a renewal falls due, an automatic summary of a month-long email exchange are concrete returns.

What a CRM with artificial intelligence offers today, in practice

Platforms on the market mainly offer four functions, each needing a different raw material in the system.

Function What it returns What it feeds on
Summary Summaries of conversations, emails and notes: whoever inherits a customer understands in minutes where things stand Interaction history linked to the record
Prediction Scores on leads and opportunities calculated on the company’s own history, not on a generic model Recorded outcomes of closed deals
Suggestion The action to take now and the customer to act on Activities and deadlines kept up to date
Generation Draft replies and offers that a person reviews and corrects before sending Previous documents and exchanges in the system

The platform we follow in this area is SugarAI, the evolution of SugarCRM: the name changes and predictive features gain weight, while the structure of the sales, marketing and service modules remains recognisable to those who already used it. Similar logic, with different scopes, applies to the service-oriented suites we discussed in the article on CRM for service companies.

CRM with artificial intelligence: how to measure master data quality at project start

Master data quality is measured with six questions: a morning, an export from the system and a spreadsheet are enough. The answers are the first working material of a well-planned project.

  1. How many customers appear two or more times, with slightly different names?
  2. How many records lack a contact with name, role and direct details?
  3. How many open opportunities have no activity recorded in the last sixty days?
  4. How many deals lost in the past year carry a reason a colleague could understand?
  5. Can emails exchanged with customers be read by people who did not write them?
  6. Who can change master data, and who checks the result of those changes?

The answers show how much clean-up is needed, which in well-planned projects is the first phase of the work inside the platform. Deduplication rules, mandatory fields and assigned responsibilities give the predictive features an archive on which forecasting makes sense.

A sales network across three countries and a forecast that did not add up

A company with its sales network spread across three countries asked us for predictive features because the forecast strayed too far from the result. The analysis showed that each country used a different definition of an advanced deal: in one case it meant offer sent, in another verbal agreement, in the third contract almost signed.

The CRM with artificial intelligence project started from that point, inside the platform. Deal stages were standardised in the CRM for all three countries, with transition criteria coded into the flows. Mandatory fields keep the scale aligned without depending on individual discipline. With a single assessment scale, the system built a consistent history over two quarters. The predictive features began reading homogeneous data, and forecast and result stopped telling two different stories.

Where data goes when a model comes in

Before switching on the generative features of a CRM with artificial intelligence, it is worth knowing which data leaves the company perimeter, towards which provider, on what legal basis and for how long it remains available. The guidance of the Italian data protection authority (Garante per la protezione dei dati personali) on the use of artificial intelligence is the starting point for the assessment, together with an up-to-date privacy notice and a list of the processing activities involved.

At Aesir Srl this part is handled jointly by the Aesir Intelligence and Aesir Business Solutions business units: whoever designs the CRM and whoever works on the data sit at the same table. Environments stay in our Tier IV data centres in the European Union with replication, under ISO 9001, 27001, 27017 and 27018 certifications; MSP and MSSP services fall within the same contract, so security does not become one more vendor to coordinate. We work on projects across Italy and, on specific request, in Switzerland, the United Kingdom and the United States: the sensitive part almost always concerns the data rather than the machines.

The indicators that show whether a CRM with artificial intelligence works

After six months the useful question concerns changed habits more than features switched on. Four measures are enough: the share of opportunities with at least one activity recorded in the last month, the gap between forecast and result, the time between a customer request and the first useful answer, and the number of duplicate records still in circulation.

When these values improve, the project is on track. When they stay flat while active features grow, the issue is adoption: it is solved by revising data entry rules and training the sales network. The link with the finance team deserves the same attention, a topic we covered when discussing modular ERP for small and mid-sized companies.

Frequently asked questions

Is SugarAI the same product as SugarCRM?

It is its evolution. Those coming from earlier versions find the same module structure plus predictive and generative features. The upgrade path must be checked against the version in use and existing customisations.

Does it integrate with company email and calendar?

Integration with the main productivity environments exists and covers email, calendar and contacts. It is worth deciding early which messages go into the CRM and which stay personal, because the choice has organisational as well as technical implications.

How much history is needed for forecasts to make sense?

No threshold applies to everyone. The practical criterion is the number of closed deals with a recorded outcome: a few dozen are not enough, a few hundred make the scores readable. Those starting without history collect data for two or three quarters, then switch on the predictive part.

Can drafts written by the system be checked before sending?

In properly planned projects human review remains mandatory. A CRM with artificial intelligence proposes the text, a person corrects and signs it; automatic sending to customers opens risks that almost no company wants to take.

Let’s talk

The six master data questions listed above can be put today to whoever runs the CRM in your company, without involving any vendor. The answers show which area the project will start from and which flows the predictive part will read first. The resulting picture makes the comparison with any vendor concrete. If you would like to explore the topic or assess your company’s situation, fill in the form at the bottom of the page or write to support@aesir-tech.it: we will book a free consultation and start from your numbers.

Vuoi parlarne applicato alla tua azienda?

Ti rispondiamo con l’esperienza di progetto, non con una brochure.

Request a call